AMD · Local AI machine
Minisforum MS-S1 Max (128 GB)
AMD machine: AMD Ryzen AI Max+ 395, 128 GB memory, 96 GB for the GPU, 256 GB/s memory speed, about 3.9 tok/s on a 70B model (estimate).
80.3Score
- #20 of 41 ranked
- #7 of 18 AMD machines
Why this score
The score adds up these parts. Each part has a weight. A part counts as the average of its sources, pulled toward 80 when there are few sources.
AI power
50% weightcounts as 80.7Average of 2 sources: 81.1
Experts
30% weightcounts as 78.4Average of 2 sources: 77.6
Users
20% weightcounts as 81.9Average of 1 source: 83.8
Shown, not counted
These sit inside an average we already count, or come from the maker's own shop.
Good: 3 reviews and buyer ratings.
80 is the field average. A part counts as 80 + (average − 80) × N ÷ (N + 1), where N is its number of sources. One source counts half of the gap from 80. No sources count as 80.
Price
€3999 at Minisforum EU shop · Maker's EU shop, ships to Austria
128 GB, 2 TB SSD, sale price (regular price 4,999 EUR); Geizhals.at lists Proshop.at at 4,192.01 EUR plus 6.99 EUR shipping, in stock
Prices checked 8 Oct 2026.
Specs
- Chip
- AMD Ryzen AI Max+ 395
- CPU
- 16 cores / 32 threads
- Memory
- 128 GB
- GPU memory
- 96 GBWindows/BIOS: AMD Variable Graphics Memory, up to 75% of RAM = 96 GB on 128 GB (AMD rule; the largest share the vendor tools allow). Tuned Linux maxima need kernel parameters and are not used: up to 124 GiB (kyuz0 toolboxes on a Framework Desktop, Fedora, amdgpu.gttsize=126976, ttm.pages_limit=32505856, 4 GiB left to the OS), about 110 GB (runaihome), about 108 GB stable (Jeff Geerling).
- Memory speed
- 256 GB/s
- FP16 TFLOPS
- 60 TFLOPS
- NPU
- 50 TOPS
- OS
- Windows, Linux
- Software
- ROCm, Vulkan
- Ethernet
- 10 Gbit/s
- Thunderbolt / USB4
- USB4 v2
- QSFP port
- No
- Idle power
- 10.5 W
- Load power
- 120 W
- Noise
- 43 dB(A)
- Volume
- 3.5 L
- Weight
- 2.8 kg
- SSD
- 2 TB
- Release
- 2025-09
Measured
- Qwen2.5 72B (dense), Q4_K_M
- 4.5 tok/s writing · Ollama (backend not stated)
- Qwen3.5 122B, Q4_K_M
- 19.2 tok/s writing · Ollama (backend not stated)
- Qwen3-Coder-Next (51.7 GB)
- 29.5 tok/s writing · Ollama (backend not stated)
Notes
One of the best-connected Strix Halo boxes: two 10 GbE ports and two USB4 v2 ports (80 Gbit/s), plus a PCIe slot, so two boxes can be linked. Fast on 100B-class MoE models, slow on dense 70B models (about 5 tok/s) because bandwidth is 256 GB/s. Fans are audible under load and the price is high.
Idle power: ServeTheHome 9 to 12 W (performance mode), NAS Compares 13 to 16 W, minipclab 18 to 22 W; midpoint of ServeTheHome used. Load: ServeTheHome about 120 W in AI inference, NAS Compares 141 to 158 W in a heavy GPU test. Noise: ServeTheHome 43 dBA at 1 m (performance mode), NAS Compares 39 to 41 idle and 51 to 53 under load, Starry Hope relays 37 idle and 48 load. Notebookcheck: the rating box shows 84% after a BIOS update raised it (read with a fetch tool); a search snippet of a Notebookcheck table showed 90 to 92%. TechRadar's 4/5 was read via Starry Hope. Price (128 GB): Minisforum EU EUR 3,999 (sale); Geizhals.at showed Proshop.at at EUR 4,192.01, but that listing's link was not recorded. A speed run from another Strix Halo box was dropped. GPU memory follows AMD's Variable Graphics Memory rule for every box with this chip (75% of RAM: 48 GB on 64 GB, 96 GB on 128 GB); tuned Linux figures are in the GPU memory note. FP16 TFLOPS is AMD's figure for the chip.
More reading:servethehome.comnotebookcheck.netnascompares.comakitaonrails.com