{"slug": "running-176b-qwen3-8-flash-next-and-284b-deepseek-v4-on-32gb-64gb-cpu-ram-0-gpus", "title": "Running 176B Qwen3.8-Flash-Next and 284B DeepSeek-V4 on 32GB/64GB CPU RAM (0 GPUs) with llama-modchip v3.0", "summary": "Synchopate released llama-modchip v3.0, which runs the 176.5B Qwen3.8-Flash-Next model on 32 GB and 64 GB of CPU RAM with 0 GPUs at roughly 14-18.2 tok/s on a desktop Ryzen 9 7950X/9950X using its built-in blk.48 MTP draft head, with under 1s multi-turn time-to-first-token in Claude Code. The same release updates the 284B DeepSeek-V4-Flash build, which fits in 64 GB of RAM at about 10.1 tok/s on DDR5, and the 30B Nemotron-3.5 build, which runs in 16 GB of RAM at 30-60 tok/s, alongside full v3.0-r2 abliterated variants for local security and red-team testing. The code and a one-line installer are on GitHub, and the 16 GB, 32 GB and 64 GB model builds are published in the MaxQ v3.0 Hugging Face collection.", "body_md": "We just shipped llama-modchip v3.0: 176.5B Qwen3.8-Flash-Next now runs on 32 GB and 64 GB CPU RAM with 0 GPUs (about 14-18.2 tok/s on a desktop Ryzen 9 7950X/9950X using its built-in blk.48 MTP draft head, <1s multi-turn TTFT in Claude Code).\n\nWe also updated our 284B DeepSeek-V4-Flash (fits in 64 GB RAM, ~10.1 tok/s on DDR5) and 30B Nemotron-3.5 (16 GB RAM, 30-60 tok/s) builds, plus full v3.0-r2 abliterated variants for local security/red-team testing.\n\n• GitHub (code + 1-line installer): [GitHub - synchopate/llama-modchip: Run 284B DeepSeek-V4-Flash (<59GiB RAM: 41.1 cold pp, 9.1 N=1 tg, up to 25.3 spec tok/s) & 30B MoEs (22.4 N=1 / 111.6 spec tok/s) on 64GB CPU (0 GPUs). · GitHub](https://github.com/synchopate/llama-modchip)\n\n• v3.0 Model Collection (16GB / 32GB / 64GB): [MaxQ v3.0 - a Synchopate Collection](https://huggingface.co/collections/Synchopate/maxq-v30-6afb6e2d63dfd573b4b1a95f)\n\nIf you test it on your CPU, let me know what hardware and tok/s you get!", "url": "https://wpnews.pro/news/running-176b-qwen3-8-flash-next-and-284b-deepseek-v4-on-32gb-64gb-cpu-ram-0-gpus", "canonical_source": "https://discuss.huggingface.co/t/running-176b-qwen3-8-flash-next-and-284b-deepseek-v4-on-32gb-64gb-cpu-ram-0-gpus-with-llama-modchip-v3-0/192008#post_1", "published_at": "2026-11-18 00:29:55+00:00", "updated_at": "2026-11-18 00:40:04.305542+00:00", "lang": "en", "topics": ["ai-infrastructure", "large-language-models", "ai-tools", "mlops"], "entities": ["Synchopate", "llama-modchip", "Qwen3.8-Flash-Next", "DeepSeek-V4-Flash", "Nemotron-3.5", "GitHub", "Hugging Face", "Ryzen 9 7950X"], "also_reported_by": [], "takeaway": "llama-modchip v3.0 runs the 176.5B Qwen3.8-Flash-Next on 32-64 GB of CPU RAM with no GPU at 14-18.2 tok/s on a Ryzen 9 7950X/9950X, using the model's built-in blk.48 MTP draft head for speculative decoding.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/running-176b-qwen3-8-flash-next-and-284b-deepseek-v4-on-32gb-64gb-cpu-ram-0-gpus", "markdown": "https://wpnews.pro/news/running-176b-qwen3-8-flash-next-and-284b-deepseek-v4-on-32gb-64gb-cpu-ram-0-gpus.md", "text": "https://wpnews.pro/news/running-176b-qwen3-8-flash-next-and-284b-deepseek-v4-on-32gb-64gb-cpu-ram-0-gpus.txt", "jsonld": "https://wpnews.pro/news/running-176b-qwen3-8-flash-next-and-284b-deepseek-v4-on-32gb-64gb-cpu-ram-0-gpus.jsonld"}}