AI & Tech··1 sources verified
NVIDIA's Vera Rubin Debuts in MLPerf With Up to 2x Training Gains Over GB300
Summarised by Relevant News AI · Read time: 3 min
NVIDIA's Vera Rubin NVL72 swept all seven MLPerf Training v6.1 benchmarks, delivering up to 2x per-GPU performance over a tuned GB300 NVL72 baseline. On DeepSeek-V3 671B, it trained in 4.14 minutes on 1,024 GPUs versus 8.03 minutes for GB300, with about 97% scaling efficiency from 256 to 1,024 GPUs. NVIDIA's claimed 10x throughput-per-watt advantage on agentic inference in MLPerf Endpoints remains provisional pending MLCommons verification, and it rests on one configuration in which Vera Rubin's time to first token was slower than GB300's.
Why it matters: The training results are the first benchmark data for the Rubin generation, so infrastructure teams planning GPU purchases should weigh the verified training gains more heavily than the unverified inference claims.
All sources