Pure-CPU DeepSeek V4.1 Inference Project Hits 6 TPS on a Xeon for Overnight Agent Jobs
Qwen30bEnjoyer · reddit · 2026-09-11
A Reddit user open-sourced Day1DeepseekV4.1-CPU, a hacky setup running DeepSeek V4.1 entirely on a lab Xeon server — 30 TPS prefill and 6 TPS decode with n-gram table offload on 50% of threads. The idea: unlimited slow tokens for overnight or multi-day agentic jobs, with a watcher script that kills the process within 15 seconds if someone else needs the machine for genome assemblies. The code was vibecoded by Opus 5.0 and unreviewed; the author hopes kernel experts will contribute PRs or forks.
More from Infra
- Meta approves nonvoting Class C shares as 2026 capex guidance hits $135B–$145B — thione · 2026-10-26
- Unsloth Ships MiniMax-M3.1-Flash Quants: 2-bit Retains 91.5% Accuracy, Runs in 150GB RAM — danielhanchen · 2026-10-26
- Engineer with thousands of fabricated devices pours cold water on photonic crystal modulator yields — jwt0625 · 2026-10-26
- Anthropic S-1 reveals 48 cents of compute cost per dollar of Q2 revenue — nmasc_ · 2026-10-26
- Persistent AI joins UR network to route private AI inference with encrypted, IP-hidden traffic — 0xSammy · 2026-10-26
- Anthropic's disclosed compute commitments: $114.5B Google, $161.2B Broadcom top the list — BenBajarin · 2026-10-26