NVIDIA Vera Rubin Driving Performance Per Watt, Lowest Token Cost for Partners Worldwide

Details

NVIDIA's Vera Rubin NVL72 has moved from announcement to active deployment at CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. The headline benchmark: CoreWeave's DeepSeek-R1 tests show 10x more throughput per megawatt than the previous Grace Blackwell NVL72 — the metric that drives AI factory capacity planning under power constraints. The platform co-designs seven chips and five rack trays as a unified system. The custom Olympus CPU core delivers 2x single-threaded performance, 3x core-to-core bandwidth, and 40% lower memory latency versus competing chiplet designs — optimized for agentic AI workloads. Sixth-generation NVLink provides 2x throughput and 3x lower latency on complex workloads. Spectrum-X Ethernet delivers 1.6x higher RDMA bandwidth over standard Ethernet. NVIDIA Photonics with co-packaged optics — the first such switch in volume manufacturing — achieves 5x lower power and 10x higher mean time between interruptions versus pluggable transceivers, with CoreWeave, Lambda, and OCI among early adopters. Supply chain spans 350+ factory sites in 30 countries. NVLink Fusion opens the platform to third-party XPUs. Source is an official NVIDIA blog post; independent benchmarks beyond CoreWeave's DeepSeek-R1 result are not yet available.