LLMs max out hardware: former xAI exec reports failing PSUs, tripping breakers
MParakhin · x · 2026-11-09
Former xAI exec MParakhin notes a new challenge: LLMs are so good at fully utilizing compute that hardware is being pushed beyond expected operating conditions — continuous CPU+GPU+memory power draw and cooling load is causing PSU failures and tripped circuit breakers.
More from Infra
- Agentic AI is orchestration of model calls; datacenter limits may push edge models — DavidLinthicum · 2026-11-09
- "Bro got distilled": meme pairs Jensen Huang photo with Huawei's Nvidia chip challenge — StephanSturges · 2026-11-09
- SpaceX plans $24 billion AI supercomputer in Starbase, Texas — beffjezos · 2026-11-09
- Megaspeed allegedly imported $4.6B of Nvidia GPUs, $2B landing in Malaysia DCs — bytebot · 2026-11-09
- GLM-130B Training Cost About $600K: Two Months on 96 A100s — ccerrato147 · 2026-11-09
- $16K CAD 4x R9700 local inference server: 190 tok/s per user, 8 users at 128K context benchmarked — Anarchaotic · 2026-11-09