Initial benchmarks put Vera Rubin NVL72 at up to ~5.7x GB300 throughput per megawatt on DeepSeek V4 Pro

Artificial Analysis reports its AA-AgentPerf benchmark found Nvidia's Vera Rubin NVL72 reached up to ~5.7x GB300 NVL72's total throughput per megawatt on DeepSeek V4 Pro. Nvidia submitted every Vera Rubin result. At peak throughput on that model, the advantage was 1.14x; the larger gap appears at faster per-user speeds. A commenter praised the user-facing inference results but questioned their value for large-scale reinforcement learning.