Skip to content
NewA calmer, faster ZeroHour
StoryAI industryinfo

NVIDIA Sweeps MLPerf Training and Endpoints as Vera Rubin Debuts With Leading Performance

1 source · 3 articlesFirst reported Updated ()1
Merged summary · from the lead articleSummaries by Vani AI

Provisional Kimi K3 results claim more than 10x throughput per watt versus GB300 NVL72 and await MLCommons verification.

NVIDIA said its Vera Rubin platform swept every MLPerf Training v6.1 benchmark and, in a first MLPerf Endpoints submission, posted more than 10x generational throughput-per-watt gains for agentic inference. Vera Rubin NVL72 delivered up to 2x more performance per GPU than GB300 NVL72. On DeepSeek-V3, scaling from 256 to 1,024 GPUs kept about 97% efficiency, cutting training from 16 minutes to about four. Twenty partner organizations submitted training results. NVIDIA also cites SemiAnalysis AgentX and Artificial Analysis AgentPerf.

Key facts
  • DeepSeek-V3 on Vera Rubin scaled 256 to 1,024 GPUs at about 97% efficiency.
  • GB300 systems ran that benchmark on 10,240 GPUs, the largest scale this round.
  • Microsoft Azure matched 10,240 NVIDIA GPUs using GB200 systems.
  • The Kimi K3 Vera Rubin endpoint result is provisional and pending MLCommons peer review.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

Coverage

oldest first · one row per article · UTC
  1. · 19h ago NVIDIA BlogFirst report

    NVIDIA announces Rubin HPC GPU for science and engineering

    The Vera Rubin-platform GPU has more than 50,000 CUDA cores and seventh-generation NVLink, with availability expected in 2028.

  2. · 19h ago NVIDIA Blog

    NVIDIA expands CUDA-X libraries for agent-assisted scientific work

    New and expanded GPU libraries speed black-hole imaging, fusion, molecular dynamics, materials and single-cell analysis, some by tens to hundreds of times.

  3. · 19h ago NVIDIA Blog

    NVIDIA Vera Rubin leads MLPerf Training and agentic endpoint results

    Provisional Kimi K3 results claim more than 10x throughput per watt versus GB300 NVL72 and await MLCommons verification.