CoreWeave puts Nvidia Vera Rubin NVL72 into production
The first customer is Cognition, maker of Devin, which measured up to 4.8x higher token throughput per GPU than the previous generation.
Results
Cognition brought up a Vera Rubin NVL72 cluster with CoreWeave in early September and measured up to 4.8x higher total token throughput per GPU for SWE-2 inference at matched interactivity, and 3.8x higher output throughput for reinforcement learning, versus GB200 NVL72.
Access
The platform is in limited availability on CoreWeave Cloud, with hundreds of Rubin GPUs across several regions, run with the same tooling as existing GB200 and GB300 fleets.
More articles
News · Policy