According to the official NVIDIA blog, its Vera Rubin platform has participated in MLPerf testing for the first time. In the Training v6.1 benchmarks, the single-GPU performance of the Rubin NVL72 reached approximately 2x that of the GB300 NVL72. The manufacturer also disclosed that, under comparable response speeds for the Kimi K3 agentic workflow, Rubin's throughput per watt exceeds 10x that of the GB300. This Rubin result remains preliminary pending MLCommons validation and peer review, and should not be viewed as a universal improvement across all models and workloads.