slb350.github.io/strix-benchmarks
Local LLM benchmark results on AMD Strix Halo (Ryzen AI MAX+ 395, Radeon 8060S, 128 GB UMA, RDNA 3.5). I test 26+ models across three Vulkan/ROCm backends (RADV, AMDVLK, ROCm) using llama.cpp, then score them on creative writing, polyglot coding, and PostgreSQL tasks.
The site includes:
- Models — per-model performance (prompt processing, token generation) and quality scores across all benchmark suites
- Methodology — how each benchmark suite works, scoring criteria, and test conditions
- Quantization — comparison of quantization formats and their effect on speed and quality
Built with Astro and deployed to GitHub Pages.