llama-modchip runs 284B DeepSeek-V4-Flash IQ1 on a pure 64GB CPU with zero GPUs
pmttyji · reddit · 2026-11-12
GitHub project synchopate/llama-modchip (936 stars) claims to run IQ1-quantized 284B DeepSeek-V4-Flash (<59GiB RAM, 44.4 tok/s prompt processing, 9.1-14.7 tok/s) and Q4 30B MoEs (25-111 tok/s) on a pure 64GB CPU-only machine at 256K context, no GPU needed. Aimed at AVX-512 servers/workstations and 64GB laptops; GGUFs are on HuggingFace.
More from Infra
- Agent harnesses are heading toward vertical integration, argues 0xblacklight — HankYeomans · 2026-11-12
- CoreWeave turns to customer prepayments and shorter contracts to fund AI growth — TechCrunch AI · 2026-11-12
- Foreign Investors Pull $50B From Asian Equities in October on AI Cash-Burn and Regulation Fears — nordicinst · 2026-11-12
- Bernstein Ranks Data Center Pipelines: META, MSFT, GOOGL Top the Credibility List — aronchick · 2026-11-12
- Photoroom reclaims ~1,000 idle GPU hours weekly using SkyPilot across Slurm and Kubernetes — skypilot_org · 2026-11-12
- UK AI chip startup Fractile raises $720M to challenge in the AI silicon race — TechCrunch AI · 2026-11-12