AI BriefingStrix Halo 128B Benchmark
Key point
A Strix Halo 128GB system ran Mistral Medium 128B at 1.57 tokens per second.
Details
The Nimo AI Mini PC is equipped with AMD Ryzen AI Max+ 395 (Strix Halo), Radeon 8060S, 128GB LPDDR5X (121Gi visible), and a 2TB NVMe SSD, and was tested with Linux Mint 22.3 / Ubuntu 24.04 combined with Mesa 25.2.8 / ROCm 7.8.0.
- Mistral-Medium-128B-Q4_K_M (about 75GB) was loaded entirely into unified memory, leaving about 40Gi for context capacity.
- Sustained token generation speed was 1.57 tok/s, and prompt processing speed was 32.10 tok/s.
- Unified memory usage was 79Gi, peak power was 145W, and maximum fan noise was 46 dBA.
In gaming tests, DOOM Eternal at 1440p Ultra Nightmare produced 137–144 FPS, and Unigine Superposition at 4K Optimized recorded a score of 7900 with an average of 59.1 FPS.
Under sustained load (150W power limit), system power reached 154.1W, GPU 88.0°C, CPU 88.5°C, and GPU clock up to 2900MHz. Idle state was 6.1W, 40.9°C, and 27 dBA.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
vLLM Publishes Performance Benchmarks for 5 Speculative Decoding Methods on AMD GPUsStep 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis Performance