CoreWeave reports the first measured silicon performance for NVIDIA's Vera Rubin NVL72, demonstrating a 10-fold increase in tokens processed per megawatt compared to Blackwell. These are real results from live hardware, not projections, indicating a significant efficiency leap for next-generation AI compute.
NVIDIA's next-gen Vera Rubin GPU delivers a 10x efficiency gain over Blackwell, materially reducing the power consumption per inference token.