jain.com

DeepSeek-V3 671B trained in 17.65 minutes on 256 Rubin GPUs

Company: CoreWeave

Subject kind
product
Statement date
2026-11-16
Promised amount
17.65 minutes
Current status
stated

The claim, verbatim

CoreWeave trained DeepSeek-V3 671B to target quality in 17.65 minutes on 256 NVIDIA Rubin GPUs across multiple racks, and was the only cloud provider to submit this benchmark on Rubin.

Source (primary)

CoreWeave Becomes First Cloud Provider to Deliver Multi-Rack NVIDIA Vera Rubin Results in MLPerf® Training v6.1 - CoreWeave (-, news_article)
View cached copy (2026-11-21)Live source ↗Archive.org ↗

Quote: “CoreWeave trained the 671-billion-parameter mixture-of-experts model to target quality in 17.65 minutes on 256 NVIDIA Rubin GPUs and NVIDIA Spectrum-X Ethernet networking across multiple racks”

How we checked this

This claim has not yet been checked assertion-by-assertion against its source. It carries a cited source and quote, but the deeper check has not run. When it does, the result appears here whatever it says.

Additional evidence

confirms CoreWeave Becomes First Cloud Provider to Deliver Multi-Rack NVIDIA Vera Rubin Results in MLPerf® Training v6.1 - CoreWeave

Quote: “CoreWeave trained the 671-billion-parameter mixture-of-experts model to target quality in 17.65 minutes on 256 NVIDIA Rubin GPUs and NVIDIA Spectrum-X Ethernet networking across multiple racks”

View cached copy (2026-11-21)Live source ↗Archive.org ↗