iCreatAI News
Skip to content
Trending storyUpdating

Vera Rubin NVL72 debuts on MLPerf training榜单

2 reports2 sources2 hr agoUpdated

Understand this story

AI overview

NVIDIA announced Vera Rubin NVL72's first results on MLPerf Training v6.1, saying its per-GPU performance reaches up to 2x that of GB300 NVL72 and that it has results on all training benchmarks. This is the first time the Vera Rubin architecture has appeared on the MLPerf training榜单. NVIDIA says the platform's CUDA software stack is already fairly mature and that it leads in training, post-training, and agentic inference. However, NVIDIA also states that the Vera Rubin results are preliminary, pending validation and peer review by the MLCommons Association, and may change; the GB300 results used for comparison have been validated by the MLCommons Association. The above performance multiples and maturity are all NVIDIA's claims.

AI overview based on source reports · Updated 2 hr ago

Coverage timeline

Follow the coverage to explore different perspectives.

Nov 17
  1. NVIDIA BlogFeatured
    NVIDIA Sweeps MLPerf Training and Endpoints as Vera Rubin Debuts With Leading Performance

    NVIDIA's Vera Rubin platform debuted in MLPerf with results on every MLPerf Training v6.1 benchmark, delivering up to 2x more performance per GPU than GB300 NVL72, and led all of its agentic benchmarks in its first MLPerf Endpoints submission, including more than 10x throughput per watt versus GB300 NVL72 on a provisional Kimi K3 result. On DeepSeek-V3, Vera Rubin NVL72 scaled from 256 to 1,024 GPUs with about 97% efficiency, cutting training time from 16 minutes to about four, while GB300 and Azure GB200 systems ran at 10,240 GPUs. Twenty organizations submitted MLPerf Training results on the NVIDIA platform.

  2. NVIDIA Technical Blog
    NVIDIA Vera Rubin NVL72 Achieves up to 2x Blackwell Ultra Performance in MLPerf Training v6.1

    MLPerf Training v6.1 marks the debut of the NVIDIA Vera Rubin architecture, with the Vera Rubin NVL72 platform delivering the fastest time in its first appearance. NVIDIA says these early results show its CUDA software stack is already highly mature, and the platform achieved up to 2x Blackwell Ultra performance.

Activity for this story

Not enough continuous observations to show a trend yet.