Meta's Prometheus: hyperscale GPU clusters run at 87% usable, 13% kept as hot spares
SumitGup · x · 2026-10-22
Meta's engineering blog details Prometheus, its gigawatt-scale supercluster in New Albany, Ohio (phase one complete, used to train the Muse models, targeting 1M GPUs). Key numbers: NVL72 racks run 64-GPU slices with 8 in-rack spares, so a 32,768-GPU run provisions 523 racks / 37,656 GPUs (13% overhead); a 32768-GPU run provisions 523 NVL72 racks / 37,656 GPUs, 13% above active footprint, plus 2% extra racks for whole-rack faults. Pretraining achieved >85% goodput, RL runs >90%, with closed-loop liquid cooling.
More from Infra
- Omarchy package server is timing out: updates failing on cursor, grok-bot, codex-desktop packages — mark_k · 2026-10-22
- DeepSeek V4.1 Flash hits 1,521 tok/s on Fireworks Ultrafast, ships Tetris in 11 seconds — nicolechirps · 2026-10-22
- Citi projects Big Four hyperscaler capex to hit $1.62T annually by 2030, $5.35T cumulative — Beth_Kindig · 2026-10-22
- NVIDIA open-sources SimReady Central: 2,000+ sim-ready 3D assets on Hugging Face — lhoestq · 2026-10-22
- From devrel to CTO: juggling 12 cloud and AI providers and still running out of compute — NERDDISCO · 2026-10-22
- Coinbase's x402 was built to kill API keys; Stripe ships Machine Payments SDK for agent payments — kleffew94 · 2026-10-22