NVIDIA Vera Rubin Sweeps MLPerf Training v6.1 and Leads Agentic Inference with 10x Throughput per Watt Gains

·2026.11.17 01:00

Key point

Vera Rubin NVL72 achieved up to 2x more performance per GPU than GB300 NVL72 in training and over 10x throughput per watt in agentic inference benchmarks.

1 / 4

Details

  • NVIDIA Vera Rubin platform swept every MLPerf Training v6.1 benchmark and led agentic inference in its MLPerf Endpoints debut.
  • Vera Rubin NVL72 delivered up to 2x more performance per GPU than the highly optimized NVIDIA GB300 NVL72 system across large language models, image generation, recommender systems, and reinforcement learning post-training.
  • In agentic inference, Vera Rubin NVL72 systems provided more than 10x the throughput per watt of NVIDIA GB300 NVL72 at the same response speed for the Kimi K3 model.
  • On the DeepSeek-V3 benchmark, Vera Rubin NVL72 scaled from 256 to 1,024 GPUs with about 97% efficiency, reducing training time from 16 minutes to about four minutes.
  • The platform supports large-scale deployments, with NVIDIA running benchmarks on 10,240 GPUs using GB300 systems and Microsoft Azure matching that scale with GB200 systems.
  • Submissions for MLPerf Training came from 20 partner organizations, including ASUS, Azure, Cisco, CoreWeave, and Google Cloud, demonstrating broad ecosystem optimization.
  • Vera Rubin also showed leading performance on independent benchmarks such as SemiAnalysis AgentX and Artificial Analysis AgentPerf.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.