AINewsnow

Stop Letting Your Model Wander Between Sockets: Practical NUMA Pinning for Local LLM Inference on Linux

A two-socket box looks like one computer until you give it a memory-hungry workload. Then it becomes two computers connected by a wire that is slower than the one inside each half — and your inference server has no idea which half it is standing on. I have watched a 32B GGUF load "successfully" and…

Read the full story at DEV Community — AI ↗

Timeline · 1 report

  1. 2026-11-01 05:03 · DEV Community — AI
    Stop Letting Your Model Wander Between Sockets: Practical NUMA Pinning for Local LLM Inference on Linux

More stories

  1. Mistral releases weights for trillion-parameter Le Chonk model with 1M-token context — Wired AI
  2. DeepSeek open-sources prototype of always-on autonomous coding agent — Techmeme
  3. Big Tech sheds nearly $900bn as AI spending fears trigger market rout — Financial Times AI
  4. DeepSeek Releases Enhanced Coding Agent to Rival Anthropic's Claude Code — Bloomberg AI
  5. Foxconn hints at new OpenAI tie-up for AI hardware, says no AI bubble — Economic Times Tech
  6. Developers Build Día de los Muertos AI Projects for Hacktoberfest Open-Source Challenge — DEV Community — AI
  7. OpenAI cuts latest funding round target to $20 billion amid tech stock selloff — Financial Times AI
  8. Renesas projects 26.4% revenue growth in 2026 on AI data center demand; Nebius leases 50 MW capacity — DigiTimes Asia

Get the daily brief of stories like this at 6:30 every morning →