CoreWeave deploys multi-rack Nvidia Vera Rubin NVL72 cluster
Summary
CoreWeave has successfully deployed a multi-rack cluster of Nvidia Vera Rubin NVL72 systems on its cloud platform, designed specifically for large agentic AI workloads. This setup incorporates hundreds of Rubin GPUs, each providing 1.6 Tb/s of network connectivity through Nvidia Spectrum-X networking, allowing for exceptional scalability. This deployment follows CoreWeave's initial implementation of a single-rack system, marking a significant advancement in AI infrastructure. In addition to this cluster rollout, CoreWeave has enhanced its AI data management capabilities with new features such as cross-region write acceleration and a lower-cost Archive tier for AI Object Storage.
Tokens
$CRWV$NVDA
Analysis
LOTA: LOTA is CoreWeave's storage accelerator technology integrated with its AI Object Storage service. It is engineered to deliver high-throughput data access optimized for GPU workloads. The accelerator supports the company's recent rollout of lower-cost archive storage tiers for AI applications. Nvidia: Nvidia develops graphics processing units and related hardware and software optimized for artificial intelligence and high-performance computing. Its Vera Rubin architecture powers the NVL72 rack-scale systems that combine GPUs and CPUs for demanding AI training and inference tasks. Nvidia's Spectrum-X networking fabric is designed to interconnect these systems at large scale. CoreWeave: CoreWeave is an AI-focused cloud provider that specializes in GPU infrastructure for large-scale machine learning and agentic workloads. It has become the first AI cloud provider to validate and deploy Nvidia's single-rack Vera Rubin NVL72 systems before scaling to multi-rack configurations. The company recently added cross-region write acceleration and a new Archive tier to its AI Object Storage offerings. Spectrum-X: Spectrum-X is Nvidia's high-speed networking technology tailored for AI data center fabrics. It enables the interconnection of multiple NVL72 racks while providing dedicated high-bandwidth links to each GPU. CoreWeave uses Spectrum-X to create scalable multi-rack clusters for its AI customers. Vera Rubin NVL72: Vera Rubin NVL72 is Nvidia's rack-scale AI platform that integrates 72 Rubin GPUs with 36 Vera CPUs in a single enclosure. It is purpose-built for next-generation agentic AI applications requiring high interconnect bandwidth. CoreWeave has deployed this hardware in both single-rack and multi-rack configurations on its cloud platform. Storage Enhancements: The provider introduced cross-region write acceleration alongside a new Archive tier to improve AI data management capabilities. Infrastructure Scaling: CoreWeave has extended its initial single-rack Vera Rubin deployments to multi-rack clusters interconnected with Nvidia networking.
Categories
techaimachine_learningai_agentscrypto