Nvidia claims Vera Rubin NVL72 proxy AI offers up to 30x improved energy efficiency.
NVIDIA has released early test data for the Vera Rubin NVL72, claiming that it can achieve up to 30 times the throughput per megawatt and up to 35 times the token cost reduction compared to the GB300 NVL72 under certain proxy coding workloads.
NVIDIA releases early performance data for Vera Rubin NVL72 for agent-based AI workloads.In the real-world proxy coding track recorded by SemiAnalysis AgentX, NVIDIA claims that when the system runs DeepSeek V4 Pro, the throughput per megawatt is up to 30 times higher than that of GB300 NVL72, and the cost per million tokens is up to 35 times lower.
These figures were measured by NVIDIA and are still awaiting review by SemiAnalysis. They do not include the performance of Vera CPUs during tool invocation and cannot be directly generalized to all models and workloads.
refer to:NVIDIA Test Specifications