CoreWeave edges up after bringing multi-rack Vera Rubin cluster online

CoreWeave, a cloud computing provider, has recently launched multi-rack Nvidia Vera Rubin clusters on its platform. This significant move enables the connection of hundreds of Nvidia Rubin GPUs into a single, scalable cluster, allowing for efficient training and inference jobs. Each Nvidia Vera Rubin NVL72 rack contains 72 GPUs, along with 36 CPUs, advanced networking components, and specialized processing units.

CoreWeave's executive vice president, Chen Goldberg, highlighted the achievement of validating and deploying this powerful configuration. The company also introduced two new capabilities within its AI Object Storage system. Cross-region write acceleration allows for local data writing with automatic background replication to other regions, while the new Archive tier offers cost-effective storage with reduced fees.

The system delivers reads at high speeds, providing up to 7 GB/s of throughput per GPU. CoreWeave's AI Object Storage ensures a unified dataset footprint across regions, with locally cached reads for enhanced performance.