# Testbench Lab: AI hardware, explained. Decisions made easier. Practical advice for choosing AI voice recorders and mini PCs. Testbench Lab helps you work out what a product can do, what it costs to keep using and which compromises matter before you buy. ## Our current picks - AI voice recorder: [Notta Memo](https://testbenchlab.com/reviews/notta-memo-review/) - Everyday AI desktop: [Geekom A9 Max](https://testbenchlab.com/reviews/geekom-a9-max-review/) - Larger local models: [Geekom A9 Mega](https://testbenchlab.com/reviews/geekom-a9-mega-review/) ## AI voice recorders A good recorder should make conversations easier to revisit. We cover dedicated card recorders, wearable devices and their transcription services, with attention to monthly allowances and the difference between recording audio and processing it. Our preferred recorder is Notta Memo, based on David Wilson’s experience using it. ## AI mini PCs A compact computer can run a useful local assistant without a large tower, but an AI badge does not tell you which models will fit. Our preferred options are Geekom A9 Max for an everyday desktop and Geekom A9 Mega for larger local models. Memory capacity and software compatibility come before headline TOPS. ## Start with the question you need answered Compare two devices if your shortlist is already small. Use the buying guides to identify the right product tier, or follow a setup guide when you own the hardware. Every recommendation should help you take a concrete next step, whether that is choosing a recorder or downloading your first local model. ## Know where the evidence comes from Hands-on opinions, manufacturer specifications and external testing are different kinds of evidence. Our pages identify them separately. A published specification is useful for choosing a configuration; it is not a substitute for a measured result under matching conditions. ## Reviews - [Anker Soundcore Work review](https://testbenchlab.com/reviews/soundcore-work-review/) - [Beelink GTR9 Pro review](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/) - [Beelink SER9 Pro review](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) - [Geekom A9 Max review: HX 470 and HX 370](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Geekom A9 Mega review: Strix Halo for local AI](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) - [HiDock P1 review](https://testbenchlab.com/reviews/hidock-p1-review/) - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) - [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/) - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Plaud Note review](https://testbenchlab.com/reviews/plaud-note-review/) - [Plaud NotePin S review](https://testbenchlab.com/reviews/plaud-notepin-s-review/) - [Pocket AI recorder review](https://testbenchlab.com/reviews/pocket-ai-review/) ## Buying guides - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) - [The best Strix Halo mini PCs](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [How we test: the Testbench Lab protocol](https://testbenchlab.com/methodology/) --- # How we assess AI recorders and mini PCs By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our evidence policy distinguishes personal experience, manufacturer figures and outside testing. This version describes how we report evidence and the conditions required for a repeatable comparison; it does not claim that every product has completed a shared bench protocol. ## How we buy and set up products An article should identify the product variant, configuration and information source. For a hardware test, record the operating system, firmware, drivers and subscription tier before collecting results. A setting change can explain a performance difference as easily as a different product can. ## How we test AI voice recorders A repeatable recorder comparison needs the same speech, room, placement and transcription language. Accuracy should be scored against a checked reference transcript, with speaker-attribution errors reported separately. Battery claims require a timed recording rundown. Convenience and summary usefulness are qualitative judgements and should be described as such. ## How we test AI mini PCs A repeatable inference comparison needs the same model file, quantisation, context, prompt and backend. Prompt processing and output generation should be recorded separately. Power and sound readings require the measurement conditions. Results from different operating systems or memory configurations should not be presented as a clean head-to-head. ## How scores are weighted The Testbench score is David Wilson’s editorial rating out of 10. It summarises the article’s recommendation, weighing the intended workload, compatibility, ownership cost and relevant limitations. It is not a measured lab result, and no weighting formula is applied to produce it; readers can see the reasoning in the article. ## Test rigs and software versions Configuration details belong beside any numerical result. A reproducible record includes RAM capacity and arrangement, power mode, BIOS, graphics driver, application build and model settings. Our current launch copy does not publish an invented test-rig inventory or treat source specifications as measurements. ## Protocol changelog Evidence policy v1.0 — 2 October 2026. Launch coverage separates editorial preference from measured results, labels advertised maximums and links to source documentation. Future benchmark datasets will include the conditions needed to interpret and reproduce them. ## Frequently asked questions ### Do you test every product the same way? A fair measured comparison requires matching conditions. The launch coverage combines hands-on opinion and researched assessments, and does not claim that every product completed the same test. ### Can I reproduce your results? A numerical result is reproducible only when its hardware, software and workload settings are supplied. The launch copy avoids publishing figures without those conditions. ### Why did a score change? The Testbench score is an editorial rating, not a lab measurement. A score or recommendation can change when a price, configuration, software feature or supporting evidence changes. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) --- # About Testbench Lab & editorial policy By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Testbench Lab covers AI voice recorders, compact computers and the software that makes them useful. David Wilson writes and edits the site, with an emphasis on practical decisions rather than impressive-sounding feature lists. ## Who runs Testbench Lab? David Wilson writes and edits the site. His coverage connects the purchase decision to the work you want to do: capturing a meeting, finding an action item or running a local assistant. Personal preferences are presented as editorial judgements, with supporting facts linked to their sources. ## How do we get the products we review? A hands-on claim means the author has used the product. A researched assessment draws on manufacturer documentation and attributed reporting. These are identified in the relevant article. Product ownership, purchases, review loans and gifts are disclosed when that information is available; a review title alone should not be taken as proof of a laboratory test. ## How does Testbench Lab make money? Purchase links are provided to help readers find the correct product and configuration. Any affiliate relationship or paid placement must be disclosed where it appears. Editorial recommendations should be assessed against the reasoning in the article, rather than the presence of a shopping button. ## What brands can and cannot do Brands can supply specifications, explain features and flag factual errors. A recommendation remains an editorial choice. Our current preferences are Notta for AI recorders and Geekom for mini PCs; those preferences do not imply that either wins every measurable category or meets every specialist requirement. ## Corrections log 2 October 2026: the launch copy uses David Wilson as its author attribution. Notta Memo coverage reflects the updated 600-minute Hardware Plan in the company’s September 2026 support guidance. Subsequent factual corrections will be recorded here with the affected page and the change. ## How to contact the lab Send factual corrections to corrections@testbenchlab.com. Include the page URL, the passage in question and a source that supports the correction. For specifications, the exact device variant and regional listing are particularly useful. ## Related reading - [How we test: the Testbench Lab protocol](https://testbenchlab.com/methodology/) - [David Wilson — writer and editor](https://testbenchlab.com/authors/david-wilson/) ## Sources 1. [Notta Hardware Plan and eligibility](https://support.notta.ai/hc/en-us/articles/38281798150683-What-Is-the-Notta-Hardware-Plan) --- # Anker Soundcore Work review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $159 at Soundcore store (checked 2026-10-02). ## Verdict Soundcore Work is a compact wearable alternative for people who want physical capture with minimal bulk. Notta Memo remains our preferred overall recorder; Soundcore is most relevant when wearing the device is part of the requirement. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Design and wearing it Soundcore Work centres on a small recorder and its carrying or charging arrangement. Consider the whole kit, rather than the recorder alone. Secure attachment and an exposed microphone position matter more than saving a few grams if the device ends up rubbing against clothing. ## Recording quality A wearable’s position affects the sound reaching it. Keep clothing away from the microphone and choose a stable location for meetings. We do not publish an invented accuracy percentage or claim that a compact device hears every participant equally well. ## Transcription and summaries Soundcore provides app-based transcription and summary features. The useful output is a note you can check, export and act on. Give a summary a clear purpose—decisions, actions or follow-up questions—rather than assuming a longer summary contains more useful information. ## Subscription and app Soundcore’s product page lists 300 Starter minutes and 1,200 Pro minutes. Purchase benefits can have activation conditions or expiry dates, so distinguish an ongoing allowance from a promotional trial. Check the terms shown for your region and purchase date. ## Frequently asked questions ### Does Soundcore Work need a subscription? A paid upgrade depends on your processing volume and features. The advertised Starter allowance supports limited use without a paid Pro plan. ### Is Soundcore Work better than Plaud? It is an alternative for a smaller wearable setup. We prefer Plaud NotePin S within this wearable shortlist, while Notta Memo is our overall recorder preference. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) ## Sources 1. [Soundcore Work product and plan information](https://www.soundcore.com/uk/products/d3200-soundcore-work-ai-voice-recorder) Cite as: Wilson, D. (2026). Anker Soundcore Work review. Testbench Lab. https://testbenchlab.com/reviews/soundcore-work-review/ --- # Beelink GTR9 Pro review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $4,349 at Beelink store (checked 2026-10-02). ## Verdict Beelink GTR9 Pro is a premium Strix Halo alternative with substantial memory and networking. Our preferred general workstation is Geekom A9 Mega; GTR9 Pro is most compelling when its specific configuration and network hardware justify the total price. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running large local models on the GTR9 Pro GTR9 Pro combines the Max+ 395 platform with high-capacity shared memory. Choose a supported runtime and inspect allocation before loading a large model. A capacity specification establishes room, not the output rate of a particular quantised model. ## Dual 10GbE and networking Fast Ethernet matters when the rest of the network can use it. A 10GbE port connected to a slower switch cannot deliver its headline link rate. Check the actual board revision, controller, drivers and retailer specification rather than assuming every unit is identical. ## Vapor-chamber cooling and noise The cooling design is intended to manage sustained loads in a compact case. The component name alone cannot determine sound level or fan character. For a desk workstation, evaluate load behaviour; for a server, also consider its location and the ventilation around it. ## Is it worth the premium? We would choose A9 Mega for the general local-workstation role. Spend more on GTR9 Pro only if the matched memory configuration or networking features are valuable to your work. Paying for an unused interface will not improve an ordinary single-user chat session. ## Frequently asked questions ### Is the Beelink GTR9 Pro worth it? It can be for a buyer who needs its particular configuration and networking. Our general preference remains Geekom A9 Mega. ### Beelink GTR9 Pro vs GMKtec EVO-X2: which is quieter? A fair answer needs the same workload, power mode, measurement distance and noise floor. Cooling specifications alone do not settle it. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/) - [Geekom A9 Mega vs Beelink GTR9 Pro](https://testbenchlab.com/compare/geekom-a9-mega-vs-beelink-gtr9-pro/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Sources 1. [Beelink GTR9 Pro](https://www.bee-link.com/products/beelink-gtr9-pro-amd-ryzen-ai-max-395) 2. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) Cite as: Wilson, D. (2026). Beelink GTR9 Pro review. Testbench Lab. https://testbenchlab.com/reviews/beelink-gtr9-pro-review/ --- # Beelink SER9 Pro review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $1,149 at Beelink store (checked 2026-10-02). ## Verdict The Beelink SER9 Pro with Ryzen AI 9 HX 370 is a compact alternative for smaller local models and ordinary desktop work. We prefer Geekom A9 Max for this project, especially where a future RAM upgrade is part of the plan. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running local LLMs on the SER9 Pro The HX 370 version pairs Ryzen AI processing with Radeon 890M graphics. Select a model that fits the installed memory and verify the backend. The same SER9 Pro name also appears on other CPU variants, so a review or offer should not be applied to every product carrying that label. ## Noise and thermals Compact systems change fan behaviour as load increases. Judge sustained model generation separately from a brief desktop task. We do not assert that SER9 Pro is quieter than A9 Max without equivalent conditions, distance and a recorded sound measurement. ## Built-in speakers/mics and ports Built-in audio can simplify a small desk, but it does not replace checking the connection you need for a display, fast storage or a network. Write down your peripherals before buying. A useful port selection is the one that matches those peripherals without an extra hub. ## Memory and upgrades The HX 370 SER9 Pro listings use onboard LPDDR5X memory. Choose enough capacity at purchase; storage expansion is not a RAM upgrade. That distinction is the main reason an upgradeable A9 Max can suit a buyer who expects model requirements to grow. ## Frequently asked questions ### Is the Beelink SER9 Pro good for AI? The HX 370 variant can support smaller local models with compatible software. Check the actual processor because the SER9 Pro name spans multiple variants. ### Can you upgrade the RAM in the Beelink SER9 Pro? The HX 370 models discussed here use onboard LPDDR5X memory rather than replaceable SODIMMs. Choose capacity at purchase. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/) - [Geekom A9 Max vs Beelink SER9 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-beelink-ser9-pro/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) ## Sources 1. [Beelink SER9 Pro HX 370](https://www.bee-link.com/products/beelink-ser9-ai-9-hx-370) 2. [Ollama hardware support](https://docs.ollama.com/gpu) Cite as: Wilson, D. (2026). Beelink SER9 Pro review. Testbench Lab. https://testbenchlab.com/reviews/beelink-ser9-pro-review/ --- # Geekom A9 Max review: HX 470 and HX 370 By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $1,499 at Geekom store (checked 2026-10-02). ## Verdict Geekom A9 Max is our preferred general-purpose AI mini PC. The HX 470 version offers a capable Windows desktop with upgradeable memory, but we would prioritise a suitable RAM configuration over the newer chip label alone. This is a researched assessment, not a claimed Testbench Lab benchmark run. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running local LLMs on the Geekom A9 Max A9 Max is a sensible starting point for small quantised models and a desktop that also handles ordinary work. Choose a supported GPU backend and confirm acceleration before judging speed. Larger dense models can load into substantial system RAM without providing the response rate you expect from a dedicated accelerator. ## Does the NPU matter? (Ryzen AI 9 HX 470) The NPU matters when the application explicitly supports it. It does not automatically accelerate every Ollama session or desktop chatbot. A runtime using the Radeon iGPU is a separate software path, so check the active backend rather than interpreting low NPU activity as a hardware fault. ## Memory: single vs dual channel and upgrading to 128GB Geekom lists DDR5 SODIMM memory with dual-channel support and capacity up to 128GB. Check whether the configuration you buy populates both channels. Shared-memory graphics depend on memory traffic, so capacity and channel arrangement are separate purchasing decisions. ## Noise, thermals and power draw A9 Max’s cooling system has to balance a compact case with sustained work. We do not assign an unsupported dBA or power figure. For a desk placement, fan character under the workload you actually run matters more than a peak CPU temperature captured during a different test. ## Ports and expansion The published specification includes USB4, dual Ethernet connections and two different M.2 storage positions. Match a replacement SSD to its physical format. USB-C shape does not identify the connection speed, display capability or PCIe tunnelling offered by a particular port. ## HX 470 vs HX 370 version: which to buy Prefer HX 470 when the overall configuration and price are close. A discounted HX 370 with suitable memory can be the better purchase than a newer unit whose RAM needs an immediate upgrade. For much larger models, move to A9 Mega rather than paying only for a small step within the HX tier. ## Frequently asked questions ### Is the Geekom A9 Max quiet? Noise depends on load, power settings and placement. We do not claim a measured quietness ranking without matching sound tests. ### How much is the Geekom A9 Max? The price depends on CPU, RAM, SSD and regional promotions. Use the linked manufacturer listing for the exact configuration rather than a price for a different variant. ### What is the RAM slot size on the Geekom A9 Max? It uses laptop-style DDR5 SODIMM memory. Geekom lists up to 128GB; verify the module and firmware compatibility for the chosen variant. ### Can the Geekom A9 Max run a 70B model? Some quantised 70B workloads may fit after a substantial memory upgrade, but fit is not the same as useful speed. A9 Mega is our preferred Geekom tier for that task. ### Is Geekom a good brand? Geekom is our preferred brand in this project. Choose the model by workload and configuration rather than assuming every product in the range is equivalent. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/) - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/) - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/) - [Geekom A9 Max vs Minisforum AI X1 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-minisforum-ai-x1-pro/) - [Geekom A9 Max vs Beelink SER9 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-beelink-ser9-pro/) - [Geekom A9 Max vs GMKtec EVO-X1](https://testbenchlab.com/compare/geekom-a9-max-vs-gmktec-evo-x1/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) - [Ryzen AI 9 HX 470 vs HX 370: what changed?](https://testbenchlab.com/guides/ryzen-ai-9-hx-470-vs-hx-370/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [AMD Ryzen AI 400 announcement and specifications](https://ir.amd.com/news-events/press-releases/detail/1270/amd-expands-ai-leadership-across-client-graphics-and-software-with-new-ryzen-ryzen-ai-and-amd-rocm-announcements-at-ces-2026) 3. [Ollama hardware support](https://docs.ollama.com/gpu) Cite as: Wilson, D. (2026). Geekom A9 Max review: HX 470 and HX 370. Testbench Lab. https://testbenchlab.com/reviews/geekom-a9-max-review/ --- # Geekom A9 Mega review: Strix Halo for local AI By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $3,799 at Geekom store (checked 2026-10-02). ## Verdict Geekom A9 Mega is our preferred Geekom for larger local AI workloads. Its Strix Halo platform is a different proposition from A9 Max: a larger shared-memory graphics system, rather than simply a faster version of the same compact desktop. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running large local models on the A9 Mega A9 Mega’s appeal is loading larger quantised models without depending on a modest discrete card’s VRAM. Model fit still depends on the memory allocated to the accelerator and the context requested. Start with a smaller test and confirm offload before attempting a demanding session. ## Max+ 395 vs Max+ 388 versions The processor suffix matters. AMD’s Max+ 395 has more CPU cores than Max+ 388, while both belong to the higher graphics tier. Check the exact offer rather than assuming the Mega name identifies one processor. Choose 395 for broader CPU work when the premium is justified. ## How much of the 128GB can the GPU use? The advertised system capacity is not dedicated VRAM. Firmware and driver settings determine the region made available to the integrated GPU, and the operating system still needs room. Check the runtime’s reported memory instead of assuming the entire installed amount is available to model weights. ## Noise, thermals and power A larger model can occupy the system for an extended period. Provide ventilation and choose a power profile that suits the desk environment. We do not attach a fabricated tokens-per-second, dBA or wall-power value to the recommendation. ## Ports and networking Geekom lists dual 2.5GbE, USB4 and two M.2 storage positions for A9 Mega. Those connections suit a single-user workstation and fast external storage. A workload built around 10GbE or internal expansion can be a reason to choose another machine despite our Geekom preference. ## Frequently asked questions ### How much VRAM does the Geekom A9 Mega have? It uses shared system memory, not a separate bank of dedicated VRAM. The GPU-visible amount depends on the configuration and allocation settings. ### Can the Geekom A9 Mega run a 70B model? A suitable high-memory configuration can hold some quantised 70B workloads. Backend, context and memory allocation determine fit and speed. ### What is the difference between Ryzen AI Max+ 395 and 388? Max+ 395 provides more CPU cores. Both use the stronger Strix Halo graphics tier; check the exact specifications and memory configuration. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom A9 Mega vs GMKtec EVO-X2](https://testbenchlab.com/compare/geekom-a9-mega-vs-gmktec-evo-x2/) - [Geekom A9 Mega vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/geekom-a9-mega-vs-minisforum-ms-s1-max/) - [Geekom A9 Mega vs Beelink GTR9 Pro](https://testbenchlab.com/compare/geekom-a9-mega-vs-beelink-gtr9-pro/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) - [Ryzen AI Max+ 395 (Strix Halo) explained](https://testbenchlab.com/guides/ryzen-ai-max-395-strix-halo-explained/) ## Sources 1. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 2. [AMD Ryzen AI software hardware table](https://www.amd.com/en/products/software/ryzen-ai-software.html) 3. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) 4. [Ollama hardware support](https://docs.ollama.com/gpu) Cite as: Wilson, D. (2026). Geekom A9 Mega review: Strix Halo for local AI. Testbench Lab. https://testbenchlab.com/reviews/geekom-a9-mega-review/ --- # GMKtec EVO-X1 review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $1,149.99 at GMKtec store (checked 2026-10-02). ## Verdict GMKtec EVO-X1 is an HX-tier mini PC worth considering for OCuLink expansion. Geekom A9 Max is our preferred general desktop; the EVO-X1 makes the most sense when its specific connector or configuration solves a requirement. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running local LLMs on the EVO-X1 The original EVO-X1 uses Ryzen AI 9 HX 370 and belongs in the smaller-model desktop tier. Check that the chosen application accelerates on the Radeon hardware. Total system RAM, GPU-visible memory and model context are separate constraints, even on an integrated system. ## OCuLink and eGPU A native OCuLink route is valuable if an external card is already in your plan. Include the dock, GPU and power supply in the buying decision. For inference, the card’s memory can be more significant than the difference between external connection standards. ## Noise and thermals Power mode and ventilation affect sustained performance and sound. Avoid assuming a higher allowed power draw is always better for a desk computer. No unsupported acoustic winner is assigned against Geekom or Minisforum. ## EVO-X1 vs EVO-X1 Pro The original and Pro have different CPU generations. Match the chip, memory arrangement and SSD when comparing prices. A useful upgrade changes the workload you can run or how well the whole configuration supports it; it should not rest only on the word Pro. ## Frequently asked questions ### Is GMKtec a good brand? GMKtec has relevant options in this shortlist. We prefer Geekom overall and assess each GMKtec model by its configuration and expansion needs. ### Does the GMKtec EVO-X1 have OCuLink? The EVO-X1 is marketed with OCuLink expansion. Confirm the exact variant and dock requirements before purchase. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/) - [Geekom A9 Max vs GMKtec EVO-X1](https://testbenchlab.com/compare/geekom-a9-max-vs-gmktec-evo-x1/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) ## Sources 1. [GMKtec EVO-X1](https://www.gmktec.com/products/amd-ryzen%E2%84%A2-ai-9-hx-370-evo-x1-ai-mini-pc) 2. [AMD Ryzen AI 400 announcement and specifications](https://ir.amd.com/news-events/press-releases/detail/1270/amd-expands-ai-leadership-across-client-graphics-and-software-with-new-ryzen-ryzen-ai-and-amd-rocm-announcements-at-ces-2026) 3. [Ollama hardware support](https://docs.ollama.com/gpu) Cite as: Wilson, D. (2026). GMKtec EVO-X1 review. Testbench Lab. https://testbenchlab.com/reviews/gmktec-evo-x1-review/ --- # GMKtec EVO-X2 review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $3,649.99 at GMKtec store (checked 2026-10-02). ## Verdict GMKtec EVO-X2 is a relevant Strix Halo alternative for large local models. Geekom A9 Mega remains our preference, but EVO-X2 should be on the shortlist when its configuration offers a meaningful saving for the same intended workload. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running large local models on the EVO-X2 EVO-X2 is offered with Ryzen AI Max+ 395 and high-capacity onboard memory. That gives it a different model-fit proposition from an HX-tier desktop. Verify backend support, available GPU memory and the chosen quantisation before interpreting a model name as a performance claim. ## Fan noise under sustained load Fan behaviour is important when the system sits beside you for a long inference run. A brief benchmark does not describe a whole working day. We do not state an acoustic figure or rank it against A9 Mega without equivalent measurements. ## Memory allocation between CPU and GPU Installed memory is shared between the system and integrated graphics. Reserving more for the GPU leaves less for other tasks; it does not create new physical memory. Use the amount reported by the model runner to plan model weights and context. ## Ports and networking Check that the selected variant supports the network and storage connections you require. A local single-user assistant may need little beyond a reliable Ethernet link. A machine serving multiple clients or retrieving datasets over a fast network has different needs. ## Price: why it changed in 2026 Memory configuration, availability and promotions change the offer. This article does not attribute a price movement to an unverified cause. Compare like-for-like RAM and SSD configurations, include shipping and tax, and use the live store price rather than an older launch bargain. ## Frequently asked questions ### Is the GMKtec EVO-X2 loud? Sound depends on load and power mode. No matching acoustic test is supplied here. ### How much does the GMKtec EVO-X2 cost? Use the current manufacturer offer for the exact RAM and SSD variant. Prices for smaller and larger configurations are not interchangeable. ### Is the EVO-X2 good for gaming as well as AI? Its Radeon 8060S is relevant for integrated-graphics gaming, but resolution, settings and the shared power budget determine the experience. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/) - [Geekom A9 Mega vs GMKtec EVO-X2](https://testbenchlab.com/compare/geekom-a9-mega-vs-gmktec-evo-x2/) - [GMKtec EVO-X2 vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/gmktec-evo-x2-vs-minisforum-ms-s1-max/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Sources 1. [GMKtec EVO-X2](https://www.gmktec.com/en/products/amd-ryzen%E2%84%A2-ai-max-395-evo-x2-ai-mini-pc) 2. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) Cite as: Wilson, D. (2026). GMKtec EVO-X2 review. Testbench Lab. https://testbenchlab.com/reviews/gmktec-evo-x2-review/ --- # HiDock P1 review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $169 at HiDock store (checked 2026-10-02). ## Verdict HiDock P1 is worth considering if earphone-call capture is central to your workflow. Notta Memo remains our preferred general recorder, but HiDock’s included core transcription is a relevant alternative for frequent users. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## What makes HiDock P1 different (Bluetooth earphone capture) HiDock P1’s distinctive use case is recording supported earphone conversations alongside in-person speech. That is a different workflow from attaching a card recorder to a phone. Check the pairing procedure and compatibility with the specific headset you use. ## Recording calls Set up P1 and the earphones before relying on them for an important call. A short recording should confirm that both sides can be heard. Bluetooth compatibility is not a universal guarantee that every calling app and headset combination behaves the same way. ## Transcription HiNotes processes recordings into text and meeting notes. Check speaker assignment and replay the passage when a name, amount or commitment matters. A good transcript remains useful when it is easy to export or compare with the original sound. ## Subscription HiDock advertises an included Member plan with unlimited core transcription and summaries. Paid plans add capabilities beyond that core. If ongoing cost is your main concern, compare the exact included features with Notta’s allowance rather than assuming a recorder must have a subscription. ## Frequently asked questions ### Does HiDock P1 need a subscription? The included HiNotes Member plan provides core transcription and summaries without a separate recurring subscription. Additional features can require a paid plan. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) ## Sources 1. [HiNotes included membership](https://hinotes.hidock.com/) 2. [HiDock P1 setup guide](https://www.hidock.com/blogs/user-guide/hidock-p1-setup-guide) Cite as: Wilson, D. (2026). HiDock P1 review. Testbench Lab. https://testbenchlab.com/reviews/hidock-p1-review/ --- # Minisforum AI X1 Pro review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $1,199 at Minisforum store (checked 2026-10-02). ## Verdict Minisforum AI X1 Pro is an expansion-oriented alternative to Geekom A9 Max. We prefer Geekom for a general desktop, but the AI X1 Pro is particularly relevant if a native OCuLink external GPU connection is part of your plan. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running local LLMs on the AI X1 Pro Treat AI X1 Pro as an HX-tier computer until an external GPU changes the configuration. A supported iGPU backend can run smaller models; the available memory and model settings decide whether a workload fits. A newer CPU variant alone does not make the system equivalent to Strix Halo. ## OCuLink and adding an external GPU OCuLink is the main reason to shortlist this system. An external GPU adds its own VRAM, power supply and dock, changing both cost and desk space. Check driver support and the GPU’s memory before buying the connection hardware: bandwidth cannot make an undersized card hold a larger model. ## Built-in power supply, noise and thermals An internal power supply changes cable management, not the amount of heat the entire system generates. Provide ventilation and compare sustained-load behaviour in the power mode you intend to use. We do not infer a quietness result from the absence of an external brick. ## Memory and storage upgrades Check the exact Pro-370 or Pro-470 specification and the installed RAM arrangement. A price for a barebones unit excludes the components that make it usable. Count compatible memory, SSD and software in the total rather than comparing a kit price with a ready-to-use Geekom. ## Pro-470 vs Pro-370 Choose Pro-470 for a small premium with matching memory and storage. Choose Pro-370 when the saving can fund a more useful upgrade. Our general preference remains A9 Max; buy this Minisforum for its specific expansion advantage rather than a generic claim of AI superiority. ## Frequently asked questions ### Does the Minisforum AI X1 Pro have OCuLink? The Pro-470 specification includes OCuLink. Check the exact regional model and connector arrangement before ordering a dock. ### Is the Minisforum AI X1 Pro loud? Sound depends on the workload and power mode. No matching noise measurements are supplied here. ### How much RAM does the AI X1 Pro support? Use the specification for the exact CPU variant and firmware. Confirm supported module sizes before purchasing an upgrade. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/) - [Geekom A9 Max vs Minisforum AI X1 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-minisforum-ai-x1-pro/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) - [What is OCuLink? OCuLink vs USB4 for an external GPU](https://testbenchlab.com/guides/what-is-oculink/) ## Sources 1. [Minisforum AI X1 Pro-470](https://store.minisforum.com/products/minisforum-ai-x1-pro-470-mini-pc) 2. [Ollama hardware support](https://docs.ollama.com/gpu) Cite as: Wilson, D. (2026). Minisforum AI X1 Pro review. Testbench Lab. https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/ --- # Minisforum MS-S1 Max review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $3,799 at Minisforum store (checked 2026-10-02). ## Verdict Minisforum MS-S1 Max is the networking-and-expansion alternative in this Strix Halo shortlist. We prefer Geekom A9 Mega for a general single-user AI workstation, but MS-S1 Max is more interesting when its infrastructure features have a defined role. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Running large local models on the MS-S1 Max The Max+ 395 platform supports a substantial shared-memory model workspace. Decide which model and context you need before choosing capacity. Results from a different chassis or backend do not establish the speed of this particular machine under your settings. ## Dual 10GbE, USB4 v2 and the PCIe slot The published networking and expansion features distinguish MS-S1 Max from simpler desktop boxes. They are useful for moving datasets, adding supported hardware or hosting services. Check lane allocation, physical clearance and electrical support before treating the PCIe slot as a route to any desktop graphics card. ## Clustering two units Two machines do not automatically become one larger-memory accelerator. You need software that distributes the model or workloads, and the communication cost can dominate. Fast networking makes experimentation possible; it does not guarantee an improvement for every inference task. ## Noise, thermals and power Long-running server use and a short interactive session place different demands on cooling. Place the chassis where the intakes remain clear and measure the behaviour at the intended load. The infrastructure feature set is not proof that it is quieter than A9 Mega. ## 64GB or 128GB? Choose memory at purchase because this platform uses onboard memory. 64GB can serve many moderate quantised workloads, while 128GB provides more room for larger models and context. Pay for capacity you can use with your backend rather than treating maximum RAM as a general performance multiplier. ## Frequently asked questions ### Is the MS-S1 Max worth it over the GMKtec EVO-X2? It can be worth a premium if you need its networking and expansion. For simple single-machine inference, additional infrastructure may offer little benefit. ### Can you cluster two MS-S1 Max units? You can connect and run distributed software on multiple units. They do not automatically pool memory for an ordinary single-machine model runner. ### What can you put in the MS-S1 Max PCIe slot? Use hardware within the slot’s electrical, power and physical limits. Confirm those limits for the exact chassis before purchasing a card. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/) - [Geekom A9 Mega vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/geekom-a9-mega-vs-minisforum-ms-s1-max/) - [GMKtec EVO-X2 vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/gmktec-evo-x2-vs-minisforum-ms-s1-max/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Sources 1. [Minisforum MS-S1 Max](https://store.minisforum.com/products/minisforum-ms-s1-max-mini-pc) 2. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) Cite as: Wilson, D. (2026). Minisforum MS-S1 Max review. Testbench Lab. https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/ --- # Notta Memo review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: hands-on. Price: $149 at Notta store (checked 2026-10-02). ## Verdict Notta Memo is our preferred AI voice recorder. David Wilson found it much better than Plaud Note Pro in everyday use, and the included Hardware Plan gives eligible owners a generous starting point without a paid subscription. ## Notta Memo vs the Notta app: what the hardware adds Notta Memo is the dedicated recorder; Notta is also the software service. The hardware gives you physical recording controls and local audio storage, so capture is not dependent on opening a phone app at the right moment. Processing and managing the notes still involves software. ## Recording calls and meetings Notta Memo provides a physical switch between room recording and smartphone-call capture. Choose the mode before recording and confirm the status on the device. For a group conversation, placement is more useful than proximity to the person taking the notes: put the recorder where it can hear the group. ## Transcription, translation and summaries Notta’s workflow turns captured speech into searchable text and structured summaries. Use the transcript to locate a passage and replay important details before acting on them. Language availability and translation allowances are separate questions; a supported language does not imply unlimited translated sessions. ## Battery and file transfer Notta lists up to 30 hours of recording, 32GB storage and Bluetooth/Wi-Fi transfer. These specifications support a portable capture workflow, but are not a promise of continuous cloud processing. Check that a transfer is complete before deleting the device copy. ## Notta plans and what is free Eligible Memo owners can receive 600 monthly transcription minutes through the Hardware Plan. It replaces the older Starter allowance. The account must meet the workspace rules and remain bound to the device. Separate limits apply to translations, summaries and other features, even when transcription time remains. ## Frequently asked questions ### Does Notta Memo need a subscription? A paid subscription is optional within the eligible Hardware Plan allowance. Upgrade when your usage or required features exceed what is included. ### Can Notta Memo record phone calls? Yes, Notta Memo has a smartphone-call recording mode. Use the appropriate attachment and mode rather than expecting ordinary room recording to capture the earpiece clearly. ### Is Notta Memo better than Plaud? Notta Memo is our overall preference after using it and Plaud Note Pro. That is a hands-on judgement, rather than a claim that it wins every technical specification. ### How many languages does Notta Memo support? Notta advertises 58-language support. Check the specific transcription or translation function you need, because availability and plan limits can differ. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) ## Sources 1. [Notta Memo specifications](https://shop.notta.ai/en-us/products/notta-memo) 2. [Notta Hardware Plan and eligibility](https://support.notta.ai/hc/en-us/articles/38281798150683-What-Is-the-Notta-Hardware-Plan) Cite as: Wilson, D. (2026). Notta Memo review. Testbench Lab. https://testbenchlab.com/reviews/notta-memo-review/ --- # Plaud Note Pro review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: hands-on. Price: $189 at Plaud store (checked 2026-10-02). ## Verdict Plaud Note Pro is a capable card-style recorder, but Notta Memo is our preferred purchase. David Wilson favoured Notta after using both; Plaud remains relevant if its automatic recording-mode selection or longer endurance mode suits your routine. ## Design and how it attaches to your phone Plaud Note Pro uses a slim card shape with a display and magnetic case. Phone attachment keeps the recorder available without adding another item to your pocket. The screen makes recording status visible, which matters when you want to confirm capture without checking the app. ## Recording phone calls vs in-person meetings Plaud Note Pro switches automatically between call and room recording. That reduces the steps before recording, but correct placement still matters. A meeting requires a microphone position that can hear everyone; call capture requires the recorder to work with the phone rather than sitting elsewhere in the room. ## Transcription and summaries Plaud turns uploaded recordings into searchable text and structured notes. Treat a summary as a navigation aid, especially where names, deadlines or commitments matter. Review the transcript and original audio before sending an action list to colleagues. ## Battery life and storage Plaud advertises 30 hours in Enhanced mode and 50 hours in Endurance mode, with a reduced pickup range in the latter. Internal storage is 64GB. These are specifications, not our timed results, and recording capacity should not be confused with monthly transcription allowance. ## Plaud plans: what the subscription adds Starter includes 300 monthly transcription minutes. Pro increases that to 1,200 and Unlimited removes the standard monthly minute cap subject to its terms. For a lighter workload, start with the included plan rather than paying before you know your usage. ## Plaud Note Pro vs Plaud Note: is the Pro worth it? Within Plaud’s card range, we favour the Pro for a new purchase when the premium is modest. Existing Note owners should upgrade for a specific useful feature, rather than expecting a completely different software service. Buyers open to another brand should read our Notta comparison first. ## Frequently asked questions ### Can you use Plaud Note Pro without a subscription? Yes, within the included Starter allowance. Recording audio and obtaining paid cloud transcription are separate parts of the workflow. ### Is there something better than Plaud? Our preferred recorder is Notta Memo, based on David Wilson’s experience. It is the first alternative we would consider. ### Which is better, Plaud or Pocket? We favour Plaud Note Pro for a conventional meeting-and-call workflow. Pocket is worth considering for its advertised unlimited core processing; the better fit depends on your workflow. ### How long does the Plaud Note Pro battery last? The advertised maximum is 50 hours in Endurance mode, or 30 hours in Enhanced mode. Actual runtime depends on settings and use. ### Does Plaud Note Pro work with Android? Plaud supports an Android companion app. Check the current app requirements and phone attachment before buying. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) - [Plaud Note vs Plaud Note Pro](https://testbenchlab.com/compare/plaud-note-vs-plaud-note-pro/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [Does Plaud need a subscription? Plans and free minutes explained](https://testbenchlab.com/guides/does-plaud-need-a-subscription/) - [How to use the Plaud Note Pro: setup to first summary](https://testbenchlab.com/guides/how-to-use-plaud-note-pro/) ## Sources 1. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) Cite as: Wilson, D. (2026). Plaud Note Pro review. Testbench Lab. https://testbenchlab.com/reviews/plaud-note-pro-review/ --- # Plaud Note review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $159 at Plaud store (checked 2026-10-02). ## Verdict The original Plaud Note still makes sense when it is meaningfully cheaper than the Pro. For a new purchase at a small price difference, we favour Plaud Note Pro; across brands, Notta Memo is our preferred recorder. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## Is the original Plaud Note still worth buying? The original Plaud Note remains useful for recording calls and meetings. Its value depends on the discount and whether you need the Pro’s newer controls. Do not replace a working recorder simply because a newer model exists; compare the feature you would gain with the full upgrade cost. ## Call recording and meetings Plaud Note uses separate modes for call capture and surrounding speech. Selecting the right mode is part of setup, especially when switching quickly between a conversation in the room and a phone call. Review a brief sample if your case or phone attachment changes. ## Battery and storage Manufacturer figures are useful for planning, but recording endurance does not determine how much transcription your plan includes. Make sure the recorder is charged and the last recording has synced before an extended session. Keeping an original audio copy also makes later transcript checking possible. ## Plaud Note vs Note Pro The Pro adds more sophisticated recording controls and an endurance option. The original Note is the economical choice within Plaud when discounted; the Pro is our preference when the premium is small. Read the comparison for the practical trade-offs rather than treating a higher model name as an automatic improvement. ## Frequently asked questions ### Is Plaud Note discontinued? Plaud’s current product comparison still lists the original Note. Availability can vary by store and country. ### Does Plaud Note need a subscription? A paid plan is optional within Plaud’s included transcription allowance. More processing can require an upgrade. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [Plaud Note vs Plaud Note Pro](https://testbenchlab.com/compare/plaud-note-vs-plaud-note-pro/) - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) ## Sources 1. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) Cite as: Wilson, D. (2026). Plaud Note review. Testbench Lab. https://testbenchlab.com/reviews/plaud-note-review/ --- # Plaud NotePin S review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $179 at Plaud store (checked 2026-10-02). ## Verdict Plaud NotePin S is our wearable choice within Plaud’s range. Its physical button and highlighting control suit people who want recording close at hand, although Notta Memo remains our overall recorder preference. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## NotePin S vs the original NotePin: what the extra $20 buys Plaud describes the S upgrade as a tactile button, press-to-highlight control and the full set of wearing accessories. The recording platform is shared, so the upgrade is mainly about interaction. Buy it for easier physical control rather than expecting a wholly new transcription engine. ## Wearing it: clip, wristband and lanyard Choose a position that keeps the recorder stable and its microphones exposed. A clip or lanyard can keep it near speech; a wrist position moves more and may be farther from other speakers. Comfortable attachment matters because a wearable only helps if you keep wearing it. ## Recording quality on the move Movement adds clothing friction, handling noise and changing distances. Avoid covering the recorder with a collar or sleeve. A short trial recording in your usual wearing position is more informative than a language count or an AI accuracy claim on the box. ## Battery life Plaud rates NotePin S at up to 20 hours of continuous recording. This is an advertised maximum. A wearable recording throughout the day has different charging needs from a device used only for scheduled meetings. ## Plaud plans NotePin S uses Plaud’s transcription service. The device is not a route around the service’s minute allowances: choose the plan according to how much audio you process. Our subscription guide explains the options without requiring you to repeat the same calculation for each Plaud model. ## Frequently asked questions ### What is the difference between Plaud NotePin and NotePin S? The S adds a tactile recording button, press-to-highlight and more included wearing accessories. Plaud lists the same core battery and storage specifications. ### Can Plaud NotePin record phone calls? It is primarily a wearable for surrounding speech, rather than a card recorder with direct phone-call capture. Use a supported call-recording method for that task. ### Is the NotePin S waterproof? Do not treat it as a waterproof swimming or shower device. Follow the current manufacturer care instructions and keep it away from immersion. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) - [Plaud Note Pro vs NotePin S: card or wearable?](https://testbenchlab.com/compare/plaud-note-pro-vs-notepin-s/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) ## Sources 1. [Plaud NotePin S specifications](https://www.plaud.ai/products/plaud-notepin-s) 2. [What changed in NotePin S](https://uk.plaud.ai/blogs/articles/plaud-notepin-vs-plaud-notepin-s) Cite as: Wilson, D. (2026). Plaud NotePin S review. Testbench Lab. https://testbenchlab.com/reviews/plaud-notepin-s-review/ --- # Pocket AI recorder review By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Assessment: researched. Price: $139 at Pocket store (checked 2026-10-02). ## Verdict Pocket is an alternative for people attracted to a conversational memory workflow and unlimited core processing. We would still start with Notta Memo for a general-purpose recorder, while Pocket’s advertised no-subscription core features deserve attention from high-volume users. This researched assessment uses published specifications. It does not claim a first-hand laboratory test; the Testbench score is an editorial rating, not a measurement. ## What Pocket is (and what it is not) Pocket here means the recorder from heypocket.com. It is not the discontinued read-it-later service or ADLINK’s Pocket AI GPU. The product combines a dedicated audio device with an app for transcripts, summaries and retrieval. ## Wearing and recording Decide how you intend to carry Pocket before buying it. Reliable access to the recording control is more important than a lifestyle photo. Keep the microphones unobstructed and use a short sample to check placement before an important conversation. ## Transcription and the AI assistant Pocket’s appeal is being able to revisit captured conversations through its app. Ask a narrow question and check the corresponding passage when an answer matters. A memory assistant is useful when it points back to evidence, rather than simply producing a plausible account of the meeting. ## Does Pocket AI need a subscription? Pocket’s current site advertises unlimited minutes and lifetime core features without a subscription. It also advertises a free subscription trial, so distinguish included core access from paid extras. Check the offer and renewal terms for the exact package you order. ## Is Pocket AI worth it? Pocket is worth considering if core processing without a monthly cap is your main concern. For our general recorder recommendation, Notta comes first. Pocket’s ownership proposition should be assessed alongside export, retention and the usefulness of its app to your own routine. ## Frequently asked questions ### Does Pocket AI require a subscription? The current manufacturer site says core features and unlimited minutes do not require a subscription. Paid features and promotional trials have separate terms. ### Is Pocket AI worth it? It can be a good fit for frequent recording and retrieval. We prefer Notta Memo overall, but Pocket’s advertised core allowance can suit a different usage pattern. ### How does Pocket AI work? The device captures audio and the companion service turns it into text and notes that can be searched and revisited. ### Is Pocket better than Plaud? Pocket is attractive for its core allowance. We favour Plaud Note Pro for a conventional card-recorder workflow, and Notta Memo over both as our general preference. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) - [Does Pocket AI need a subscription? What is free and what is not](https://testbenchlab.com/guides/does-pocket-ai-need-a-subscription/) ## Sources 1. [Pocket product and membership information](https://heypocket.com/) Cite as: Wilson, D. (2026). Pocket AI recorder review. Testbench Lab. https://testbenchlab.com/reviews/pocket-ai-review/ --- # Geekom A9 Max vs Beelink SER9 Pro By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom A9 Max. ## Short answer Choose Geekom A9 Max over the HX 370 Beelink SER9 Pro as our preferred general AI desktop. The A9 Max’s replaceable memory is particularly relevant if your local-model requirements may grow after purchase. | Question | Buying implication | |---|---| | Might RAM needs increase? | Prefer Geekom A9 Max | | Is a fixed installed capacity sufficient? | SER9 Pro can remain an alternative | | Do you need large-model shared graphics memory? | Consider the Strix Halo tier | ## Which is faster for local AI? A reliable answer needs a matched memory setup and backend, not simply a comparison of processor labels. SER9 Pro’s onboard memory and A9 Max’s SODIMM arrangement make capacity and bandwidth part of the decision. Confirm the SER9 Pro CPU because other variants share the name. ## Which is quieter? We do not assign a noise winner without comparable recordings or readings. Compare sustained model generation in a similar power mode and room, rather than mixing an idle measurement from one review with a load result from another. ## Which should you buy? We prefer A9 Max for the upgradeable desktop route. SER9 Pro suits buyers who are confident the installed memory will remain sufficient. Check the supported RAM upgrade before buying modules, and remember that storage expansion cannot solve a RAM limit. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Beelink SER9 Pro review](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [Beelink SER9 Pro HX 370](https://www.bee-link.com/products/beelink-ser9-ai-9-hx-370) --- # Geekom A9 Max vs GMKtec EVO-X1 By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom A9 Max. ## Short answer Geekom A9 Max is our preferred general desktop over GMKtec EVO-X1. The EVO-X1 deserves a place on the shortlist when a native OCuLink connection or a materially better matched offer is important. | Requirement | Our choice | |---|---| | General AI desktop | Geekom A9 Max | | Native OCuLink for a planned eGPU | Consider GMKtec EVO-X1 | | Substantially larger models without eGPU | Consider Geekom A9 Mega | ## Which is faster for local AI? Match the original EVO-X1 against the correct A9 Max processor generation. An HX 370-versus-HX 470 comparison can also hide different memory and power configurations. Use the same model, context and backend to compare actual inference; this page does not invent a benchmark result. ## Which should you buy? Choose A9 Max for a straightforward desktop with room to develop your local-AI workflow. Choose EVO-X1 when the connector or price is a concrete advantage for your intended configuration. If an external GPU is involved, compare the completed setup rather than the computer price in isolation. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [GMKtec EVO-X1](https://www.gmktec.com/products/amd-ryzen%E2%84%A2-ai-9-hx-370-evo-x1-ai-mini-pc) --- # Geekom A9 Max vs Minisforum AI X1 Pro By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom A9 Max. ## Short answer Geekom A9 Max is our preferred purchase over Minisforum AI X1 Pro for an everyday AI desktop. Choose the Minisforum when native OCuLink is a defined requirement; our Geekom preference is not an unsupported claim of higher inference speed. | Decision | Preference | |---|---| | General compact desktop | Geekom A9 Max | | Native OCuLink requirement | Minisforum AI X1 Pro | | Measured inference or noise winner | Not established under matching conditions | ## Which is faster for local AI? There is no clean speed winner without matching chip variant, RAM arrangement, model and backend. CPU generation is only part of the system. Verify GPU use first and compare the same quantisation and context before interpreting tokens per second. ## OCuLink vs USB4: does it matter? It matters if you intend to add a discrete GPU. Native OCuLink supplies a more direct PCIe path, while USB4 supports a broader range of peripherals and docks. If the integrated GPU is all you plan to use, an unused connector should carry little weight. ## Which is quieter? Quietness needs a matching workload and measurement setup. Manufacturer cooling descriptions do not establish the winner. Consider both the steady fan level and changes in pitch when the system switches between idle and generation. ## Which should you buy? Choose A9 Max for the general desktop role. Choose AI X1 Pro for a properly planned OCuLink setup. For large models without an external graphics card, consider Geekom A9 Mega rather than treating two HX-tier machines as substitutes for Strix Halo. ## Frequently asked questions ### Geekom A9 Max or Minisforum AI X1 Pro: which is better? A9 Max is our preferred general desktop. AI X1 Pro is the specialist choice here for native OCuLink expansion. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) - [What is OCuLink? OCuLink vs USB4 for an external GPU](https://testbenchlab.com/guides/what-is-oculink/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [Minisforum AI X1 Pro-470](https://store.minisforum.com/products/minisforum-ai-x1-pro-470-mini-pc) --- # Geekom A9 Mega vs Beelink GTR9 Pro By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom A9 Mega. ## Short answer Geekom A9 Mega is our preferred purchase over Beelink GTR9 Pro for a general Strix Halo workstation. Consider the Beelink when its matched configuration or networking meets a requirement that the Geekom does not. | Question | Decision | |---|---| | General Strix Halo workstation? | Prefer Geekom A9 Mega | | Need GTR9 Pro network features? | Compare the exact board and configuration | | Choosing model capacity? | Match usable RAM before price | ## Same chip: so what is different? The Max+ 395 platform is shared, but installed memory, networking hardware, firmware and cooling vary. Compare the exact GTR9 Pro board and seller listing. A high-capacity premium offer should not be compared with a lower-memory Geekom as though only the brand changed. ## Which should you buy? Choose A9 Mega as our general recommendation. Choose GTR9 Pro when its particular networking or delivered capacity is essential and the complete offer is acceptable. We do not claim that a lower price or an editorial preference proves a faster, quieter or more reliable machine. ## Related reading - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [Beelink GTR9 Pro review](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/) - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/) ## Sources 1. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 2. [Beelink GTR9 Pro](https://www.bee-link.com/products/beelink-gtr9-pro-amd-ryzen-ai-max-395) --- # Geekom A9 Mega vs GMKtec EVO-X2 By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom A9 Mega. ## Short answer Geekom A9 Mega is our preferred Strix Halo workstation over GMKtec EVO-X2. That recommendation is an editorial choice; the shared Max+ 395 platform does not establish that Geekom generates every model faster. | Decision | Verdict | |---|---| | Overall editorial preference | Geekom A9 Mega | | Model-fit comparison | Match memory and allocation | | Measured speed and sound winner | Not established here | ## Same chip: so what is different? Installed memory, firmware, cooling, power profile, ports and seller terms still vary between machines with the same chip. Compare the complete configuration. A 128GB system and a 64GB system answer different model-fit questions even when the processor matches. ## Which runs large models faster? Matching backend and memory allocation are essential. Long-context processing and output generation can respond differently to the same settings. Without a controlled result, the processor name supports a platform comparison rather than a precise chassis performance ranking. ## Which is quieter? Noise is not established by the shared silicon or by the cooling component name. Long-running generation can produce a different fan pattern from a synthetic burst. We do not claim a sound winner without a matching workload and measurement method. ## Which should you buy? Our choice is A9 Mega for a general local workstation. EVO-X2 is worth considering when the same usable capacity and storage come at a meaningful saving. Keep the model-fit requirement fixed rather than stepping down in memory to make a price comparison look favourable. ## Frequently asked questions ### Is the Geekom A9 Mega better than the GMKtec EVO-X2? A9 Mega is our preferred purchase. A matched benchmark is needed to establish a speed or acoustic advantage. ## Related reading - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Sources 1. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 2. [GMKtec EVO-X2](https://www.gmktec.com/en/products/amd-ryzen%E2%84%A2-ai-max-395-evo-x2-ai-mini-pc) --- # Geekom A9 Mega vs Minisforum MS-S1 Max By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom A9 Mega. ## Short answer Geekom A9 Mega is our preference for a single-user local AI workstation. Choose Minisforum MS-S1 Max instead when 10GbE or internal expansion is a requirement you can use, rather than a specification you might want later. | Workload | Recommended choice | |---|---| | Single-user local workstation | Geekom A9 Mega | | Defined 10GbE or internal-card requirement | Minisforum MS-S1 Max | | Basic chat inference | Prioritise usable memory and backend | ## Same chip: so what is different? Both can be configured around Max+ 395, but their surrounding hardware serves different priorities. A9 Mega is our desktop starting point; MS-S1 Max adds infrastructure-oriented features. Match RAM, SSD and CPU suffix so the offer reflects the workstation you actually intend to buy. ## Do you need 10GbE and a PCIe slot? A faster network helps if datasets or clients move enough data and the rest of the network supports it. Internal expansion helps when a compatible card fits a defined task. Neither feature automatically accelerates a model already loaded into memory. ## Which should you buy? Choose A9 Mega for the general role. Choose MS-S1 Max for a genuine expansion or network requirement. Our Geekom preference does not override compatibility: a required interface can be a better reason to buy than an overall brand verdict. ## Related reading - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Sources 1. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 2. [Minisforum MS-S1 Max](https://store.minisforum.com/products/minisforum-ms-s1-max-mini-pc) --- # Geekom vs Beelink: which AI mini PC brand? By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom. ## Short answer Geekom is our preference over Beelink for this AI mini PC shortlist. We favour A9 Max for a general desktop and A9 Mega for larger models, while Beelink remains relevant for buyers satisfied with its fixed-memory configurations or specialist networking. | Tier | Our preference | Important comparison check | |---|---|---| | Ryzen AI 9 | Geekom A9 Max | Exact SER9 Pro CPU and memory type | | Strix Halo | Geekom A9 Mega | Matched RAM, SSD and network features | ## Geekom vs Beelink at the Ryzen AI 9 tier Compare A9 Max with the HX 370 SER9 Pro specifically. The SER9 Pro label also covers other processors, so do not mix their specifications. We prefer the A9 Max’s upgrade path for buyers who expect memory needs to grow; a suitable onboard-memory Beelink can still serve a settled workload. ## Geekom vs Beelink at the Ryzen AI Max+ 395 tier A9 Mega is our general local-workstation choice. GTR9 Pro offers a different networking and configuration proposition. Match capacity, SSD and the actual board before judging price or accepting a result from another review as directly comparable. ## Which brand should you buy? Choose Geekom as the default in this project. Choose Beelink for a configuration or feature you specifically need, rather than a vague expectation of better reliability. We do not have fleet failure data establishing that either brand lasts longer. ## Frequently asked questions ### Is Geekom better than Beelink? Geekom is our preferred choice for the use cases covered here. Compare the exact products rather than extending that verdict to every model. ### Which is more reliable, Geekom or Beelink? No comparable long-term failure dataset is supplied here. Warranty terms and a personal brand preference do not establish a reliability ranking. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Beelink SER9 Pro review](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [Beelink GTR9 Pro review](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/) - [Geekom A9 Max vs Beelink SER9 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-beelink-ser9-pro/) - [Geekom A9 Mega vs Beelink GTR9 Pro](https://testbenchlab.com/compare/geekom-a9-mega-vs-beelink-gtr9-pro/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 3. [Beelink SER9 Pro HX 370](https://www.bee-link.com/products/beelink-ser9-ai-9-hx-370) 4. [Beelink GTR9 Pro](https://www.bee-link.com/products/beelink-gtr9-pro-amd-ryzen-ai-max-395) --- # Geekom vs GMKtec: which AI mini PC brand? By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom. ## Short answer Geekom is our preferred brand over GMKtec for a general AI desktop or local-model workstation. GMKtec is worth comparing when a matched configuration costs meaningfully less or OCuLink expansion is already part of the plan. | Intended use | Our preference | GMKtec alternative | |---|---|---| | Everyday AI desktop | Geekom A9 Max | EVO-X1 / EVO-X1 Pro | | Larger local models | Geekom A9 Mega | EVO-X2 | ## Geekom vs GMKtec at the Ryzen AI 9 tier We choose A9 Max as the general desktop. EVO-X1 and EVO-X1 Pro need to be separated by processor generation and configuration. A native external-GPU connector can be useful, but only after including the dock and card in the ownership cost. ## Geekom vs GMKtec at the Ryzen AI Max+ 395 tier Our preference is A9 Mega. EVO-X2 is a direct alternative in the same platform class. Match memory capacity and the model backend before attributing differences to the logo on the case. A lower-priced configuration can be good value without becoming a measured performance winner. ## Which brand should you buy? Buy Geekom unless a specific GMKtec offer or feature changes your decision. Preserve enough budget for the memory configuration your models require. A brand preference is useful for narrowing a shortlist; it does not make an unsuitable configuration suitable. ## Frequently asked questions ### Is Geekom better than GMKtec? Geekom is our preference for this shortlist. A particular GMKtec configuration can still make sense for price or expansion. ### Is GMKtec a good brand? It offers relevant alternatives in both tiers. Evaluate the exact product, seller and warranty instead of assuming the entire range has identical strengths. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) - [Geekom A9 Max vs GMKtec EVO-X1](https://testbenchlab.com/compare/geekom-a9-max-vs-gmktec-evo-x1/) - [Geekom A9 Mega vs GMKtec EVO-X2](https://testbenchlab.com/compare/geekom-a9-mega-vs-gmktec-evo-x2/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 3. [GMKtec EVO-X1](https://www.gmktec.com/products/amd-ryzen%E2%84%A2-ai-9-hx-370-evo-x1-ai-mini-pc) 4. [GMKtec EVO-X2](https://www.gmktec.com/en/products/amd-ryzen%E2%84%A2-ai-max-395-evo-x2-ai-mini-pc) --- # Geekom vs Minisforum: which AI mini PC brand? By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Geekom. ## Short answer Geekom is our preferred brand for a general AI mini PC purchase. Choose A9 Max for an everyday desktop or A9 Mega for larger local models; Minisforum deserves attention when native OCuLink or workstation networking is a firm requirement. | Workload | Our preferred starting point | Minisforum reason to switch | |---|---|---| | Everyday desktop and smaller models | Geekom A9 Max | Native OCuLink requirement | | Larger local models | Geekom A9 Mega | MS-S1 Max networking or expansion | ## Geekom vs Minisforum at the Ryzen AI 9 HX 470 tier A9 Max and AI X1 Pro-470 belong in a similar processor tier, but the complete configurations differ. We favour A9 Max as the general desktop starting point. Choose the Minisforum when OCuLink is part of a costed external-GPU plan, rather than assuming it will be useful someday. ## Geekom vs Minisforum at the Ryzen AI Max+ 395 tier Our preference is A9 Mega for a single-user workstation. MS-S1 Max targets buyers who can use its network and expansion features. Match installed memory before comparing the machines: a higher-priced 128GB offer is not equivalent to a 64GB alternative. ## Warranty and support Compare the written warranty for your country and seller. Duration matters, but so do the return route, covered faults and responsibility for shipping. A policy is not a measured service-quality result, and this comparison does not claim a tested support winner. ## Which brand should you buy? Start with Geekom unless a Minisforum feature solves a specific requirement. That is our editorial recommendation, not a claim that Geekom is faster in every task. The matched model, memory arrangement and supported backend remain the basis of any performance comparison. ## Frequently asked questions ### Is Geekom better than Minisforum? Geekom is our overall preference. Minisforum can be a better specialist fit for an expansion or networking requirement. ### Which mini PC brand has the best warranty? Compare the exact regional warranties and seller terms. A longer published duration alone does not establish a better support experience. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) - [Geekom A9 Max vs Minisforum AI X1 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-minisforum-ai-x1-pro/) - [Geekom A9 Mega vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/geekom-a9-mega-vs-minisforum-ms-s1-max/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 3. [Minisforum AI X1 Pro-470](https://store.minisforum.com/products/minisforum-ai-x1-pro-470-mini-pc) 4. [Minisforum MS-S1 Max](https://store.minisforum.com/products/minisforum-ms-s1-max-mini-pc) --- # GMKtec EVO-X2 vs Minisforum MS-S1 Max By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: GMKtec EVO-X2. ## Short answer GMKtec EVO-X2 is our pick over Minisforum MS-S1 Max for straightforward single-machine inference when matched capacity costs less. MS-S1 Max is worth the extra when its networking and expansion will be used. | Priority | Our recommendation | |---|---| | Simple single-machine inference | GMKtec EVO-X2, subject to matched total cost | | Networking and internal expansion | Minisforum MS-S1 Max | | Wider shortlist | Geekom A9 Mega is our preference | ## Is the MS-S1 Max worth the extra? The premium should buy a needed feature, not just a longer specification list. Fast networking helps when moving data or serving a networked workload. A PCIe slot matters when a supported card fits the job. Neither benefit is essential to every local assistant. ## Which should you buy? Start with EVO-X2 for a simpler inference workstation and compare like-for-like RAM and SSD offers. Buy MS-S1 Max for the infrastructure role. If neither special feature matters and the shortlist is open, our overall Strix Halo preference is Geekom A9 Mega. ## Related reading - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Sources 1. [GMKtec EVO-X2](https://www.gmktec.com/en/products/amd-ryzen%E2%84%A2-ai-max-395-evo-x2-ai-mini-pc) 2. [Minisforum MS-S1 Max](https://store.minisforum.com/products/minisforum-ms-s1-max-mini-pc) --- # Notta Memo vs Plaud Note Pro By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Notta Memo. ## Short answer **Notta Memo is our overall winner after using both recorders.** At $149, Notta Memo costs $40 less than the $189 Plaud Note Pro, and eligible owners now receive twice as much free monthly transcription. Choose Notta Memo for value; choose Plaud Note Pro if its longer advertised recording endurance is your priority. Our hands-on verdict comes from David Wilson’s experience using both products. Specifications and subscription allowances below come from manufacturer documentation checked on 2 October 2026. Battery figures are advertised maximums, rather than our measured runtimes. ## Notta Memo vs Plaud Note Pro at a glance | Feature | Notta Memo | Plaud Note Pro | |---|---|---| | US device price | $149, including leather case | $189, including magnetic case | | Recording hardware | Four MEMS microphones plus a bone-conduction microphone | Four MEMS microphones plus a voice pickup unit | | Recording modes | Physical switch for meetings or calls | Automatic switching between meetings and calls | | Advertised recording battery life | Up to 30 hours | Up to 30 hours in Enhanced mode; 50 hours in Endurance mode | | Internal storage | 32GB | 64GB | | Advertised weight | 28g | Approximately 30g | | File transfer | Bluetooth and Wi-Fi | Bluetooth and Wi-Fi | | Free monthly transcription | 600 minutes for eligible Hardware Plan users | 300 minutes on Starter | | Overall recommendation | Our pick | Consider for recording endurance | Sources: [Notta Memo specifications](https://shop.notta.ai/en-us/products/notta-memo), [Notta Hardware Plan](https://support.notta.ai/hc/en-us/articles/38281798150683-What-Is-the-Notta-Hardware-Plan) and [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro). ## Why Notta Memo wins our comparison Notta Memo was the better recorder in our hands-on experience, and its lower ownership cost strengthens that recommendation. The preference comes from using the products; the financial advantage can be checked directly against their published prices. For a buyer, the question is whether the more expensive device offers a benefit they will actually use. We would keep the $40 saving unless a specific Plaud feature matters to the way you record. The free allowance is especially useful for occasional meetings and interviews. Ten hours of monthly transcription gives eligible Notta owners more room before a paid upgrade becomes necessary. ## Which is better for phone calls? Notta Memo is our overall preference, but the available evidence does not establish a measured call-quality winner. Both products support phone-call recording; Notta uses a physical mode switch, while Plaud detects the recording mode automatically. Notta’s switch lets you select the mode explicitly before starting. Plaud’s approach removes that manual step. These are different controls, rather than proof that one produces a more accurate call transcript. Independent experience supports Notta’s suitability for this job. In its February 2026 [Notta Memo review](https://www.techradar.com/pro/notta-memo-ai-voice-recorder-review), TechRadar used the recorder for calls, a class and a meeting, and praised its transcription and straightforward controls. That was a standalone review, rather than a controlled comparison with Plaud Note Pro. ## Which transcribes more accurately? A reliable numerical accuracy winner cannot be established from the evidence available for this comparison. Our preference for Notta Memo is a hands-on judgement, rather than a measured word-error-rate result. TechRadar reported very high accuracy across its Notta Memo recordings. However, the review did not publish a scored reference transcript or a repeatable word-error-rate calculation. Its findings are useful evidence of successful everyday use, but should not be read as a guarantee for every voice, accent or room. For your own recordings, pay particular attention to names, figures and technical terms. A transcript that reads smoothly can still contain a mistake that changes what was said. Check important details against the audio before relying on the summary. ## Subscription cost: Notta vs Plaud plans over a year Notta Memo has the lower entry cost and the larger free transcription allowance for eligible owners. The latest Notta support guidance replaces the former 300-minute Starter allowance with a 600-minute Hardware Plan. | Option | Device | Annual subscription | First-year total | Monthly transcription | |---|---:|---:|---:|---:| | Notta Memo with Hardware Plan | $149 | $0 | $149 | 600 minutes | | Plaud Note Pro with Starter | $189 | $0 | $189 | 300 minutes | | Notta Memo with Notta Pro | $149 | $97.99 | $246.99 | 1,800 minutes | | Plaud Note Pro with Pro | $189 | $99.99 | $288.99 | 1,200 minutes | US dollar prices checked on 2 October 2026. Totals combine one device with one year of the listed plan, excluding any applicable taxes, shipping, add-ons or promotional discounts. Notta Pro uses the annual price displayed on the [Notta pricing page](https://www.notta.ai/en/pricing). Paid-plan allowances are shown independently; free hardware minutes have not been added to them. Notta Pro offers 50% more monthly transcription minutes than Plaud Pro at the displayed annual prices. That makes Notta attractive for regular use as well as occasional recording. The Hardware Plan requires an eligible account and a bound device. Workspace ownership and subscription status affect eligibility; it is different from Notta’s ordinary software-only Free plan. Translation and other features also have separate usage limits. For a fuller explanation of Plaud’s options, see [Does Plaud need a subscription?](/guides/does-plaud-need-a-subscription/). ## Who should buy the Notta Memo? Buy Notta Memo if you want our preferred recorder with a lower initial price and more free transcription time. It is the stronger value choice for buyers whose monthly recording workload fits within the Hardware Plan allowance. Notta also makes sense if you already use its software and want a dedicated recorder feeding that workspace. Its physical controls provide a direct way to start recording and select the type of conversation. Our recommendation is straightforward: start with Notta Memo unless you can identify a Plaud feature that justifies spending more. ## Who should buy the Plaud Note Pro? Buy Plaud Note Pro if longer advertised endurance or its automatic mode switching is particularly useful to you. Its Endurance mode trades pickup range for additional recording time, so the 50-hour headline needs that qualification. Plaud also deserves consideration if you already have a library of recordings in its app. Switching systems has a practical cost, even when the alternative hardware is cheaper. Those are valid reasons to choose Plaud. They do not change our overall preference for Notta Memo. ## Frequently asked questions ### Is Notta Memo better than Plaud Note Pro? Notta Memo is our overall pick based on hands-on use. Its lower price and larger free transcription allowance support that verdict, although Plaud offers longer advertised recording endurance. ### Do Notta Memo and Plaud Note Pro both need subscriptions? Neither requires a paid plan for recording and transcription within its included allowance. Eligible Notta Hardware Plan users receive 600 minutes monthly; Plaud Starter includes 300. Heavier use can require a paid upgrade. ### Which has better battery life, Notta Memo or Plaud Note Pro? Plaud Note Pro has the higher advertised maximum: 50 hours in Endurance mode, compared with Notta Memo’s 30 hours. Plaud’s Enhanced mode is also rated at 30 hours. These are manufacturer figures, not a comparative battery rundown. ## Related reading - [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/) - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Plaud vs Pocket](https://testbenchlab.com/compare/plaud-vs-pocket/) - [Plaud Note vs Plaud Note Pro](https://testbenchlab.com/compare/plaud-note-vs-plaud-note-pro/) - [Best AI voice recorders](https://testbenchlab.com/best/best-ai-voice-recorders/) - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) --- # Plaud Note Pro vs NotePin S: card or wearable? By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Verdict: Depends on use. ## Short answer Choose Plaud Note Pro for phone-based recording and NotePin S for wearing throughout your day. We prefer the Note Pro for a conventional calls-and-meetings workflow; Notta Memo remains our overall recorder choice. | Requirement | Preferred Plaud model | |---|---| | Phone-call capture | Note Pro | | Visible status on a card recorder | Note Pro | | Recording while wearing the device | NotePin S | | Overall recorder across brands | Notta Memo | ## Card or wearable: which fits how you work? A card recorder belongs with the phone or on the meeting table. A wearable belongs on you, which makes it easier to reach during a walk or an unscheduled conversation. These positions produce different audio conditions. Choose the capture situation first and the product second. ## Which should you buy? Buy Note Pro if calls and fixed-position recording dominate. Buy NotePin S if attachment and quick physical access are the main benefit. Neither device makes recording permission automatic, and a wearable should not be assumed to capture phone audio that never leaves an earpiece. ## Frequently asked questions ### Should I get Plaud Note Pro or NotePin? Get Note Pro for a phone-and-table workflow or NotePin S for wearable access. We favour Note Pro as the more general choice within Plaud. ## Related reading - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Plaud NotePin S review](https://testbenchlab.com/reviews/plaud-notepin-s-review/) - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) ## Sources 1. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) 2. [Plaud NotePin S specifications](https://www.plaud.ai/products/plaud-notepin-s) --- # Plaud Note vs Plaud Note Pro By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Plaud Note Pro. ## Short answer Plaud Note Pro is our preferred new purchase within Plaud’s card range when the premium is modest. Keep the original Note if it already meets your needs, and consider Notta Memo if you are open to changing brands. | Option | Why choose it? | |---|---| | Plaud Note | A discounted entry into Plaud or an existing device that already works | | Plaud Note Pro | Our preferred new Plaud card recorder | | Notta Memo | Our preferred recorder across brands | ## What does the Note Pro add? Plaud Note Pro adds a display, automatic mode selection, a broader microphone setup and an endurance option. The value is in how those features change capture, rather than a completely separate transcription service. Both models feed the Plaud software ecosystem. ## Is the Plaud Note Pro worth $30 more? At the published $159 versus $189 device prices, we would spend the additional $30 for the Pro on a first purchase. That conclusion is an editorial judgement about the feature difference. It does not imply that every recording or transcript becomes better by the same amount. ## Which should you buy? Buy the Pro if the small premium is affordable and you expect to use its controls. Buy the original Note at a stronger discount or keep an existing unit. Our first choice across brands is Notta Memo, so the decision need not stay inside Plaud’s range. ## Frequently asked questions ### Is Plaud Note Pro worth it over Plaud Note? For a first purchase with a $30 price gap, we favour the Pro. An existing Note owner should upgrade only for a feature they expect to use. ### Is the original Plaud Note discontinued? Plaud’s current comparison still lists it. Check stock in your country and the exact retailer offer. ## Related reading - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Plaud Note review](https://testbenchlab.com/reviews/plaud-note-review/) - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) ## Sources 1. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) --- # Plaud vs Pocket: which AI recorder should you buy? By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Our pick: Plaud Note Pro. ## Short answer We favour Plaud Note Pro over Pocket for a conventional meeting-and-call recorder. Pocket is more interesting if its advertised unlimited core processing is your priority. Our preferred recorder across the wider category remains Notta Memo. | Decision | Our recommendation | |---|---| | Conventional calls and meetings | Plaud Note Pro | | Wearable within Plaud | Plaud NotePin S | | Unlimited core processing proposition | Pocket | | Wider-category preference | Notta Memo | ## Which is better, Plaud or Pocket? Plaud is our pick for buyers who want a card recorder with visible status and a clear call-capture workflow. Pocket’s strongest buying argument is its core-processing proposition. Neither choice should be made solely from a feature count: the app is where most of the long-term value sits. ## Plaud Note Pro vs Pocket The Note Pro is the main Plaud model in this comparison because it covers conventional desk and phone use. Pocket emphasises captured conversations as a searchable memory. Choose the former for a defined recording session; consider the latter if ongoing retrieval is the feature you intend to use. ## Plaud NotePin S vs Pocket NotePin S is the closer wearable alternative, with physical recording and highlighting controls. If your first requirement is keeping the recorder on your person, compare attachment, access to the button and microphone placement. That is a separate decision from choosing between card-style call recorders. ## Subscription costs compared Plaud uses monthly transcription allowances and optional paid tiers. Pocket’s current site advertises unlimited core minutes without a subscription, alongside subscription-trial offers. That means a simple paid-plan price comparison misses an important difference: what you can keep using after the trial ends. ## Which should you buy? Choose Plaud Note Pro for the conventional recording workflow we favour here. Choose Pocket when its included processing and retrieval approach solve a specific need. Before buying either, compare Notta Memo: it is our overall preference and offers substantial free processing for eligible owners. ## Frequently asked questions ### Which is better, Plaud or Pocket? We favour Plaud Note Pro for the conventional card-recorder workflow. Pocket may fit frequent users who prioritise its advertised core processing. ### Does Pocket AI require a subscription? Pocket currently advertises unlimited core processing without a subscription. Check separate paid-feature and trial terms. ### Is Pocket cheaper than Plaud? Compare the exact device offer and the processing you need. Pocket’s core-processing model can reduce recurring cost, but a promotional hardware price alone is not the total. ## Related reading - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Plaud NotePin S review](https://testbenchlab.com/reviews/plaud-notepin-s-review/) - [Pocket AI recorder review](https://testbenchlab.com/reviews/pocket-ai-review/) - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) ## Sources 1. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) 2. [Pocket product and membership information](https://heypocket.com/) 3. [Plaud NotePin S specifications](https://www.plaud.ai/products/plaud-notepin-s) --- # Best Plaud alternatives By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. US prices checked 2026-10-02. Notta Memo is our first-choice Plaud alternative. It is the recorder David Wilson preferred in hands-on use, and the alternatives below offer different reasons to move beyond Plaud: wearable convenience, earphone-call capture or included core processing. ## Quick list | # | Pick | Product | Price | |---|---|---|---| | 1 | Preferred alternative | [Notta Memo](https://testbenchlab.com/reviews/notta-memo-review/) | $149 | | 2 | Different processing model | [Pocket](https://testbenchlab.com/reviews/pocket-ai-review/) | $139 | | 3 | Earphone calls | [HiDock P1](https://testbenchlab.com/reviews/hidock-p1-review/) | $169 | | 4 | Compact wearable | [Anker Soundcore Work](https://testbenchlab.com/reviews/soundcore-work-review/) | $159 | ## Our preferred alternative: Notta Memo Notta Memo is the closest starting point for someone considering a Plaud card recorder. It supports a dedicated calls-and-meetings workflow and feeds Notta’s note-management service. Our preference comes from use, with the account’s included processing adding a useful ownership advantage. ## For a different processing model: Pocket Pocket is worth considering when an included, uncapped core-processing proposition matters more than staying in a familiar recorder ecosystem. Confirm which features remain available without the paid tier. A trial banner and a lifetime-core claim describe different parts of the offer. ## For earphone calls: HiDock P1 HiDock P1 targets a more specific capture workflow. Its included HiNotes core features are attractive for people who record often, but headset compatibility should be checked before buying. The right alternative solves the task Plaud does not fit, rather than simply costing less. ## For a compact wearable: Soundcore Work Soundcore Work belongs on the shortlist if wearing a smaller recorder is your priority. Compare attachment and charging convenience as well as the app. A wearable format is not automatically a substitute for card-style phone-call recording. ## What to check before switching Export your existing recordings and check how long the new service keeps files, what happens when a trial ends and whether a transcript can leave the app. Switching hardware is easy; moving an archive of conversations is often the more important decision. ## Frequently asked questions ### What is better than Plaud? Our preferred alternative is Notta Memo. We would start there, then consider other devices for a specific wearable or headset-based requirement. ### Is there a cheaper alternative to Plaud? Notta Memo’s $149 standard device price is below Plaud Note Pro’s $189. Compare the processing needed as well as the hardware price. ## Related reading - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/) - [Pocket AI recorder review](https://testbenchlab.com/reviews/pocket-ai-review/) - [Anker Soundcore Work review](https://testbenchlab.com/reviews/soundcore-work-review/) - [HiDock P1 review](https://testbenchlab.com/reviews/hidock-p1-review/) ## Sources 1. [Notta Hardware Plan and eligibility](https://support.notta.ai/hc/en-us/articles/38281798150683-What-Is-the-Notta-Hardware-Plan) 2. [Pocket product and membership information](https://heypocket.com/) 3. [HiNotes included membership](https://hinotes.hidock.com/) 4. [Soundcore Work product and plan information](https://www.soundcore.com/uk/products/d3200-soundcore-work-ai-voice-recorder) --- # The best AI voice recorders of 2026 By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. US prices checked 2026-10-02. Notta Memo is our preferred AI voice recorder for 2026. David Wilson’s hands-on preference anchors the recommendation; the other picks below are researched alternatives for particular workflows, rather than a claimed common-bench ranking. ## Quick list | # | Pick | Product | Price | |---|---|---|---| | 1 | Best overall | [Notta Memo](https://testbenchlab.com/reviews/notta-memo-review/) | $149 | | 2 | Best Plaud card recorder | [Plaud Note Pro](https://testbenchlab.com/reviews/plaud-note-pro-review/) | $189 | | 3 | Best wearable | [Plaud NotePin S](https://testbenchlab.com/reviews/plaud-notepin-s-review/) | $179 | | 4 | Unlimited core processing | [Pocket](https://testbenchlab.com/reviews/pocket-ai-review/) | $139 | | 4 | Unlimited core processing | [HiDock P1](https://testbenchlab.com/reviews/hidock-p1-review/) | $169 | ## Best overall: Notta Memo Notta Memo is the first recorder we would recommend. The dedicated hardware suits scheduled meetings and calls, while its eligible Hardware Plan offers room for regular transcription before upgrading. The recommendation reflects David Wilson’s use of the product; it is not a numerical accuracy award. ## Best Plaud card recorder: Plaud Note Pro Plaud Note Pro is the Plaud model we would buy for a conventional card-recorder setup. It combines visible recording status with automatic mode selection and an endurance option. Choose it for those capture features rather than assuming its higher price guarantees better notes. ## Best wearable in our shortlist: Plaud NotePin S NotePin S suits people who want the recording control on their clothing or wrist. We favour its tactile interaction within this wearable shortlist. A wearable is less useful when your main task is direct phone-call capture, so do not treat form factor as a cosmetic choice. ## For unlimited core processing: Pocket and HiDock P1 Pocket advertises unlimited core minutes, while HiDock includes core HiNotes transcription and summaries. Their value depends on the workflow you need: conversation retrieval for Pocket, or supported earphone-call capture for P1. Read the included-feature terms alongside the hardware description. ## How to choose Start with what you record, then estimate monthly processing time. Check export, retention and attachment before committing to an ecosystem. Five hours of included transcription may be enough for occasional interviews and too little for a busy meeting calendar. The best recorder is the one whose complete workflow you keep using. ## Frequently asked questions ### Which AI voice recorder is best? Notta Memo is our overall preference. Different form factors and processing plans can make another recorder a better specialist fit. ### Are AI voice recorders legal? Recording rules depend on consent, location and context. The device being sold legally does not authorise every recording. ### Is there an AI voice recorder with no subscription? Several include core processing or a free monthly allowance. Pocket and HiDock advertise included core access; Notta offers an eligible Hardware Plan. ### What is the best AI voice recorder for meetings? Notta Memo is our preferred general meeting recorder. Place any recorder where the group can be heard, rather than next to only one speaker. ### Are AI voice recorders any good? They can make audio searchable and easier to revisit. Check significant details against the recording before relying on an AI summary. ## Related reading - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/) - [Plaud NotePin S review](https://testbenchlab.com/reviews/plaud-notepin-s-review/) - [Pocket AI recorder review](https://testbenchlab.com/reviews/pocket-ai-review/) - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) - [One-party vs two-party consent states: can you use an AI recorder?](https://testbenchlab.com/guides/recording-consent-laws-by-state/) ## Sources 1. [Notta Hardware Plan and eligibility](https://support.notta.ai/hc/en-us/articles/38281798150683-What-Is-the-Notta-Hardware-Plan) 2. [Pocket product and membership information](https://heypocket.com/) 3. [HiNotes included membership](https://hinotes.hidock.com/) 4. [Plaud NotePin S specifications](https://www.plaud.ai/products/plaud-notepin-s) --- # The best mini PCs for AI in 2026 By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. US prices checked 2026-10-02. Geekom leads our mini PC recommendations for 2026. Choose A9 Max for an everyday computer that also runs smaller local models, or A9 Mega when larger-model memory is the main requirement. This is a researched shortlist, not a claimed shared-bench ranking. ## Quick list | # | Pick | Product | Price | |---|---|---|---| | 1 | Best everyday AI desktop | [Geekom A9 Max](https://testbenchlab.com/reviews/geekom-a9-max-review/) | from $1,499 | | 2 | Best for larger local models | [Geekom A9 Mega](https://testbenchlab.com/reviews/geekom-a9-mega-review/) | $3,799 | | 3 | Best for native eGPU expansion | [Minisforum AI X1 Pro](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) | from $1,199 | | 4 | Best for networking and expansion | [Minisforum MS-S1 Max](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) | $3,799 | ## Best everyday AI desktop: Geekom A9 Max A9 Max is our general starting point because it combines an ordinary desktop role with an upgradeable memory route. Choose the RAM arrangement with care and confirm that your runtime uses the Radeon GPU. Cloud-based assistants alone do not justify paying for the maximum local-AI configuration. ## Best for larger local models: Geekom A9 Mega A9 Mega moves to the Strix Halo platform. Choose it for the memory-and-graphics proposition rather than as a default upgrade for every office task. The amount of accelerator-visible memory and your intended model context should determine capacity. ## Best for native eGPU expansion: Minisforum AI X1 Pro A native OCuLink connection is useful when a discrete card is part of a costed setup. The card changes the relevant memory limit and the whole ownership cost. If you do not intend to use an eGPU, choose the mini PC on its own merits. ## Best for networking and internal expansion: Minisforum MS-S1 Max MS-S1 Max makes sense when fast networking or a compatible internal card has a role. Two connected systems do not automatically pool memory for an ordinary model runner. Buy the infrastructure only when the software and workload can use it. ## How we chose The shortlist prioritises workload fit, usable memory, software compatibility and relevant expansion. Geekom is our editorial preference. We do not turn unmatched external results into numerical scores or describe this group as completing a common Testbench Lab protocol. ## Frequently asked questions ### Which mini PC is good for AI? Geekom A9 Max is our preferred everyday option, and A9 Mega is our choice for larger local models. ### What is the best budget mini PC with AI capabilities? Choose sufficient memory for a small model and a supported runtime. For cloud AI, a capable ordinary desktop can be enough. ### What are the best mini PCs for AI agents? Choose according to whether the agent calls a cloud service or hosts a local model. Browser tools, containers and concurrent sessions can add requirements beyond the model itself. ### Can a mini PC run a local LLM? Yes, with a model that fits the available memory and software that supports the processor or GPU. ### How much RAM does a mini PC need for AI? 32GB is a useful smaller-model starting point. Larger models and longer context can require 64GB or 128GB, plus a backend able to use that capacity. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) - [Beelink SER9 Pro review](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) - [How much RAM do you need to run a local LLM?](https://testbenchlab.com/guides/how-much-ram-for-local-llm/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 3. [Minisforum AI X1 Pro-470](https://store.minisforum.com/products/minisforum-ai-x1-pro-470-mini-pc) 4. [Minisforum MS-S1 Max](https://store.minisforum.com/products/minisforum-ms-s1-max-mini-pc) --- # The best Ryzen AI 9 HX 370 and HX 470 mini PCs By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. US prices checked 2026-10-02. Geekom A9 Max is our preferred Ryzen AI 9 HX 370/470 mini PC. This tier works best as an everyday desktop with local-AI capability, rather than a substitute for the larger shared-memory Strix Halo platform. ## Quick list | # | Pick | Product | Price | |---|---|---|---| | 1 | Our first choice | [Geekom A9 Max](https://testbenchlab.com/reviews/geekom-a9-max-review/) | from $1,499 | | 2 | For OCuLink | [Minisforum AI X1 Pro](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) | from $1,199 | | 2 | For OCuLink | [GMKtec EVO-X1](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) | $1,149.99 | | 3 | Fixed-memory workload | [Beelink SER9 Pro](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) | $1,149 | ## Our first choice: Geekom A9 Max Prioritise the exact memory arrangement and a supported graphics backend. HX 470 is attractive when the full configuration costs little more; HX 370 can remain a good option when the saving funds useful RAM or storage. Our preference concerns the complete buying decision, not an invented inference ranking. ## For OCuLink: Minisforum AI X1 Pro or GMKtec EVO-X1 These alternatives are most relevant when external graphics is part of the plan. Compare the dock, card and power supply as a complete setup. Buying a connector without knowing which GPU and model will use it can add cost without solving a real limit. ## For a settled fixed-memory workload: Beelink SER9 Pro The HX 370 variant uses onboard memory, so capacity is a purchase decision. Other processors also appear under the SER9 Pro label. Confirm the actual chip before applying a review or comparing a sale price. ## What matters more than the new CPU label Model fit, memory channels, supported drivers and actual hardware use come first. A newer chip running on the CPU can lose the practical advantage you expected from AI marketing. Check the backend before spending more to solve what may be a software configuration problem. ## Frequently asked questions ### Is the Ryzen AI 9 HX 370 good for AI? It is useful for smaller local models with a supported backend. NPU capability does not automatically accelerate every application. ### HX 370 or HX 470: which should I buy? Choose HX 470 for a modest premium in an otherwise matched configuration. A better RAM setup or a substantial discount can make HX 370 the better buy. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) - [Beelink SER9 Pro review](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) - [Ryzen AI 9 HX 470 vs HX 370: what changed?](https://testbenchlab.com/guides/ryzen-ai-9-hx-470-vs-hx-370/) ## Sources 1. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 2. [Minisforum AI X1 Pro-470](https://store.minisforum.com/products/minisforum-ai-x1-pro-470-mini-pc) 3. [GMKtec EVO-X1](https://www.gmktec.com/products/amd-ryzen%E2%84%A2-ai-9-hx-370-evo-x1-ai-mini-pc) 4. [Beelink SER9 Pro HX 370](https://www.bee-link.com/products/beelink-ser9-ai-9-hx-370) 5. [AMD Ryzen AI 400 announcement and specifications](https://ir.amd.com/news-events/press-releases/detail/1270/amd-expands-ai-leadership-across-client-graphics-and-software-with-new-ryzen-ryzen-ai-and-amd-rocm-announcements-at-ces-2026) --- # The best Strix Halo mini PCs By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. US prices checked 2026-10-02. Geekom A9 Mega is our preferred Strix Halo mini PC. The right alternative depends on matched memory, expansion and network needs; the Max+ 395 chip shared by several models does not make their complete configurations identical. ## Quick list | # | Pick | Product | Price | |---|---|---|---| | 1 | Our first choice | [Geekom A9 Mega](https://testbenchlab.com/reviews/geekom-a9-mega-review/) | $3,799 | | 2 | Matched lower-cost offer | [GMKtec EVO-X2](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) | $3,649.99 | | 3 | For infrastructure | [Minisforum MS-S1 Max](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) | $3,799 | | 4 | Premium network configuration | [Beelink GTR9 Pro](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/) | $4,349 | ## Our first choice: Geekom A9 Mega A9 Mega is our starting point for a single-user workstation. Check the exact processor variant and installed memory, then confirm that your model runner can use the allocation. Capacity selected at purchase is especially important because this class uses onboard memory. ## Alternative for a matched lower-cost offer: GMKtec EVO-X2 EVO-X2 is relevant when the same intended model fits at a better total cost. Compare memory and storage rather than a headline starting price. A smaller-capacity bargain can become the wrong purchase if the model you want leaves too little room for context. ## Alternative for infrastructure: Minisforum MS-S1 Max MS-S1 Max serves buyers who can use fast networking and internal expansion. Treat those as workload features. An additional network interface cannot make a model fit into insufficient memory, and clustering requires appropriate software. ## Alternative for a premium network configuration: Beelink GTR9 Pro GTR9 Pro is worth checking for the exact memory and networking package offered. Verify the actual board and seller terms. We prefer Geekom overall, while recognising that a necessary interface can determine the purchase. ## What to match before comparing results Use the same model, quantisation, context and backend. Record memory allocation and power mode. Without those conditions, a tokens-per-second figure is a result for one setup, rather than a reliable rank for the chassis. ## Frequently asked questions ### Which Strix Halo mini PC is the best? Geekom A9 Mega is our preferred overall choice for a general local workstation. ### Who makes Strix Halo mini PCs? Our shortlist includes Geekom A9 Mega, GMKtec EVO-X2, Minisforum MS-S1 Max and Beelink GTR9 Pro. ### Is Strix Halo good for gaming? Its larger integrated graphics platform is relevant for gaming, but settings, memory and shared power limits still determine performance. ### What is Strix Halo equivalent to? There is no single universal desktop-GPU equivalent. Workload, power and memory behaviour change the comparison. ## Related reading - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) - [Beelink GTR9 Pro review](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/) - [Geekom A9 Mega vs GMKtec EVO-X2](https://testbenchlab.com/compare/geekom-a9-mega-vs-gmktec-evo-x2/) - [Geekom A9 Mega vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/geekom-a9-mega-vs-minisforum-ms-s1-max/) - [GMKtec EVO-X2 vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/gmktec-evo-x2-vs-minisforum-ms-s1-max/) - [Ryzen AI Max+ 395 (Strix Halo) explained](https://testbenchlab.com/guides/ryzen-ai-max-395-strix-halo-explained/) ## Sources 1. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 2. [GMKtec EVO-X2](https://www.gmktec.com/en/products/amd-ryzen%E2%84%A2-ai-max-395-evo-x2-ai-mini-pc) 3. [Minisforum MS-S1 Max](https://store.minisforum.com/products/minisforum-ms-s1-max-mini-pc) 4. [Beelink GTR9 Pro](https://www.bee-link.com/products/beelink-gtr9-pro-amd-ryzen-ai-max-395) 5. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) --- # AI mini PCs An AI mini PC can be an everyday desktop, a local assistant or a compact model server. Our preferred brand is Geekom: A9 Max is the general desktop choice, while A9 Mega is the more suitable starting point for large local models. ## Our picks - Everyday AI desktop: [Geekom A9 Max](https://testbenchlab.com/reviews/geekom-a9-max-review/) — from $1,499 - Larger local models: [Geekom A9 Mega](https://testbenchlab.com/reviews/geekom-a9-mega-review/) — $3,799 - Native eGPU expansion: [Minisforum AI X1 Pro](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) — from $1,199 ## All reviews | Product | Chip tier | Assessment | Price | |---|---|---|---| | [Beelink SER9 Pro](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) | Ryzen AI 9 HX 370 | Researched | $1,149 | | [GMKtec EVO-X1](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) | Ryzen AI 9 HX 370 | Researched | $1,149.99 | | [Minisforum AI X1 Pro](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) | Ryzen AI 9 HX 370/470 | Researched | from $1,199 | | [Geekom A9 Max](https://testbenchlab.com/reviews/geekom-a9-max-review/) | Ryzen AI 9 HX 370/470 | Researched | from $1,499 | | [GMKtec EVO-X2](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) | Ryzen AI Max+ 395 | Researched | $3,649.99 | | [Geekom A9 Mega](https://testbenchlab.com/reviews/geekom-a9-mega-review/) | Ryzen AI Max+ 395 | Researched | $3,799 | | [Minisforum MS-S1 Max](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) | Ryzen AI Max+ 395 | Researched | $3,799 | | [Beelink GTR9 Pro](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/) | Ryzen AI Max+ 395 | Researched | $4,349 | ## Comparisons - [Geekom A9 Max vs Beelink SER9 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-beelink-ser9-pro/) - [Geekom A9 Max vs GMKtec EVO-X1](https://testbenchlab.com/compare/geekom-a9-max-vs-gmktec-evo-x1/) - [Geekom A9 Max vs Minisforum AI X1 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-minisforum-ai-x1-pro/) - [Geekom A9 Mega vs Beelink GTR9 Pro](https://testbenchlab.com/compare/geekom-a9-mega-vs-beelink-gtr9-pro/) - [Geekom A9 Mega vs GMKtec EVO-X2](https://testbenchlab.com/compare/geekom-a9-mega-vs-gmktec-evo-x2/) - [Geekom A9 Mega vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/geekom-a9-mega-vs-minisforum-ms-s1-max/) - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/) - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/) - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/) - [GMKtec EVO-X2 vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/gmktec-evo-x2-vs-minisforum-ms-s1-max/) ## Buying guides - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) - [The best Strix Halo mini PCs](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) ## Guides - [NVIDIA DGX Spark vs Strix Halo mini PCs](https://testbenchlab.com/guides/dgx-spark-vs-strix-halo/) - [How much RAM do you need to run a local LLM?](https://testbenchlab.com/guides/how-much-ram-for-local-llm/) - [How to install Ollama on Windows 11 (Ryzen AI mini PC)](https://testbenchlab.com/guides/install-ollama-windows-11-ryzen-ai-mini-pc/) - [llama.cpp with Vulkan on a Radeon iGPU](https://testbenchlab.com/guides/llama-cpp-vulkan-radeon-igpu/) - [Mini PC vs Mac mini for AI](https://testbenchlab.com/guides/mini-pc-vs-mac-mini-for-ai/) - [NPU vs GPU: which runs AI better on a mini PC?](https://testbenchlab.com/guides/npu-vs-gpu/) - [Radeon 890M vs Radeon 8060S: what the iGPU means for AI and gaming](https://testbenchlab.com/guides/radeon-890m-vs-radeon-8060s/) - [Run local LLMs on the Ryzen AI NPU with Lemonade Server](https://testbenchlab.com/guides/run-llms-on-ryzen-ai-npu-lemonade-server/) - [Ryzen AI 9 HX 470 vs HX 370: what changed?](https://testbenchlab.com/guides/ryzen-ai-9-hx-470-vs-hx-370/) - [Ryzen AI Max+ 395 (Strix Halo) explained](https://testbenchlab.com/guides/ryzen-ai-max-395-strix-halo-explained/) - [What is an AI PC? (and is a Copilot+ PC different?)](https://testbenchlab.com/guides/what-is-an-ai-pc/) - [What is an NPU, and does your PC need one?](https://testbenchlab.com/guides/what-is-an-npu/) - [What is OCuLink? OCuLink vs USB4 for an external GPU](https://testbenchlab.com/guides/what-is-oculink/) - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) ## Which tier do you need: Ryzen AI 9 HX 370/470 or Ryzen AI Max+ 395? The HX tier suits a compact desktop that also runs smaller local models. Strix Halo systems provide a substantially wider graphics and memory platform for more demanding workloads. Choose according to the model and software you intend to run; an NPU number alone cannot identify the right tier. ## How much memory do you need for local AI? Allow room for model weights, context, runtime and the operating system. A model fitting in total system memory does not guarantee it fits in the memory available to the chosen accelerator. Start with a smaller model, verify hardware use and increase capacity only when the workload calls for it. ## Mini PC or Mac mini for AI? Choose a Windows or Linux mini PC when that is the environment you need. Apple silicon offers a different software path through Metal and MLX. Compare the exact memory configuration and supported backend, rather than treating the operating system as a minor difference after purchase. ## Frequently asked questions ### Which mini PC is good for AI? Geekom A9 Max is our preferred general desktop, and A9 Mega is our preferred starting point for larger local models. ### What is the best budget mini PC with AI capabilities? For a limited budget, choose enough memory for a small model and verify software compatibility. A premium HX or Strix Halo machine is unnecessary for cloud-only AI use. ### Can a mini PC run a local LLM? Yes, when the model fits available memory and the runtime supports the hardware. Practical speed depends on more than the AI label. ### Who makes Strix Halo mini PCs? The planned shortlist includes Geekom, Minisforum, GMKtec and Beelink. Match the processor and memory variant before comparing offers. --- # AI voice recorders An AI voice recorder captures audio with dedicated hardware and uses software to turn it into transcripts and summaries. Our preferred choice is Notta Memo; the right alternative depends on whether you need a wearable, phone-call capture or a different subscription model. ## Our picks - Our pick: [Notta Memo](https://testbenchlab.com/reviews/notta-memo-review/) — $149 - Best Plaud card recorder: [Plaud Note Pro](https://testbenchlab.com/reviews/plaud-note-pro-review/) — $189 - Best wearable: [Plaud NotePin S](https://testbenchlab.com/reviews/plaud-notepin-s-review/) — $179 ## All reviews | Product | Type | Assessment | Price | |---|---|---|---| | [Pocket](https://testbenchlab.com/reviews/pocket-ai-review/) | Card recorder (MagSafe) | Researched | $139 | | [Notta Memo](https://testbenchlab.com/reviews/notta-memo-review/) | Card recorder | Hands-on | $149 | | [Plaud Note](https://testbenchlab.com/reviews/plaud-note-review/) | Card recorder | Researched | $159 | | [Anker Soundcore Work](https://testbenchlab.com/reviews/soundcore-work-review/) | Wearable | Researched | $159 | | [HiDock P1](https://testbenchlab.com/reviews/hidock-p1-review/) | Earphone call recorder | Researched | $169 | | [Plaud NotePin S](https://testbenchlab.com/reviews/plaud-notepin-s-review/) | Wearable pin | Researched | $179 | | [Plaud Note Pro](https://testbenchlab.com/reviews/plaud-note-pro-review/) | Card recorder | Hands-on | $189 | ## Comparisons - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) - [Plaud Note Pro vs NotePin S: card or wearable?](https://testbenchlab.com/compare/plaud-note-pro-vs-notepin-s/) - [Plaud Note vs Plaud Note Pro](https://testbenchlab.com/compare/plaud-note-vs-plaud-note-pro/) - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) ## Buying guides - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) ## Guides - [Does Plaud need a subscription? Plans and free minutes explained](https://testbenchlab.com/guides/does-plaud-need-a-subscription/) - [Does Pocket AI need a subscription? What is free and what is not](https://testbenchlab.com/guides/does-pocket-ai-need-a-subscription/) - [How to record a phone call on iPhone (built-in, apps and AI recorders)](https://testbenchlab.com/guides/how-to-record-a-phone-call-on-iphone/) - [How to record a Microsoft Teams meeting](https://testbenchlab.com/guides/how-to-record-a-teams-meeting/) - [How to record a Zoom meeting (and get an AI transcript)](https://testbenchlab.com/guides/how-to-record-a-zoom-meeting/) - [How to use the Plaud Note Pro: setup to first summary](https://testbenchlab.com/guides/how-to-use-plaud-note-pro/) - [Is Plaud HIPAA compliant?](https://testbenchlab.com/guides/is-plaud-hipaa-compliant/) - [One-party vs two-party consent states: can you use an AI recorder?](https://testbenchlab.com/guides/recording-consent-laws-by-state/) ## What is an AI voice recorder? An AI voice recorder separates capture from processing. The device stores sound, while its companion service recognises speech and organises the text. Offline recording does not necessarily mean offline transcription. Consider where processing happens and whether you can export both the original audio and the finished notes. ## Do AI voice recorders need a subscription? Some include a free allowance, while others advertise unlimited core processing with the hardware. Notta’s eligible Hardware Plan owners receive 600 transcription minutes monthly. Paid plans may add usage or features. Compare the allowance with your actual recording workload, rather than assuming every recorder needs a recurring payment. ## Card recorder or wearable pin? A card recorder suits a phone-based workflow or a fixed position on the table. A wearable is easier to keep available while moving, but clothing and body position affect capture. A phone-call recording feature must be checked separately; wearing a recorder does not give it access to audio inside your phone. ## Frequently asked questions ### Are AI voice recorders any good? They can save time finding and summarising conversations. Their usefulness depends on the audio, the transcription service and whether the notes remain easy to export and revisit. ### Are AI voice recorders legal? The hardware itself does not decide legality. Consent, location, workplace rules and how the recording is used all matter. See our recording-consent guide before recording other people. ### Can an AI voice recorder record phone calls? Some card recorders, including Notta Memo and Plaud Note Pro, provide a call-capture mode. Many wearables are designed primarily for speech in the surrounding room. ### Is there an AI voice recorder without a subscription? Yes. Some devices include core processing without a recurring fee, and others offer a limited free allowance. Included features and eligibility still matter. --- # Does Plaud need a subscription? Plans and free minutes explained Explainer · AI recorders · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. You can use Plaud without a paid subscription within its included transcription allowance. A paid plan becomes useful when you need more processing time; paying for a higher tier does not change the microphone hardware you already own. | Plaud plan | Monthly transcription | Annual subscription | |---|---:|---:| | Starter | 300 minutes | $0 | | Pro | 1,200 minutes | $99.99 | | Unlimited | Unlimited under plan terms | $239.99 | US prices checked 2 October 2026. ## Can you use Plaud without a subscription? Plaud’s included Starter plan provides 300 transcription minutes monthly. You can record audio independently of that cloud-processing quota. Estimate the minutes you actually intend to transcribe before paying for a tier you may not need. ## Plaud plans compared (Starter, Pro, Unlimited) Starter provides a limited allowance, Pro expands it and Unlimited supports a heavier workload under its usage terms. Check the current feature table as well as minutes: plan names alone do not identify every tool, export option or administrative control. ## How much does Plaud cost per year? The annual Pro price shown is $99.99 and Unlimited is $239.99. With the $189 Note Pro device, first-year totals are $288.99 and $428.99 respectively, before tax, shipping or discounts. Hardware is a first-year purchase; subscription renewal is a separate ongoing cost. ## Is the Plaud subscription worth it? It is worthwhile when the time saved exceeds the recurring cost and the free quota is insufficient. If you are still choosing hardware, our preference is Notta Memo. Choose the service around your actual workload rather than subscribing because the recorder has an AI label. ## Frequently asked questions ### Can you use Plaud Note Pro without a subscription? Yes. Use the included Starter allowance and upgrade only if your processing needs exceed it. ### What happens when you run out of Plaud minutes? Further cloud transcription requires additional available quota, an upgrade or the next allowance renewal. Recording audio is separate from consuming transcription minutes. ### Does Plaud have a free plan? Yes, Starter is included with eligible Plaud hardware and currently provides 300 monthly transcription minutes. ## Related reading - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Plaud NotePin S review](https://testbenchlab.com/reviews/plaud-notepin-s-review/) - [Plaud Note review](https://testbenchlab.com/reviews/plaud-note-review/) - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) ## Sources 1. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) --- # Does Pocket AI need a subscription? What is free and what is not Explainer · AI recorders · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Pocket’s current product site says core features and unlimited minutes do not require a subscription. Paid extras and subscription trials are separate, so read the offer as two questions: what stays included, and what renews at a cost. ## Does Pocket AI work without a subscription? Pocket advertises lifetime core access without a monthly minute cap. That is the relevant starting point if you simply want to capture and process conversations. Included access still depends on the service continuing to operate and on its current terms; it is not the same thing as fully offline processing. ## What Pocket Pro adds Paid access is for capabilities beyond the advertised core. The current public offer does not expose a sufficiently stable, complete dollar-price table to reproduce here. Check the plan presented with your order, including the features, trial duration and renewal amount. A founding-member offer may differ from a new purchase. ## Pocket vs Plaud subscription cost Plaud offers included monthly minutes and priced tiers for heavier processing. Pocket’s advertised core model can avoid that same minute-based calculation. Choose according to the features you need after the trial, and include hardware, tax and any recurring paid extras in your total. ## Frequently asked questions ### How much is the Pocket AI subscription? The renewal price depends on the current paid offer and account package. Read the amount shown before accepting a trial; core features are currently advertised without a subscription. ### Is Pocket AI worth it? It is worth considering for frequent processing and retrieval. Notta Memo remains our preferred recorder for general use. ## Related reading - [Pocket AI recorder review](https://testbenchlab.com/reviews/pocket-ai-review/) - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) - [Does Plaud need a subscription? Plans and free minutes explained](https://testbenchlab.com/guides/does-plaud-need-a-subscription/) ## Sources 1. [Pocket product and membership information](https://heypocket.com/) 2. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) --- # How much RAM do you need to run a local LLM? Explainer · Local AI · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Budget local-model memory for weights, context cache and runtime, not just the download. 32GB is a useful smaller-model starting point; larger dense models can require 64GB or 128GB depending on quantisation and context. | Dense model | Approximate Q4 weights | System-memory planning range | |---|---:|---:| | 7B | 4–5GB | 16–32GB | | 14B | 8–10GB | 24–32GB | | 32B | 18–22GB | 32–64GB | | 70B | 40–48GB | 64–128GB | Editorial estimates for typical four-bit formats and moderate context. Architecture, cache precision and runtime change the requirement. ## How much RAM for 7B, 14B, 32B and 70B models? Use the table as a planning range, not a guaranteed minimum. Four-bit storage is roughly half a byte per parameter before overhead. Longer context, multimodal inputs and concurrent sessions can exceed these suggested ranges. ## What quantisation does to memory Quantisation reduces weight precision and file size, with possible quality and speed trade-offs. A theoretical four-bit 70B payload is about 35GB before overhead. Actual formats also store scales, metadata and higher-precision tensors; runtime and context require more capacity. ## Which mini PC tier fits which model? We prefer Geekom A9 Max for smaller-model desktop work and A9 Mega for larger shared-memory workloads. Verify accelerator-visible capacity. Fitting weights in system RAM is different from keeping the intended workload on the GPU at useful speed. ## Frequently asked questions ### Can 32GB of RAM run a 70B model? A conventional four-bit dense 70B workload exceeds that capacity after weights and working memory are included. More aggressive formats change the trade-offs. ### Is 64GB enough for local AI? It is enough for many quantised workloads, but not every model or context. Leave room for system work and check GPU-visible memory. ## Related reading - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) ## Sources 1. [llama.cpp build documentation](https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md) 2. [Ollama hardware support](https://docs.ollama.com/gpu) 3. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 4. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) --- # How to install Ollama on Windows 11 (Ryzen AI mini PC) How-to · Local AI · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Install Ollama, run a small model and check which processor is used. A successful response does not prove Radeon acceleration. Current documentation includes a Vulkan route for additional GPU support. **Requirements:** Windows 11, current graphics drivers, storage and internet for initial downloads. This procedure follows upstream documentation; it is not a claimed execution test on a specific device. ## Install Ollama Download the official Windows installer from ollama.com. After installation, open a new PowerShell window and run: ```powershell ollama --version ollama run qwen2.5:3b ``` The first run downloads model data. Complete the download before judging response speed. ## Make Ollama use the Radeon iGPU Install the appropriate AMD graphics driver, restart Ollama and run a prompt. Current documentation says Vulkan is enabled by default when the backend is installed. Inspect `ollama ps` while the model is loaded. CPU-only use calls for a driver/backend check; the NPU is a separate path. ## Pick a model that fits your memory Begin with the 3B example. Context and other applications increase memory use beyond downloaded weights. Close unnecessary workloads and reduce context before assuming a larger computer is the only answer to an allocation error. ## Add a chat UI (Open WebUI) With uv installed, use the documented Windows route: ```powershell $env:DATA_DIR="C:\open-webui\data" uvx --python 3.11 open-webui@latest serve ``` Open the address printed by the process and connect to Ollama at `http://localhost:11434`. Keep access local unless you deliberately configure authenticated wider use. ## Frequently asked questions ### Does Ollama work on AMD GPUs? Yes, through supported backends and drivers. Check current platform support and verify actual offload. ### How much RAM do I need for Ollama? Small quantised models use modest memory, but leave room for Windows and context. 32GB is a useful starting point for broader experimentation. ## Related reading - [How much RAM do you need to run a local LLM?](https://testbenchlab.com/guides/how-much-ram-for-local-llm/) - [Run local LLMs on the Ryzen AI NPU with Lemonade Server](https://testbenchlab.com/guides/run-llms-on-ryzen-ai-npu-lemonade-server/) - [llama.cpp with Vulkan on a Radeon iGPU](https://testbenchlab.com/guides/llama-cpp-vulkan-radeon-igpu/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) ## Sources 1. [Ollama Windows documentation](https://docs.ollama.com/windows) 2. [Ollama hardware support](https://docs.ollama.com/gpu) 3. [Open WebUI quick start](https://docs.openwebui.com/getting-started/quick-start/) --- # How to record a Microsoft Teams meeting How-to · AI recorders · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. In a supported Teams meeting, open More actions, choose Record and transcribe, then start recording. Whether you can do this depends on your licence, organisation’s policy and meeting role; participants are notified when recording begins. ## How to record a Teams meeting Join the meeting, open More actions and select Record and transcribe → Start recording. Confirm the prompt. Use the same menu to stop recording. Check the meeting chat after processing and play a short section to verify the saved audio. ## Who can record in Teams? Access depends on the organiser’s settings, eligible licensing and administrator policy. Guests or people from another organisation may not have meeting-recording permission. If the option is absent, ask the organiser or IT administrator to check the role and policy rather than treating it as a broken microphone. ## Where Teams recordings are saved Teams stores recordings in OneDrive or SharePoint according to the meeting type. The meeting chat or channel provides the practical route to the finished recording. Download and sharing permissions, as well as expiry settings, may be controlled by the organisation. ## Getting a transcript Use Record and transcribe to start a supported live transcript and confirm the spoken language. If an approved external tool is needed, obtain the audio through the permitted export route. Our preferred general note workflow is Notta, but organisational data rules take precedence over a personal product preference. ## Frequently asked questions ### Why can't I record a Teams meeting? Check your licence, administrator recording policy, meeting role and organiser settings. A guest account or external organisation can limit recording permission. ### Does Teams notify people when you record? Yes. Teams notifies meeting participants when recording starts. ## Related reading - [How to record a Zoom meeting (and get an AI transcript)](https://testbenchlab.com/guides/how-to-record-a-zoom-meeting/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) ## Sources 1. [Microsoft Teams recording instructions](https://support.microsoft.com/en-us/teams/meetings/start-stop-and-find-meeting-recordings-in-microsoft-teams) 2. [Microsoft Teams recording policies](https://learn.microsoft.com/en-us/microsoftteams/teams-recording-policy) --- # How to record a phone call on iPhone (built-in, apps and AI recorders) How-to · AI recorders · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Use iPhone’s built-in call recording when it is available for your device, language and region. If you need a dedicated recorder workflow, Notta Memo is our preferred hardware option. Recording permission still needs to be addressed whichever method you choose. ## How to record a call with the built-in iPhone feature On a supported iPhone, start the call and open the call controls. Choose Call Recording from the available controls and allow the announcement to finish. Stop recording when needed. Find the saved recording in Notes and check whether a transcript is available. Feature availability depends on the current iOS, language and regional support. ## Does the other person know you are recording? Apple’s built-in feature announces recording to the participants. A hardware recorder or a third-party service may behave differently, so make disclosure part of your own routine. A screen indicator visible only to you is not the same as telling the other caller. ## Recording calls with an AI recorder (Plaud Note Pro, Notta Memo) Notta Memo is our preferred choice; Plaud Note Pro is another card-recorder option. Attach the device as instructed, choose or verify call mode and make a short test call. Do not assume an ordinary room microphone can hear audio confined to an earpiece or Bluetooth headset. ## Is it legal to record a phone call? Permission rules differ by jurisdiction and can be more complicated when callers are in different places. Ask everyone for permission before recording and check the rules that apply to the conversation. Our state-law guide explains why a one-party label is not a universal exemption. ## Frequently asked questions ### Can you record a phone call on iPhone without an app? Yes, where Apple’s built-in feature is available. Device, software, language and region affect access. ### Does iPhone notify the other person when you record? Apple’s built-in recording feature announces recording. Do not assume that third-party tools behave the same way. ### Can Plaud record phone calls? Supported Plaud card recorders provide a call-capture mode. Set up the attachment correctly and verify the audio with a test call. ## Related reading - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/) - [One-party vs two-party consent states: can you use an AI recorder?](https://testbenchlab.com/guides/recording-consent-laws-by-state/) - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) ## Sources 1. [Apple call recording guide](https://support.apple.com/guide/iphone/record-and-transcribe-a-call-iph57c6590e9/ios) 2. [Notta Memo specifications](https://shop.notta.ai/en-us/products/notta-memo) 3. [Reporters Committee state recording-law guide](https://www.rcfp.org/reporters-recording-guide/) --- # How to record a Zoom meeting (and get an AI transcript) How-to · AI recorders · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Use Zoom’s recording controls to capture a meeting, then obtain a transcript through the supported cloud feature or an approved transcription service. Recording permission, file ownership and transcript access should be agreed before the meeting starts. ## How to record a Zoom meeting as the host In the desktop app, start the meeting and select Record. Choose computer or cloud recording when your account offers both. Stop when the session is finished and allow processing to complete. Make a short test before an important meeting to confirm audio and the view being captured. ## How to record a Zoom meeting as a participant Ask the host for permission to record to your computer, or ask them to make and share a cloud recording. Host and account settings determine which options appear. If the controls are unavailable, check permissions before assuming that another recording method is authorised. ## Where Zoom saves recordings Computer recordings are converted after the meeting and saved to the configured local folder. Cloud recordings appear in the account’s recording area after processing. Locate the finished files before closing down your workflow, and check who is allowed to view or download them. ## How to get a transcript of a Zoom meeting Use the supported cloud audio-transcription feature if your plan and settings include it. Otherwise export the audio and upload it to an approved transcription service. Notta is our preferred general note workflow. Direct audio capture is preferable to putting a room recorder beside laptop speakers. ## Frequently asked questions ### Can you record a Zoom meeting without the host knowing? Use the host-authorised recording workflow and make participants aware. The existence of another capture method does not establish permission. ### Can you record a Zoom meeting on the free plan? Zoom supports computer recording in eligible desktop sessions when recording is enabled. Cloud recording and transcription have separate plan requirements. ### How do I get a transcript of a Zoom recording? Use the account’s available cloud transcription feature, or process exported audio in an approved service. ## Related reading - [How to record a Microsoft Teams meeting](https://testbenchlab.com/guides/how-to-record-a-teams-meeting/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [One-party vs two-party consent states: can you use an AI recorder?](https://testbenchlab.com/guides/recording-consent-laws-by-state/) ## Sources 1. [Zoom computer recording instructions](https://support.zoom.com/hc/en/article?id=zm_kb&sysparm_article=KB0076922) --- # How to use the Plaud Note Pro: setup to first summary How-to · AI recorders · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Set up Plaud Note Pro in the official app, make a short recording and check the full capture-to-summary workflow before relying on it. The first useful test is hearing the audio and finding the corresponding transcript, rather than exploring every template. ## Set up the Plaud Note Pro Charge the device, install the official Plaud app and create or sign into your account. Follow the app’s device-binding instructions and apply any offered firmware update. Confirm the account’s available transcription allowance before starting a long session. ## Record a meeting Place the recorder where it can hear the group and start recording with the physical control. Confirm the display status. Stop after a short sample, sync it and listen for quiet speakers or handling noise. Adjust placement before beginning the full meeting. ## Record a phone call Attach the recorder to the phone as instructed and verify call mode. Tell the other caller you intend to record. Test your actual phone and case combination; changing the attachment or switching to a headset can change capture conditions. ## Turn it off and charge it Stop the active recording and confirm it has saved before handling power controls. Use the supplied charging arrangement and follow the current manual for shutdown. Ending a recording, going into standby and powering down are different states; the display helps identify which state the device is in. ## Export and share Sync the audio, choose the transcription language and generate a summary suited to the session. Review names, decisions and dates against the audio. Export through the app’s available controls and check that the recipient has the intended file or access permission. ## Frequently asked questions ### How do I turn off the Plaud Note Pro? Finish recording first, then follow the current device manual’s power procedure. Do not treat ending a recording as proof that the recorder is fully powered down. ### How does Plaud work? The recorder captures audio; the app transfers it to the service for transcription, summaries and retrieval. ## Related reading - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [Does Plaud need a subscription? Plans and free minutes explained](https://testbenchlab.com/guides/does-plaud-need-a-subscription/) ## Sources 1. [Plaud Note Pro specifications and plans](https://www.plaud.ai/products/plaud-note-pro) --- # Is Plaud HIPAA compliant? Explainer · AI recorders · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Plaud states that it supports HIPAA compliance and now provides instructions for requesting a signed Business Associate Agreement. That does not make every consumer account or clinical recording workflow automatically compliant: the contract, configuration and organisation’s controls still matter. ## Is Plaud HIPAA compliant? Plaud’s current support guidance directs customers to privacy@plaud.ai to obtain a signed BAA. This is important because older articles say no BAA is available. Confirm the agreement covers the exact service and account being used before sending protected health information to it. ## What that means for clinicians A covered organisation needs appropriate agreements and safeguards, not merely a compliance badge. HHS guidance requires risk analysis and management for cloud processing of electronic protected health information. Review access, retention, deletion, subprocessors and device handling with your organisation’s compliance team. A transcript or summary can retain sensitive identifiers from the original audio. ## Alternatives for medical use Choose a service approved for your clinical workflow with the required contractual and operational controls. Our general preference for Notta is a consumer editorial recommendation, not a HIPAA certification or clinical approval. Local processing can reduce transfers, but it still requires suitable security and governance. ## Frequently asked questions ### Can doctors use Plaud? Potentially, in an approved workflow with the appropriate BAA and safeguards. Do not upload patient recordings through an ordinary account on the strength of a marketing claim alone. ## Related reading - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) ## Sources 1. [Plaud guidance on patient identifiers and signed BAAs](https://support.plaud.ai/hc/en-us/articles/53443476408985-Does-Plaud-automatically-mask-remove-patient-identifiable-information-before-using-AI) 2. [HHS guidance on cloud computing and HIPAA](https://www.hhs.gov/hipaa/for-professionals/special-topics/health-information-technology/cloud-computing/index.html) --- # llama.cpp with Vulkan on a Radeon iGPU How-to · Local AI · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Vulkan is a useful llama.cpp route for Radeon integrated graphics. ROCm is another option on supported combinations. Choose a working backend, then compare speed with the same model and settings. The command expects the downloaded model file named `model.gguf`. This documentation-based procedure is not presented as a hardware validation run. ## Vulkan or ROCm? Support differs by operating system and build. Vulkan provides a broad graphics-compute route; ROCm requires a supported driver and hardware combination. A working Linux setup does not guarantee the same installation route on Windows. ## Build or download llama.cpp with Vulkan Install Git, Visual Studio C++ build tools, CMake and the Vulkan SDK. In a developer PowerShell, build the official repository: ```powershell git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build -DGGML_VULKAN=ON cmake --build build --config Release ``` An official Windows Vulkan release package is an alternative; keep its DLLs beside the executables. ## Run your first model Use a small publisher-supplied GGUF and save it as `model.gguf` in the project folder: ```powershell .\build\bin\Release\llama-cli.exe -m .\model.gguf -ngl 99 -c 2048 -n 128 -p "Explain unified memory in three sentences." ``` Check startup output for Vulkan detection and layer offload. Reduce offload or model size if allocation fails. Record the model file and build when comparing speed. ## Frequently asked questions ### Is Vulkan or ROCm faster on Strix Halo? There is no universal winner. Match model, quantisation, context, driver and power settings on supported builds. ## Related reading - [How to install Ollama on Windows 11 (Ryzen AI mini PC)](https://testbenchlab.com/guides/install-ollama-windows-11-ryzen-ai-mini-pc/) - [How much RAM do you need to run a local LLM?](https://testbenchlab.com/guides/how-much-ram-for-local-llm/) ## Sources 1. [llama.cpp build documentation](https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md) 2. [Ollama hardware support](https://docs.ollama.com/gpu) --- # Mini PC vs Mac mini for AI Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Choose a mini PC for a Windows or Linux workflow and a Mac mini for software built around macOS and Apple silicon. Our preferred Windows/Linux shortlist starts with Geekom, but compatibility should determine the platform. | Workflow | Starting point | |---|---| | Windows desktop and smaller models | Geekom A9 Max | | Larger AMD shared-memory workload | Geekom A9 Mega | | macOS/Metal/MLX requirement | Suitably configured Mac mini | ## Which runs local AI better? Neither category has one universal winner. Apple silicon commonly uses Metal and MLX; AMD systems use supported graphics or Ryzen AI backends. Match model and context. A result for one Mac mini generation does not describe every current model. ## Memory and price Compare capacity at purchase, especially for onboard memory. The M4 models in older comparisons are a specific reference, not the whole current range. Budget for model weights and context rather than comparing entry prices alone. ## Software support Choose the environment supporting your task with least friction. A9 Max is our preferred general desktop; A9 Mega is our choice for larger shared-memory models. Mac mini is more relevant where the required application explicitly targets Apple silicon. ## Related reading - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) ## Sources 1. [Apple Mac mini specifications, 2024 models](https://support.apple.com/en-gb/121555) 2. [Ollama hardware support](https://docs.ollama.com/gpu) 3. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 4. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) --- # NPU vs GPU: which runs AI better on a mini PC? Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. For general local-model experimentation, start with a supported GPU backend. An NPU is useful when the software and model are explicitly prepared for it. The two processors use different execution paths rather than competing on one universal speed scale. | Buying check | GPU | NPU | |---|---|---| | Software route | Graphics/compute backend | Neural runtime and supported model | | Constraints | Driver, memory and offload | Driver, model catalogue and backend | | AI badge proves use? | No | No | ## What is the difference between an NPU and a GPU? A GPU is a broadly programmable parallel processor used for graphics and compute. An NPU specialises in supported neural operations. Their usefulness depends on the software exposing them, not simply on both being labelled AI hardware. ## Which is better for local LLMs? GPU backends provide a practical starting point for a broad model workflow. NPU paths can be useful within their supported catalogue. A GGUF file suitable for a GPU runner is not automatically an NPU model; check formats before buying. ## Which is better for background AI features? An NPU can suit supported background features where efficiency matters. That does not imply it wins a large-model generation task. Check whether your feature runs locally and whether it uses this processor or a remote service. ## Radeon 890M vs XDNA 2 NPU on Ryzen AI These are separate accelerators. A Radeon workload uses a graphics-compute backend; an NPU workload uses a compatible neural runtime. Watch the application’s diagnostics to identify the active path rather than relying on a total-AI number. ## Frequently asked questions ### Is an NPU faster than a GPU? Not universally. A matched model and runtime are needed to compare useful speed. ### Can an NPU replace a GPU? It can perform supported AI tasks but does not replace the GPU’s broader graphics and compute roles. ## Related reading - [What is an NPU, and does your PC need one?](https://testbenchlab.com/guides/what-is-an-npu/) - [Radeon 890M vs Radeon 8060S: what the iGPU means for AI and gaming](https://testbenchlab.com/guides/radeon-890m-vs-radeon-8060s/) - [Run local LLMs on the Ryzen AI NPU with Lemonade Server](https://testbenchlab.com/guides/run-llms-on-ryzen-ai-npu-lemonade-server/) ## Sources 1. [Ollama hardware support](https://docs.ollama.com/gpu) 2. [AMD Lemonade getting-started guide](https://developer.amd.com/playbooks/lemonade-getting-started/) 3. [AMD Ryzen AI 400 announcement and specifications](https://ir.amd.com/news-events/press-releases/detail/1270/amd-expands-ai-leadership-across-client-graphics-and-software-with-new-ryzen-ryzen-ai-and-amd-rocm-announcements-at-ces-2026) --- # NVIDIA DGX Spark vs Strix Halo mini PCs Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Choose DGX Spark for a supported CUDA-focused development workflow. For a general x86 Windows or Linux Strix Halo workstation, our preference is Geekom A9 Mega. This is a platform comparison rather than a claimed head-to-head benchmark. | Requirement | Preferred starting point | |---|---| | General x86 Windows/Linux workstation | Geekom A9 Mega | | Defined CUDA-focused development | NVIDIA DGX Spark | | Model-fit comparison | Match allocation and context | ## CUDA vs ROCm DGX Spark targets NVIDIA’s ecosystem, while Strix Halo uses AMD and cross-vendor backends. Dependencies can settle the purchase before specifications do. DGX Spark’s CPU architecture also differs from an x86 desktop; check the whole software stack. ## Memory and bandwidth NVIDIA lists 128GB unified memory and 273GB/s bandwidth. Compare the exact Strix Halo capacity and allocation. Equal capacity does not imply equal speed, and advertised compute at a particular precision is not a dense-model generation result. ## Price Compare delivered configurations, storage, tax and support in your country. Launch prices and regional promotions make unqualified dollar comparisons misleading. Keep the intended workload fixed when deciding whether a premium is useful. ## Which should you buy? Our default is A9 Mega for the general workstation role. Choose DGX Spark for a defined NVIDIA dependency. A personal preference should not override software compatibility that your project genuinely needs. ## Related reading - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) ## Sources 1. [NVIDIA DGX Spark hardware guide](https://docs.nvidia.com/dgx/dgx-spark/hardware.html) 2. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) 3. [Ollama hardware support](https://docs.ollama.com/gpu) --- # One-party vs two-party consent states: can you use an AI recorder? Explainer · AI recorders · 3 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. One-party consent generally means one participant can consent to recording; all-party consent generally requires every participant’s permission. US state rules vary by type of communication and circumstances, so a simple two-column map cannot settle every recording decision. | State or issue | Practical distinction to check | |---|---| | [California](https://www.rcfp.org/reporters-recording-guide/california/) | All-party requirements for confidential communications. | | [Florida](https://www.rcfp.org/reporters-recording-guide/florida/) | All-party requirements for protected conversations and calls. | | [Illinois](https://www.rcfp.org/reporters-recording-guide/illinois/) | All-party requirements for private conversations and protected communications. | | [Maryland](https://www.rcfp.org/reporters-recording-guide/maryland/) | All-party requirements for private speech and calls. | | [Massachusetts](https://www.rcfp.org/reporters-recording-guide/massachusetts/) | The prohibition on secret recording is important; awareness affects the analysis. | | [Montana](https://www.rcfp.org/reporters-recording-guide/montana/) | Hidden recording and the knowledge of participants are important. | | [New Hampshire](https://www.rcfp.org/reporters-recording-guide/new-hampshire/) | All-party requirements, with distinctions affecting penalties. | | [Pennsylvania](https://www.rcfp.org/reporters-recording-guide/pennsylvania/) | All-party requirements for protected conversations and communications. | | [Washington](https://www.rcfp.org/reporters-recording-guide/washington/) | Private-conversation rules and statutory exceptions need checking. | | [Connecticut](https://www.rcfp.org/reporters-recording-guide/connecticut/) | Phone-call civil liability differs from the criminal one-party rule and in-person recording. | | [Nevada](https://www.rcfp.org/reporters-recording-guide/nevada/) | Phone calls and in-person conversations have different consent rules. | | [Oregon](https://www.rcfp.org/reporters-recording-guide/oregon/) | Distinguish in-person notice rules, electronic communications and applicable exceptions. | | [Michigan](https://www.rcfp.org/reporters-recording-guide/michigan/) | Participant recording and outsider eavesdropping require a careful current-law check. | | [Delaware](https://www.rcfp.org/reporters-recording-guide/delaware/) | Different recording and privacy provisions require a careful current-law check. | This is a guide to states and distinctions requiring particular attention, not an exhaustive classification of every possible conversation. Consult the linked state chapter and current statutes for the circumstances. Source chapters have their own update dates; checking a page today does not make it a new legal opinion. ## What is one-party consent? One-party consent does not mean that any bystander can record any conversation. The consent normally comes from a participant, and other restrictions may apply. Federal and state rules must be considered together. The purpose and use of a recording can also affect the legal analysis. ## Which states require all-party consent? California and Florida are examples of states with all-party requirements for protected conversations. Other states have distinctions between calls, in-person speech, secrecy and reasonable privacy. Use the linked state guide rather than relying on an unqualified list of “two-party states”; two-party is shorthand even when more than two people are present. ## Does this apply to AI voice recorders? Yes. A dedicated recorder is still recording a conversation. Processing the audio into a transcript, uploading it to a cloud service and sharing it are additional actions to consider. Ask permission openly and explain the intended use, especially in a workplace or confidential meeting. ## Recording calls across state lines A cross-state call can engage more than one jurisdiction. Do not assume your own state’s rule is the only relevant one. Obtain everyone’s agreement and seek advice for a disputed, sensitive or consequential recording. This guide provides general context, not a determination for a particular call. ## Frequently asked questions ### Are AI voice recorders legal? They can be used lawfully, but consent and context determine whether a particular recording is permitted. The AI feature does not create an exception. ### Is it legal to record a conversation at work? It depends on applicable law, the participants and workplace rules. Permission to attend a meeting does not automatically include permission to record and upload it. ## Related reading - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) - [How to record a phone call on iPhone (built-in, apps and AI recorders)](https://testbenchlab.com/guides/how-to-record-a-phone-call-on-iphone/) - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/) ## Sources 1. [Reporters Committee state recording-law guide](https://www.rcfp.org/reporters-recording-guide/) --- # Radeon 890M vs Radeon 8060S: what the iGPU means for AI and gaming Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Radeon 8060S belongs to the larger Strix Halo graphics tier, while 890M belongs to the smaller Ryzen AI 9 tier. For local models, compare memory capacity and bandwidth alongside graphics resources. A compute-unit ratio does not directly predict inference speed. | Specification | Radeon 890M | Radeon 8060S | |---|---|---| | Architecture | RDNA 3.5 | RDNA 3.5 | | Compute units | 16 | 40 | | Relevant tier | Ryzen AI 9 HX | Ryzen AI Max+ | | Preferred Geekom | A9 Max | A9 Mega | ## Radeon 890M vs 8060S specs Both use RDNA 3.5 graphics, but the designs differ substantially in scale. Compute units describe resources, not usable model memory or backend efficiency. Check the processor generation because clocks and supported memory configurations vary. ## Which mini PCs use each? Geekom A9 Max is our preferred 890M desktop, and A9 Mega is our preferred 8060S workstation. Minisforum, GMKtec and Beelink also offer relevant options. Match the exact CPU suffix before assuming the graphics configuration. ## Local AI performance 8060S systems are interesting for the memory platform surrounding the GPU. Larger quantised models may fit where an HX desktop configuration cannot. Compare prompt processing and generation separately, using the same model and backend. ## Gaming performance Resolution, settings, power and memory all affect the experience. Identify frame generation and upscaling when quoting results. There is no single discrete-GPU equivalent across every game and AI workload. ## Frequently asked questions ### Is the Radeon 8060S good for gaming? It is a substantial integrated GPU. Settings, power and the complete configuration determine performance. ### What is the Radeon 890M equivalent to? No single equivalent covers every game and compute task. Use matched tests for your application. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) - [Ryzen AI Max+ 395 (Strix Halo) explained](https://testbenchlab.com/guides/ryzen-ai-max-395-strix-halo-explained/) ## Sources 1. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) 2. [AMD Ryzen AI 400 announcement and specifications](https://ir.amd.com/news-events/press-releases/detail/1270/amd-expands-ai-leadership-across-client-graphics-and-software-with-new-ryzen-ryzen-ai-and-amd-rocm-announcements-at-ces-2026) 3. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 4. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) --- # Run local LLMs on the Ryzen AI NPU with Lemonade Server How-to · Local AI · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Lemonade provides local model tools for several backends, including supported Ryzen AI routes. Selecting Lemonade does not automatically mean the NPU is active: the model and backend determine execution. **Requirements:** compatible Windows hardware, Ryzen AI drivers and a model supported by the intended backend. This is a documentation-based procedure, not a hardware validation claim. ## What is Lemonade Server? Lemonade organises model execution behind an application and service interface. A model using llama.cpp is different from one prepared for a Ryzen AI NPU backend, even when both appear in the same application. ## Install Lemonade Server Download and run the current official Windows MSI installer. Update Ryzen AI drivers as directed by the compatible AMD guide. Open the application and check the detected hardware and backends. Use the current installer rather than an executable name from an older tutorial. ## Run a model on the NPU Select a catalogue model explicitly supported by the detected Ryzen AI backend, download it and start a chat. Inspect backend status and logs. A response proves that a model ran, not that it used the NPU. ## NPU vs iGPU in Lemonade Choose the route according to model support. The current command interface can be checked with: ```powershell lemonade --help lemonade status lemonade backends ``` A hybrid path can use multiple processors and should be labelled accordingly. Speed comparisons need matching model conditions. ## Frequently asked questions ### Does Lemonade Server use the NPU? It can with compatible hardware, drivers and a supported model/backend. Other models use GPU or CPU routes. ## Related reading - [What is an NPU, and does your PC need one?](https://testbenchlab.com/guides/what-is-an-npu/) - [NPU vs GPU: which runs AI better on a mini PC?](https://testbenchlab.com/guides/npu-vs-gpu/) - [How to install Ollama on Windows 11 (Ryzen AI mini PC)](https://testbenchlab.com/guides/install-ollama-windows-11-ryzen-ai-mini-pc/) ## Sources 1. [Lemonade installation documentation](https://lemonade-server.ai/docs/guide/install/) 2. [AMD Lemonade getting-started guide](https://developer.amd.com/playbooks/lemonade-getting-started/) 3. [Lemonade current command interface](https://lemonade-server.ai/docs/guide/cli/) --- # Ryzen AI 9 HX 470 vs HX 370: what changed? Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. HX 470 refreshes the HX 370 platform with higher headline clocks and NPU capability. It remains an HX-tier processor, rather than a move to Strix Halo. Choose a suitable memory configuration before paying a large premium for the newer suffix. | Manufacturer maximum | HX 370 | HX 470 | |---|---|---| | Cores / threads | 12 / 24 | 12 / 24 | | CPU boost | Up to 5.1GHz | Up to 5.2GHz | | Graphics | Radeon 890M | Radeon 890M | | NPU | Up to 50 TOPS | Up to 55 TOPS | ## HX 470 vs HX 370 specs Both are 12-core, 24-thread Ryzen AI 9 processors with Radeon 890M graphics. The newer chip raises the headline CPU and NPU specifications. The chassis still determines power, cooling and installed memory. ## Does the HX 470 run AI faster? A higher NPU figure matters only to a workload using that NPU. GPU inference also depends on memory and software. No universal percentage improvement is justified for every local model. ## Which mini PCs use the HX 470? Geekom A9 Max 2026 is our preferred option. Minisforum AI X1 Pro-470 and GMKtec EVO-X1 Pro are alternatives. Store URLs may persist across generations; inspect the variant selected in the cart. ## Should you buy the HX 470 version? Prefer it for a modest premium in an otherwise matched configuration. A discounted HX 370 with suitable RAM can be a better buy. For much larger models, a tier change matters more than this refresh. ## Frequently asked questions ### Is the HX 470 worth it over the HX 370? For a small premium with matching RAM and storage, it is our preference. Configuration and a substantial discount can outweigh the generation change. ## Related reading - [Geekom A9 Max (2026, Ryzen AI 9 HX 470) review](https://testbenchlab.com/reviews/geekom-a9-max-review/) - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) ## Sources 1. [AMD Ryzen AI 400 announcement and specifications](https://ir.amd.com/news-events/press-releases/detail/1270/amd-expands-ai-leadership-across-client-graphics-and-software-with-new-ryzen-ryzen-ai-and-amd-rocm-announcements-at-ces-2026) 2. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) 3. [Minisforum AI X1 Pro-470](https://store.minisforum.com/products/minisforum-ai-x1-pro-470-mini-pc) --- # Ryzen AI Max+ 395 (Strix Halo) explained Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Strix Halo is the common name for AMD’s Ryzen AI Max platform. Its appeal is a larger integrated GPU and high-capacity shared memory. The Max+ 395 is one option, but CPU and graphics resources vary across the range. | Processor | CPU cores / threads | GPU | GPU compute units | |---|---|---|---:| | Max+ 395 | 16 / 32 | Radeon 8060S | 40 | | Max+ 388 | 8 / 16 | Radeon 8060S | 40 | | Max 390 | 12 / 24 | Radeon 8050S | 32 | | Max 385 | 8 / 16 | Radeon 8050S | 32 | ## What is Strix Halo? The platform combines CPU cores, integrated Radeon graphics and an NPU around shared memory. It is relevant to compact workstations because large model weights can use a substantial pool without installing a separate high-VRAM graphics card. ## Ryzen AI Max+ 395 vs 390 vs 388 vs 385 Check the exact designation: Max+ 388 is different from Max 385. CPU and GPU resources do not increase in the same order. A lower-core model retaining stronger graphics may suit GPU-focused work differently from a higher-core model with fewer graphics units. ## Which mini PCs use it? Geekom A9 Mega is our preferred workstation. Minisforum MS-S1 Max, GMKtec EVO-X2 and Beelink GTR9 Pro are alternatives. Inspect the selected processor and installed capacity rather than assuming one chassis name guarantees a specification. ## What it is good (and bad) at The platform suits memory-heavy local inference and compact workstation use. It does not supply every CUDA-specific software route, and capacity does not guarantee rapid generation. Check the backend before treating a large RAM figure as a universal solution. ## Frequently asked questions ### What is Strix Halo equivalent to? No single desktop CPU or graphics card is equivalent across all tasks. Compare the exact workload and power settings. ### Is Ryzen AI Max+ 395 good for gaming? Its Radeon 8060S makes it relevant for integrated-graphics gaming. Resolution, settings and shared power still matter. ## Related reading - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) - [Radeon 890M vs Radeon 8060S: what the iGPU means for AI and gaming](https://testbenchlab.com/guides/radeon-890m-vs-radeon-8060s/) - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) ## Sources 1. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) 2. [AMD Ryzen AI software hardware table](https://www.amd.com/en/products/software/ryzen-ai-software.html) 3. [Geekom A9 Mega specifications](https://www.geekompc.com/geekom-a9-mega-ai-mini-pc/) --- # What is an AI PC? (and is a Copilot+ PC different?) Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. AI PC is a broad description for a computer offering AI-related hardware or software. Copilot+ describes Microsoft’s published hardware-and-feature class. Neither term tells you which model fits or whether your application uses the NPU. ## What makes a PC an "AI PC"? The label can refer to an NPU, software or a wider platform. Ask which feature you need, whether it runs locally and which processor it uses. A cloud assistant can require little specialised local hardware. ## What is a Copilot+ PC? Microsoft’s published class calls for a 40+ TOPS NPU and other system requirements. Branding and feature rollout can evolve. The threshold alone does not establish that every related feature is available in your region. ## Does Windows Recall work on AMD mini PCs? Recall requires an eligible system and supported Windows configuration, with availability affected by rollout and policy. Installing Windows on a machine with an AMD chip does not by itself unlock the feature. Check the exact machine and current eligibility information. ## Do you need an AI PC? Buy for a supported local feature or workload that benefits. For ordinary cloud tools, a capable desktop can be sufficient. Geekom A9 Max is our preferred general mini PC, but memory and application support matter more than an AI sticker. ## Frequently asked questions ### How many TOPS do you need for Copilot+? Microsoft’s published requirement calls for a 40+ TOPS NPU alongside its other hardware and software requirements. ## Related reading - [What is an NPU, and does your PC need one?](https://testbenchlab.com/guides/what-is-an-npu/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) ## Sources 1. [Microsoft Copilot+ hardware requirements](https://www.microsoft.com/en-gb/windows/copilot-plus-pcs) 2. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) --- # What is an NPU, and does your PC need one? Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. An NPU is a specialised processor for supported AI operations. It can handle selected local features efficiently, but it does not automatically run every chatbot. Buy for the application you need rather than the TOPS figure alone. ## What does an NPU do? An NPU executes supported neural-network workloads through an appropriate runtime. Examples include selected audio, image and language features. The application decides which accelerator is used; a PC containing an NPU can still run a particular model entirely on its CPU or GPU. ## What do TOPS ratings mean? TOPS expresses a theoretical operation rate at a specified numerical precision. It is not a universal measure of useful output. Workloads use different operations, memory patterns and software. Do not convert a TOPS ratio into an expected tokens-per-second ratio. ## What software actually uses the NPU? Applications need explicit driver and runtime support. Windows AI features and compatible Ryzen AI tools provide examples. Check the current application documentation and active backend: installing a generic chatbot does not establish NPU use. ## Does an NPU help run local LLMs? It can with supported models and software, including compatible Ryzen AI backends. Many local-model workflows use GPU acceleration instead. Prioritise memory, model support and practical fit before the headline NPU specification. ## Frequently asked questions ### What is an NPU used for? Supported neural-network operations, including selected local audio, image and language features. ### How many TOPS do you need for Copilot+? Microsoft’s published hardware class calls for a 40+ TOPS NPU, alongside other system requirements. ### Is an NPU better than a GPU for AI? Neither is universally better. The supported model and application determine the useful accelerator. ## Related reading - [NPU vs GPU: which runs AI better on a mini PC?](https://testbenchlab.com/guides/npu-vs-gpu/) - [Ryzen AI 9 HX 470 vs HX 370: what changed?](https://testbenchlab.com/guides/ryzen-ai-9-hx-470-vs-hx-370/) - [Run local LLMs on the Ryzen AI NPU with Lemonade Server](https://testbenchlab.com/guides/run-llms-on-ryzen-ai-npu-lemonade-server/) - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/) ## Sources 1. [AMD Ryzen AI 400 announcement and specifications](https://ir.amd.com/news-events/press-releases/detail/1270/amd-expands-ai-leadership-across-client-graphics-and-software-with-new-ryzen-ryzen-ai-and-amd-rocm-announcements-at-ces-2026) 2. [AMD Lemonade getting-started guide](https://developer.amd.com/playbooks/lemonade-getting-started/) 3. [Microsoft Copilot+ hardware requirements](https://www.microsoft.com/en-gb/windows/copilot-plus-pcs) --- # What is OCuLink? OCuLink vs USB4 for an external GPU Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. OCuLink carries PCIe outside a system and is often used for an external GPU dock. USB4 is a broader peripheral connection that can tunnel PCIe. Choose GPU memory and compatible software before choosing the cable standard. | Buying check | OCuLink | USB4 | |---|---|---| | Main use | Direct external PCIe | General peripherals and compatible PCIe docks | | Verify | Lanes, dock and power procedure | PCIe tunnelling and controller support | | Other peripherals | Separate connections usually needed | Broader dock ecosystem | | Connecting an eGPU | Power off unless explicitly supported otherwise | Follow the supported dock procedure | ## What is OCuLink? A common mini PC implementation exposes PCIe 4.0 x4 externally. This is focused expansion, rather than a general USB port. Confirm generation, lane count, cable and dock because the connector alone does not describe the completed system. ## OCuLink vs USB4 / Thunderbolt for eGPUs A direct link can avoid some tunnelling overhead, while USB4 offers broader peripheral use. Real behaviour depends on the controller, lanes and application. Neither theoretical link rate nor a gaming result establishes a universal inference advantage. ## Which mini PCs have OCuLink? Minisforum AI X1 Pro and GMKtec EVO-X1 are relevant options. Geekom A9 Max is our preferred general desktop, with USB4 rather than an advertised native OCuLink connector. A required connector is a valid reason to choose an alternative. ## Is an eGPU worth it for local AI? It can add memory and supported acceleration. Include the card, dock and power supply in the total. A faster cable cannot make a model fit insufficient VRAM; also compare a high-memory Strix Halo computer. ## Frequently asked questions ### Is OCuLink faster than Thunderbolt? A direct PCIe implementation can offer more useful eGPU bandwidth than some tunneled links. Compare actual generations and controllers. ### Is OCuLink hot-pluggable? Treat common mini PC eGPU setups as power-off connections unless both manufacturers explicitly support another procedure. ## Related reading - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) - [Geekom A9 Max vs Minisforum AI X1 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-minisforum-ai-x1-pro/) - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) ## Sources 1. [Minisforum AI X1 Pro-470](https://store.minisforum.com/products/minisforum-ai-x1-pro-470-mini-pc) 2. [GMKtec EVO-X1](https://www.gmktec.com/products/amd-ryzen%E2%84%A2-ai-9-hx-370-evo-x1-ai-mini-pc) 3. [Geekom A9 Max specifications](https://www.geekompc.com/geekom-a9-max-mini-pc/) --- # What is unified memory, and why does it matter for local AI? Explainer · AI PC hardware · 2 min read. By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-10-02. Unified memory is a shared physical pool that multiple processors can access. For local AI, its attraction is capacity: an integrated GPU can potentially work with more model data than a small graphics card holds. Allocation and software still set limits. ## What is unified memory? The CPU and integrated accelerator share a memory system rather than relying on entirely separate physical RAM banks. Details vary by platform. The operating system, applications and model runtime still compete for capacity and bandwidth. ## Unified memory vs VRAM Dedicated VRAM belongs to a graphics card; unified memory also serves ordinary system work. Equal capacity is not equivalent performance. A large shared pool is valuable for fit, while a smaller fast card can excel on a workload that fits its memory. ## How much of the memory can the GPU use on Strix Halo? The GPU-visible amount depends on firmware, capacity and drivers. A 128GB listing does not supply 128GB exclusively for model weights. Inspect the runtime’s allocation and leave room for the operating system, context and temporary buffers. ## Unified memory on Mac vs AMD Mac workflows commonly use Metal or MLX; AMD systems use supported Vulkan, ROCm or other backends. Choose the software environment first. A similar memory number does not make applications and model formats interchangeable. ## Frequently asked questions ### Is unified memory the same as VRAM? No. It can serve graphics workloads but remains a shared system pool rather than separate card memory. ### What does "96GB VRAM" mean on a mini PC? It usually describes an allocation from shared memory. Confirm the platform and settings rather than assuming separate physical graphics memory. ## Related reading - [Geekom A9 Mega (Ryzen AI Max+ 395) review](https://testbenchlab.com/reviews/geekom-a9-mega-review/) - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) - [The best Strix Halo mini PCs (Ryzen AI Max+ 395)](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) - [How much RAM do you need to run a local LLM?](https://testbenchlab.com/guides/how-much-ram-for-local-llm/) ## Sources 1. [AMD Ryzen AI Max specifications](https://ir.amd.com/news-events/press-releases/detail/1232/amd-announces-expanded-consumer-and-commercial-ai-pc-portfolio-at-ces) 2. [Ollama hardware support](https://docs.ollama.com/gpu) 3. [Apple Mac mini specifications, 2024 models](https://support.apple.com/en-gb/121555) --- # David Wilson Writer and editor, Testbench Lab. David Wilson writes and edits Testbench Lab’s coverage of AI voice recorders, mini PCs and local AI. His focus is the practical difference between a feature on a product page and a tool you would want to use regularly. ## What David covers David’s work covers recorder workflows, transcription plans, mini PC configurations and local language-model setup. The guides explain the decisions behind the steps, including how much memory a model needs and when a paid transcription plan becomes worthwhile. ## An approach built around everyday use David prefers Notta Memo after using it and Plaud Note Pro. Geekom is his preferred mini PC brand for this project. His articles explain those choices while acknowledging where another product has a relevant feature or a better fit for a specialist task. ## How to read David’s recommendations A personal verdict is identified as a judgement. Technical claims are supported by manufacturer documentation or attributed external evidence. His byline identifies the author responsible for the article and its recommendations. ## Published work - [Anker Soundcore Work review](https://testbenchlab.com/reviews/soundcore-work-review/) — Review - [Beelink GTR9 Pro review](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/) — Review - [Beelink SER9 Pro review](https://testbenchlab.com/reviews/beelink-ser9-pro-review/) — Review - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/) — Buying guide - [Does Plaud need a subscription? Plans and free minutes explained](https://testbenchlab.com/guides/does-plaud-need-a-subscription/) — Guide - [Does Pocket AI need a subscription? What is free and what is not](https://testbenchlab.com/guides/does-pocket-ai-need-a-subscription/) — Guide - [Geekom A9 Max review: HX 470 and HX 370](https://testbenchlab.com/reviews/geekom-a9-max-review/) — Review - [Geekom A9 Max vs Beelink SER9 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-beelink-ser9-pro/) — Comparison - [Geekom A9 Max vs GMKtec EVO-X1](https://testbenchlab.com/compare/geekom-a9-max-vs-gmktec-evo-x1/) — Comparison - [Geekom A9 Max vs Minisforum AI X1 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-minisforum-ai-x1-pro/) — Comparison - [Geekom A9 Mega review: Strix Halo for local AI](https://testbenchlab.com/reviews/geekom-a9-mega-review/) — Review - [Geekom A9 Mega vs Beelink GTR9 Pro](https://testbenchlab.com/compare/geekom-a9-mega-vs-beelink-gtr9-pro/) — Comparison - [Geekom A9 Mega vs GMKtec EVO-X2](https://testbenchlab.com/compare/geekom-a9-mega-vs-gmktec-evo-x2/) — Comparison - [Geekom A9 Mega vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/geekom-a9-mega-vs-minisforum-ms-s1-max/) — Comparison - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/) — Comparison - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/) — Comparison - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/) — Comparison - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/) — Review - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/) — Review - [GMKtec EVO-X2 vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/gmktec-evo-x2-vs-minisforum-ms-s1-max/) — Comparison - [HiDock P1 review](https://testbenchlab.com/reviews/hidock-p1-review/) — Review - [How much RAM do you need to run a local LLM?](https://testbenchlab.com/guides/how-much-ram-for-local-llm/) — Guide - [How to install Ollama on Windows 11 (Ryzen AI mini PC)](https://testbenchlab.com/guides/install-ollama-windows-11-ryzen-ai-mini-pc/) — Guide - [How to record a Microsoft Teams meeting](https://testbenchlab.com/guides/how-to-record-a-teams-meeting/) — Guide - [How to record a phone call on iPhone (built-in, apps and AI recorders)](https://testbenchlab.com/guides/how-to-record-a-phone-call-on-iphone/) — Guide - [How to record a Zoom meeting (and get an AI transcript)](https://testbenchlab.com/guides/how-to-record-a-zoom-meeting/) — Guide - [How to use the Plaud Note Pro: setup to first summary](https://testbenchlab.com/guides/how-to-use-plaud-note-pro/) — Guide - [Is Plaud HIPAA compliant?](https://testbenchlab.com/guides/is-plaud-hipaa-compliant/) — Guide - [llama.cpp with Vulkan on a Radeon iGPU](https://testbenchlab.com/guides/llama-cpp-vulkan-radeon-igpu/) — Guide - [Mini PC vs Mac mini for AI](https://testbenchlab.com/guides/mini-pc-vs-mac-mini-for-ai/) — Guide - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/) — Review - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/) — Review - [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/) — Review - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/) — Comparison - [NPU vs GPU: which runs AI better on a mini PC?](https://testbenchlab.com/guides/npu-vs-gpu/) — Guide - [NVIDIA DGX Spark vs Strix Halo mini PCs](https://testbenchlab.com/guides/dgx-spark-vs-strix-halo/) — Guide - [One-party vs two-party consent states: can you use an AI recorder?](https://testbenchlab.com/guides/recording-consent-laws-by-state/) — Guide - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/) — Review - [Plaud Note Pro vs NotePin S: card or wearable?](https://testbenchlab.com/compare/plaud-note-pro-vs-notepin-s/) — Comparison - [Plaud Note review](https://testbenchlab.com/reviews/plaud-note-review/) — Review - [Plaud Note vs Plaud Note Pro](https://testbenchlab.com/compare/plaud-note-vs-plaud-note-pro/) — Comparison - [Plaud NotePin S review](https://testbenchlab.com/reviews/plaud-notepin-s-review/) — Review - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/) — Comparison - [Pocket AI recorder review](https://testbenchlab.com/reviews/pocket-ai-review/) — Review - [Radeon 890M vs Radeon 8060S: what the iGPU means for AI and gaming](https://testbenchlab.com/guides/radeon-890m-vs-radeon-8060s/) — Guide - [Run local LLMs on the Ryzen AI NPU with Lemonade Server](https://testbenchlab.com/guides/run-llms-on-ryzen-ai-npu-lemonade-server/) — Guide - [Ryzen AI 9 HX 470 vs HX 370: what changed?](https://testbenchlab.com/guides/ryzen-ai-9-hx-470-vs-hx-370/) — Guide - [Ryzen AI Max+ 395 (Strix Halo) explained](https://testbenchlab.com/guides/ryzen-ai-max-395-strix-halo-explained/) — Guide - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/) — Buying guide - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/) — Buying guide - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/) — Buying guide - [The best Strix Halo mini PCs](https://testbenchlab.com/best/best-strix-halo-mini-pcs/) — Buying guide - [What is an AI PC? (and is a Copilot+ PC different?)](https://testbenchlab.com/guides/what-is-an-ai-pc/) — Guide - [What is an NPU, and does your PC need one?](https://testbenchlab.com/guides/what-is-an-npu/) — Guide - [What is OCuLink? OCuLink vs USB4 for an external GPU](https://testbenchlab.com/guides/what-is-oculink/) — Guide - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) — Guide --- # Reviews Every AI voice recorder and mini PC we cover. Each review states whether it is hands-on or researched, and links its sources. - [Anker Soundcore Work review](https://testbenchlab.com/reviews/soundcore-work-review/). Soundcore Work is a compact wearable alternative for people who want physical capture with minimal bulk. - [Beelink GTR9 Pro review](https://testbenchlab.com/reviews/beelink-gtr9-pro-review/). Beelink GTR9 Pro is a premium Strix Halo alternative with substantial memory and networking. - [Beelink SER9 Pro review](https://testbenchlab.com/reviews/beelink-ser9-pro-review/). The Beelink SER9 Pro with Ryzen AI 9 HX 370 is a compact alternative for smaller local models and ordinary desktop work. - [Geekom A9 Max review: HX 470 and HX 370](https://testbenchlab.com/reviews/geekom-a9-max-review/). Geekom A9 Max is our preferred general-purpose AI mini PC. The HX 470 version offers a capable Windows desktop with upgradeable memory, but we would… - [Geekom A9 Mega review: Strix Halo for local AI](https://testbenchlab.com/reviews/geekom-a9-mega-review/). Geekom A9 Mega is our preferred Geekom for larger local AI workloads. Its Strix Halo platform is a different proposition from A9 Max: a larger shared-memory… - [GMKtec EVO-X1 review](https://testbenchlab.com/reviews/gmktec-evo-x1-review/). GMKtec EVO-X1 is an HX-tier mini PC worth considering for OCuLink expansion. Geekom A9 Max is our preferred general desktop; the EVO-X1 makes the most sense… - [GMKtec EVO-X2 review](https://testbenchlab.com/reviews/gmktec-evo-x2-review/). GMKtec EVO-X2 is a relevant Strix Halo alternative for large local models. Geekom A9 Mega remains our preference, but EVO-X2 should be on the shortlist when… - [HiDock P1 review](https://testbenchlab.com/reviews/hidock-p1-review/). HiDock P1 is worth considering if earphone-call capture is central to your workflow. Notta Memo remains our preferred general recorder, but HiDock’s… - [Minisforum AI X1 Pro review](https://testbenchlab.com/reviews/minisforum-ai-x1-pro-review/). Minisforum AI X1 Pro is an expansion-oriented alternative to Geekom A9 Max. We prefer Geekom for a general desktop, but the AI X1 Pro is particularly… - [Minisforum MS-S1 Max review](https://testbenchlab.com/reviews/minisforum-ms-s1-max-review/). Minisforum MS-S1 Max is the networking-and-expansion alternative in this Strix Halo shortlist. - [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/). Notta Memo is our preferred AI voice recorder. David Wilson found it much better than Plaud Note Pro in everyday use, and the included Hardware Plan gives… - [Plaud Note Pro review](https://testbenchlab.com/reviews/plaud-note-pro-review/). Plaud Note Pro is a capable card-style recorder, but Notta Memo is our preferred purchase. - [Plaud Note review](https://testbenchlab.com/reviews/plaud-note-review/). The original Plaud Note still makes sense when it is meaningfully cheaper than the Pro. For a new purchase at a small price difference, we favour Plaud Note… - [Plaud NotePin S review](https://testbenchlab.com/reviews/plaud-notepin-s-review/). Plaud NotePin S is our wearable choice within Plaud’s range. Its physical button and highlighting control suit people who want recording close at hand… - [Pocket AI recorder review](https://testbenchlab.com/reviews/pocket-ai-review/). Pocket is an alternative for people attracted to a conversational memory workflow and unlimited core processing. --- # Comparisons Two products or brands side by side, with a clear recommendation and the reasoning behind it. - [Geekom A9 Max vs Beelink SER9 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-beelink-ser9-pro/). Choose Geekom A9 Max over the HX 370 Beelink SER9 Pro as our preferred general AI desktop. - [Geekom A9 Max vs GMKtec EVO-X1](https://testbenchlab.com/compare/geekom-a9-max-vs-gmktec-evo-x1/). Geekom A9 Max is our preferred general desktop over GMKtec EVO-X1. The EVO-X1 deserves a place on the shortlist when a native OCuLink connection or a… - [Geekom A9 Max vs Minisforum AI X1 Pro](https://testbenchlab.com/compare/geekom-a9-max-vs-minisforum-ai-x1-pro/). Geekom A9 Max is our preferred purchase over Minisforum AI X1 Pro for an everyday AI desktop. - [Geekom A9 Mega vs Beelink GTR9 Pro](https://testbenchlab.com/compare/geekom-a9-mega-vs-beelink-gtr9-pro/). Geekom A9 Mega is our preferred purchase over Beelink GTR9 Pro for a general Strix Halo workstation. - [Geekom A9 Mega vs GMKtec EVO-X2](https://testbenchlab.com/compare/geekom-a9-mega-vs-gmktec-evo-x2/). Geekom A9 Mega is our preferred Strix Halo workstation over GMKtec EVO-X2. That recommendation is an editorial choice; the shared Max+ 395 platform does not… - [Geekom A9 Mega vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/geekom-a9-mega-vs-minisforum-ms-s1-max/). Geekom A9 Mega is our preference for a single-user local AI workstation. Choose Minisforum MS-S1 Max instead when 10GbE or internal expansion is a… - [Geekom vs Beelink: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-beelink/). Geekom is our preference over Beelink for this AI mini PC shortlist. We favour A9 Max for a general desktop and A9 Mega for larger models, while Beelink… - [Geekom vs GMKtec: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-gmktec/). Geekom is our preferred brand over GMKtec for a general AI desktop or local-model workstation. - [Geekom vs Minisforum: which AI mini PC brand?](https://testbenchlab.com/compare/geekom-vs-minisforum/). Geekom is our preferred brand for a general AI mini PC purchase. Choose A9 Max for an everyday desktop or A9 Mega for larger local models; Minisforum… - [GMKtec EVO-X2 vs Minisforum MS-S1 Max](https://testbenchlab.com/compare/gmktec-evo-x2-vs-minisforum-ms-s1-max/). GMKtec EVO-X2 is our pick over Minisforum MS-S1 Max for straightforward single-machine inference when matched capacity costs less. - [Notta Memo vs Plaud Note Pro](https://testbenchlab.com/compare/notta-memo-vs-plaud-note-pro/). Notta Memo ($149) beats Plaud Note Pro ($189) in our hands-on verdict. Compare free transcription, annual costs, recording features and battery life. - [Plaud Note Pro vs NotePin S: card or wearable?](https://testbenchlab.com/compare/plaud-note-pro-vs-notepin-s/). Choose Plaud Note Pro for phone-based recording and NotePin S for wearing throughout your day. - [Plaud Note vs Plaud Note Pro](https://testbenchlab.com/compare/plaud-note-vs-plaud-note-pro/). Plaud Note Pro is our preferred new purchase within Plaud’s card range when the premium is modest. - [Plaud vs Pocket: which AI recorder should you buy?](https://testbenchlab.com/compare/plaud-vs-pocket/). We favour Plaud Note Pro over Pocket for a conventional meeting-and-call recorder. Pocket is more interesting if its advertised unlimited core processing is… --- # Best of Our picks in each category, with who each one suits and what to check before you buy. - [Best Plaud alternatives](https://testbenchlab.com/best/plaud-alternatives/). Notta Memo is our first-choice Plaud alternative. It is the recorder David Wilson preferred in hands-on use, and the alternatives below offer different… - [The best AI voice recorders of 2026](https://testbenchlab.com/best/best-ai-voice-recorders/). Notta Memo is our preferred AI voice recorder for 2026. David Wilson’s hands-on preference anchors the recommendation; the other picks below are researched… - [The best mini PCs for AI in 2026](https://testbenchlab.com/best/best-mini-pcs-for-ai/). Geekom leads our mini PC recommendations for 2026. Choose A9 Max for an everyday computer that also runs smaller local models, or A9 Mega when larger-model… - [The best Ryzen AI 9 HX 370 and HX 470 mini PCs](https://testbenchlab.com/best/best-ryzen-ai-9-hx-370-mini-pcs/). Geekom A9 Max is our preferred Ryzen AI 9 HX 370/470 mini PC. This tier works best as an everyday desktop with local-AI capability, rather than a substitute… - [The best Strix Halo mini PCs](https://testbenchlab.com/best/best-strix-halo-mini-pcs/). Geekom A9 Mega is our preferred Strix Halo mini PC. The right alternative depends on matched memory, expansion and network needs; the Max+ 395 chip shared… --- # Categories Browse every review, comparison, buying guide and setup guide by the kind of product it covers. - [AI mini PCs](https://testbenchlab.com/categories/ai-mini-pcs/). An AI mini PC can be an everyday desktop, a local assistant or a compact model server. Our preferred brand is Geekom: A9 Max is the general desktop choice… - [AI voice recorders](https://testbenchlab.com/categories/ai-voice-recorders/). An AI voice recorder captures audio with dedicated hardware and uses software to turn it into transcripts and summaries. --- # Authors Who writes and edits Testbench Lab, and the reviews, comparisons and guides each author is responsible for. - [David Wilson — writer and editor](https://testbenchlab.com/authors/david-wilson/). David Wilson writes and edits Testbench Lab’s coverage of AI voice recorders, mini PCs and local AI. --- # Guides: AI recorders, local AI and mini PCs Clear guides to recording conversations and running AI on your own computer. Start with the task you want to complete, then choose the hardware or software that supports it. ## AI recorders Learn how to record calls and online meetings, how free transcription allowances work and what to check before sharing a recording. Our setup guides keep hardware recording separate from the cloud services that turn speech into searchable notes. - [Does Plaud need a subscription? Plans and free minutes explained](https://testbenchlab.com/guides/does-plaud-need-a-subscription/) — explainer, 2 min - [Does Pocket AI need a subscription? What is free and what is not](https://testbenchlab.com/guides/does-pocket-ai-need-a-subscription/) — explainer, 2 min - [How to record a Microsoft Teams meeting](https://testbenchlab.com/guides/how-to-record-a-teams-meeting/) — how-to, 2 min - [How to record a phone call on iPhone (built-in, apps and AI recorders)](https://testbenchlab.com/guides/how-to-record-a-phone-call-on-iphone/) — how-to, 2 min - [How to record a Zoom meeting (and get an AI transcript)](https://testbenchlab.com/guides/how-to-record-a-zoom-meeting/) — how-to, 2 min - [How to use the Plaud Note Pro: setup to first summary](https://testbenchlab.com/guides/how-to-use-plaud-note-pro/) — how-to, 2 min - [Is Plaud HIPAA compliant?](https://testbenchlab.com/guides/is-plaud-hipaa-compliant/) — explainer, 2 min - [One-party vs two-party consent states: can you use an AI recorder?](https://testbenchlab.com/guides/recording-consent-laws-by-state/) — explainer, 3 min ## Local AI Install a model runner, choose a model that fits your memory and check which processor is actually doing the work. These guides cover Ollama, Lemonade and llama.cpp, with particular attention to AMD integrated graphics and Ryzen AI hardware. - [How much RAM do you need to run a local LLM?](https://testbenchlab.com/guides/how-much-ram-for-local-llm/) — explainer, 2 min - [How to install Ollama on Windows 11 (Ryzen AI mini PC)](https://testbenchlab.com/guides/install-ollama-windows-11-ryzen-ai-mini-pc/) — how-to, 2 min - [llama.cpp with Vulkan on a Radeon iGPU](https://testbenchlab.com/guides/llama-cpp-vulkan-radeon-igpu/) — how-to, 2 min - [Run local LLMs on the Ryzen AI NPU with Lemonade Server](https://testbenchlab.com/guides/run-llms-on-ryzen-ai-npu-lemonade-server/) — how-to, 2 min ## AI PC hardware Understand NPUs, unified memory, integrated graphics and external GPU connections. The goal is to translate specifications into decisions: how large a model you can load, whether a particular backend supports your chip and what expansion would cost. - [Mini PC vs Mac mini for AI](https://testbenchlab.com/guides/mini-pc-vs-mac-mini-for-ai/) — explainer, 2 min - [NPU vs GPU: which runs AI better on a mini PC?](https://testbenchlab.com/guides/npu-vs-gpu/) — explainer, 2 min - [NVIDIA DGX Spark vs Strix Halo mini PCs](https://testbenchlab.com/guides/dgx-spark-vs-strix-halo/) — explainer, 2 min - [Radeon 890M vs Radeon 8060S: what the iGPU means for AI and gaming](https://testbenchlab.com/guides/radeon-890m-vs-radeon-8060s/) — explainer, 2 min - [Ryzen AI 9 HX 470 vs HX 370: what changed?](https://testbenchlab.com/guides/ryzen-ai-9-hx-470-vs-hx-370/) — explainer, 2 min - [Ryzen AI Max+ 395 (Strix Halo) explained](https://testbenchlab.com/guides/ryzen-ai-max-395-strix-halo-explained/) — explainer, 2 min - [What is an AI PC? (and is a Copilot+ PC different?)](https://testbenchlab.com/guides/what-is-an-ai-pc/) — explainer, 2 min - [What is an NPU, and does your PC need one?](https://testbenchlab.com/guides/what-is-an-npu/) — explainer, 2 min - [What is OCuLink? OCuLink vs USB4 for an external GPU](https://testbenchlab.com/guides/what-is-oculink/) — explainer, 2 min - [What is unified memory, and why does it matter for local AI?](https://testbenchlab.com/guides/what-is-unified-memory/) — explainer, 2 min