{"slug": "frankenstein-strata-b650m-qwen-3-8-iqxs3-32k-205k-context-32gb-ram-pcie-3-0x4-s", "title": "FRANKENSTEIN: Strata, B650M, Qwen 3.8 IQXS3, 32K/205K Context, 32GB RAM, PCIE 3.0x4?, MI25's?,  70+Tok/s 700+ Prefill? Yes. It can do that, with a little help.", "summary": "A hardware tinkerer reported running the Qwen 3.8 IQXS3 model at 32K/205K context on a 32GB RAM system with a B650M motherboard and PCIe 3.0 x4, achieving over 70 tokens per second generation and over 700 tokens per second prefill, and published the configuration and benchmark notes in the GitHub repository onewireout/mi25-llm-inference. The same repository documents testing of AMD Instinct MI25 cards for llama.cpp inference, where the cards fell well short of their theoretical performance, and the author is building a headless inference node using 4x MI25 GPUs.", "body_md": "Also working on getting inferencing working on MI25’s, did a lot of testing with them, to see where the math breaks down, as they hit no where near their theoretical.\n\nCurrently building a headless inferencing node that uses 4x MI25’s, it’s taking a long time to build it out, but it’s promising. The cards are old and slow, but that’s not a huge issue when it can chew away in the background.\n\nGot some numbers in case your interested, throw this into gpt and see if it would help you in anyway ~ [GitHub - onewireout/mi25-llm-inference: LLM inference on an AMD Instinct MI25 with llama.cpp: benchmarks, dials, patches · GitHub](https://github.com/onewireout/mi25-llm-inference)\n\nNice to see others running older cards, keep up the great work.", "url": "https://wpnews.pro/news/frankenstein-strata-b650m-qwen-3-8-iqxs3-32k-205k-context-32gb-ram-pcie-3-0x4-s", "canonical_source": "https://forum.level1techs.com/t/frankenstein-strata-b650m-qwen-3-8-iqxs3-32k-205k-context-32gb-ram-pcie-3-0x4-mi25s-70-tok-s-700-prefill-yes-it-can-do-that-with-a-little-help/259880#post_2", "published_at": "2026-11-01 09:28:50+00:00", "updated_at": "2026-11-01 09:49:55.656152+00:00", "lang": "en", "topics": ["large-language-models", "ai-infrastructure", "ai-chips", "ai-tools"], "entities": ["Qwen 3.8 IQXS3", "AMD Instinct MI25", "llama.cpp", "B650M", "onewireout/mi25-llm-inference"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/frankenstein-strata-b650m-qwen-3-8-iqxs3-32k-205k-context-32gb-ram-pcie-3-0x4-s", "markdown": "https://wpnews.pro/news/frankenstein-strata-b650m-qwen-3-8-iqxs3-32k-205k-context-32gb-ram-pcie-3-0x4-s.md", "text": "https://wpnews.pro/news/frankenstein-strata-b650m-qwen-3-8-iqxs3-32k-205k-context-32gb-ram-pcie-3-0x4-s.txt", "jsonld": "https://wpnews.pro/news/frankenstein-strata-b650m-qwen-3-8-iqxs3-32k-205k-context-32gb-ram-pcie-3-0x4-s.jsonld"}}