BOSGAME M5 AI Mini PC (Ryzen AI Max+ 395)
$3,299.99 – $3,599.99Price range: $3,299.99 through $3,599.99
AMD Ryzen AI Max+ 395, up to 128GB LPDDR5X, 2TB SSD, Radeon 8060S — all-metal AI mini desktop.
The BOSGAME M5 packs AMD’s flagship Ryzen AI Max+ 395 “Strix Halo” processor into an all-metal mini desktop, aimed at developers and AI hobbyists who want serious local-LLM memory capacity and genuine desktop performance without workstation pricing or a workstation footprint. If your priority is the highest unified-memory-per-pound configuration in a compact chassis, with enough I/O to double as a daily-driver desktop, this is built for exactly that trade-off.
Key Features
- AMD Ryzen AI Max+ 395 Processor — 16 Zen 5 cores / 32 threads, paired with the integrated Radeon 8060S GPU (40 RDNA 3.5 compute units) for real gaming and compute-shader performance alongside AI acceleration.
- Up to 126 TOPS Combined AI Performance — the XDNA 2 NPU delivers 50 TOPS on its own, with the 126 TOPS platform figure reflecting combined CPU, GPU, and NPU throughput on real mixed AI workloads.
- 96GB or 128GB LPDDR5X-8000 Unified Memory — configure the memory pool to your model size; the 128GB configuration is enough to load and run open-weight LLMs well beyond what a typical discrete GPU’s VRAM allows.
- Dual M.2 PCIe Gen4 SSD Slots, 2TB Included — fast storage out of the box with a second slot free for expansion as your model and dataset library grows.
- Dual USB4 (40Gbps) Ports — one front, one rear, for high-speed external storage, docking, or DisplayPort Alt Mode output, alongside three additional USB 3.2 Gen2 Type-A ports and two USB 2.0 ports.
- Quad Display Support — HDMI 2.1, full-size DisplayPort with 8K@60Hz, and DisplayPort Alt Mode over the USB4 ports combine for up to four simultaneous displays.
- All-Metal Chassis with a 3-Way Performance Switch — a physical Silent / Balanced / Extreme switch lets you trade noise for sustained throughput on demand, with a bundled vertical stand for a PS5-style upright desk footprint.
- Wi-Fi 7 and Full-Size SD 4.0 Card Reader — modern wireless plus a genuinely useful card reader for creators moving footage or datasets off removable media.
Maximum Local Memory, Minimum Desk Footprint
The single number that matters most for local LLM work on the M5 is memory: 128GB of LPDDR5X-8000 shared coherently across CPU, GPU, and NPU is enough to load quantized versions of genuinely large open-weight models that simply will not fit on a consumer discrete GPU’s 16GB or 24GB of VRAM, all inside a chassis that fits on a desk rather than under one. For developers doing local inference, RAG pipeline testing, or agentic workflow development against models in the 30B–70B range, that memory ceiling is the difference between “runs on this machine” and “needs a cloud GPU rental.”
The physical performance switch is a genuinely practical feature for anyone who splits time between quiet, everyday use and demanding sustained AI workloads — Silent mode for background inference or general desktop work, Extreme mode when you need every bit of throughput a fine-tuning or batch-inference job can extract from the Ryzen AI Max+ 395’s 40 RDNA 3.5 compute units. Dual PCIe Gen4 M.2 slots mean the included 2TB drive doesn’t have to be your ceiling — add a second SSD as your model checkpoint library grows.
Being straightforward about where the M5 fits: its 2.5GbE networking, rather than 10GbE, is worth noting for anyone planning to move very large datasets or checkpoints across a local network regularly — it’s a genuine gap at this price and performance tier that reviewers have flagged. And as with every Strix Halo device, ROCm software support, while substantially improved and actively developed by AMD, still trails CUDA’s maturity for some of the newest inference frameworks — budget time for setup if you’re coming from an NVIDIA-first workflow. Within that honest framing, the M5’s combination of maximum memory capacity, genuine RDNA 3.5 graphics performance, and an all-metal build at a mini-PC footprint makes it one of the strongest value plays in the Strix Halo category for local AI development that doesn’t need enterprise support contracts.
Specifications
- Processor: AMD Ryzen AI Max+ 395, 16 cores / 32 threads (Zen 5)
- GPU: AMD Radeon 8060S, 40 RDNA 3.5 compute units
- NPU: AMD XDNA 2, up to 50 TOPS (up to 126 TOPS combined platform AI performance)
- Memory: 96GB or 128GB LPDDR5X-8000 unified memory
- Storage: Dual M.2 PCIe Gen4 NVMe slots, 2TB SSD included
- USB: 2x USB4 (40Gbps, front + rear), 3x USB 3.2 Gen2 Type-A (10Gbps), 2x USB 2.0 Type-A
- Display: 1x HDMI 2.1, 1x DisplayPort (8K@60Hz), DisplayPort Alt Mode via USB4 — up to 4 displays
- Networking: 2.5 Gigabit Ethernet, Wi-Fi 7, Bluetooth
- Other I/O: Full-size SD 4.0 card reader, 3.5mm headphone and microphone jacks
- Chassis: All-metal construction with angled vent design; bundled vertical stand
- Unique Feature: Physical 3-way performance switch (Silent / Balanced / Extreme)
- Dimensions: Approx. 220.8 x 199.5 x 57.1mm
- Configurations Available: 96GB RAM / 2TB SSD; 128GB RAM / 2TB SSD
| Configuration | 96GB RAM / 2TB SSD, 128GB RAM / 2TB SSD |
|---|
5 reviews for BOSGAME M5 AI Mini PC (Ryzen AI Max+ 395)
Fast, Tracked Worldwide Shipping
In-stock orders are processed within 1–3 business days and shipped with full tracking. Delivery costs and timeframes depend on your destination:
Delivery Estimates & Costs
- United Kingdom — 2–4 business days, $9.95 (free over $199)
- Europe — 5–10 business days, $19.95 (free over $199)
- United States & Canada — 6–12 business days, $24.95 (free over $199)
- Rest of world — 10–20 business days, $49.95 (free over $499)
Free shipping is applied automatically at checkout once your order passes the threshold for your destination — no code needed. Every order ships with a tracking number emailed to you as soon as it leaves our warehouse.
International orders may be subject to import duties or customs fees set by your local authorities; these are not included in the checkout price and are the buyer’s responsibility.
Read our full Shipping Policy for details on processing, customs, damaged or lost parcels, and delivery addresses.
Related products
Beelink EQi Pro AMD Ryzenâ„¢ 7 7735HS (OpenClaw Pre-Installed)
Beelink EQi13 Intel® Core™ 13500H/13420H
Beelink EQi Wildcat Lake Core 3 304
Beelink EQi12 Pro Intel® Core 1235U (OpenClaw Pre-Installed)
Beelink GTi14 Ultra AI PC Intel® Core™ UItra 9 185H (OpenClaw Pre-Installed)
Beelink SER9 PRO AMD Pro Ryzenâ„¢ AI 9 HX 370 (OpenClaw & Local LLM Pre-installed )
Beelink SER8 AMD Ryzenâ„¢ 7 8845HS

Ingrid L. –
Fan noise becomes noticeable under sustained multi-hour fine-tuning runs, and I did notice mild thermal throttling on the longest jobs (4+ hours continuous). Still functional and I haven’t had a hard crash, but if your workloads run long, budget for extra desk fan noise.
Sofia M. –
Running quantized 70B models locally for a fraction of what a DGX-class box costs, and the all-metal chassis feels far more premium than I expected from a smaller brand. Genuinely impressed given the price point.
Nadia F. –
Needed to spend an evening in Linux forums getting ROCm properly configured with the right environment variables — not plug and play, but the Strix Halo community has documented the process well by now. Once set up it’s been dependable for daily inference work.
Jakub W. –
Dual USB4 plus quad 8K display support made this an easy pick for my multi-monitor dev setup — didn’t expect a mini PC in this price bracket to handle four displays without a hitch.
Hassan A. –
Doubles as a genuinely capable gaming machine when I’m not running inference jobs, which isn’t something I expected to say about an AI-focused mini PC. Good value for a dual-purpose desk machine.