8x RTX 5090 training run hits ~$11/b tokens amid ~50% GPU failure rates
jon_durbin · x · 2026-09-11
Developer jondurbin shared results from a distributed training run on 8x RTX 5090s: roughly $11 per billion tokens, excellent MFU, plus a work-in-progress monitoring dashboard.
He chose 5090s deliberately as a stress test, calling them "the worst GPUs on earth for stability" at sustained utilization — with about a 50% failure rate in his experience, as at least one of eight cards would randomly drop off the bus or produce NaNs. Some nodes were borrowed from @liumio.
More from Infra
- DeepSpeed shows AutoEP/AutoTP/AutoSP parallelism gains, hosts office hours — PyTorch · 2026-10-27
- Running a local 9B as a coding agent: brief the big model to plan, let the small model execute — Electrical_Ninja3805 · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- Gary Marcus: With Q3 data added, Anthropic's compute-efficiency growth shows deceleration — GaryMarcus · 2026-10-27
- SpaceX locks in $40B GPU financing deal with Apollo, the largest ever committed — TechCrunch AI · 2026-10-27
- Inference is just Kubernetes with a multi-trillion-dollar market, and the communities overlap — matt_slotnick · 2026-10-27