Pyyan / News / 16 November 2026

BenchmarksNVIDIA · MLCommons

NVIDIA's Vera Rubin NVL72 debuted in MLPerf Training v6.1 at up to 2x GB300 per GPU

2xup to this much performance per GPU over GB300 NVL72, by NVIDIA's reading

MLCommons released MLPerf Training v6.1 results on 16 November. MLPerf Training times how fast systems train models to a set accuracy. NVIDIA says it was the only platform with results on every benchmark. It says Vera Rubin NVL72 gave up to 2x the per GPU performance of GB300 NVL72. On DeepSeek-V3, Vera Rubin went from 256 to 1,024 GPUs at about 97% efficiency. Training time fell from 16 minutes to about four. These readings are NVIDIA's own summary of the results.

NVIDIAfrom the source itself

Related

← All the news, newest first