Artificial Intelligence1 min read

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

By · Published by Everything Blog

In short

NVIDIA's new platform established top training times on six of the seven test categories while submitting results across the full suite. The architecture showed its greatest gains on communication-heavy models, completing the DeepSeek-V3 671B test in 4.14 minutes with 1,024 GPUs compared to 8.03 minutes on an equal GB300 configuration. In addition to high per-GPU performance, the platform maintained exceptional scaling efficiency exceeding 97 percent from 256 up to 1,024 GPUs. NVIDIA also demonstrated massive cluster scalability by utilizing 10,240 GB300 GPUs to capture the fastest overall DeepSeek-V3 run.

Key points

  • NVIDIA remained the sole platform provider to submit validated results across the entire…: NVIDIA remained the sole platform provider to submit validated results across the entire seven-workload benchmark round.
  • The Vera Rubin architecture proved highly capable of scaling efficiently, retaining over…: The Vera Rubin architecture proved highly capable of scaling efficiently, retaining over 97 percent efficiency between 256 and 1,024 GPUs.
  • Extreme scaling capabilities were proven at the high end by running a 10,240-GPU GB300 NV…: Extreme scaling capabilities were proven at the high end by running a 10,240-GPU GB300 NVL72 cluster for the round's fastest overall benchmark time.

Sources checked by our analysis pipeline: 1

As AI agents tackle increasingly complex workflows, reducing time-to-train is critical for rapidly iterating on both models and the agents they power. MLPerf Training v6.1 marks the debut of the NVIDIA Vera Rubin architecture, and these early results demonstrate that its CUDA software stack is already highly mature. In its first appearance, the Vera Rubin NVL72 platform delivered the fastest time to train on six of the seven benchmarks, achieving up to 2x more performance than a highly optimized NVIDIA GB300 NVL72 baseline.

For the seventh benchmark, DeepSeek-V3 671B, NVIDIA captured the absolute fastest time of the round by demonstrating massive scale on a 10,240-GPU GB300 NVL72 cluster. This dual-platform approach deliberately showcases the strengths of both systems. While the GB300 results highlight the massive, reliable cluster scale NVIDIA is deploying today, Vera Rubin demonstrated its own exceptional scaling architecture by maintaining above 97% overall efficiency when scaling from 256 to 1,024 GPUs. Together, NVIDIA platforms hold the fastest time to train on every benchmark in v6.1.

NVIDIA platforms were also the only ones to submit results across all seven…

Artificial Intelligence