NVIDIA reports Vera Rubin leads MLPerf Training and agentic inference benchmarks
Reported by NVIDIA, Traders Union, Business Insider and 2 more outlets
NVIDIA says Vera Rubin had the fastest results on six of seven MLPerf Training v6.1 benchmarks and led agentic inference tests. Its reported 10x-plus throughput-per-watt result is provisional and awaits verification.
Key points
- NVIDIA says Vera Rubin had the fastest result on six of seven Training benchmarks
- NVIDIA reports up to twice the per-GPU performance of its GB300 system
- The 10x-plus throughput-per-watt comparison is provisional and pending MLCommons verification
- The new training round includes MLPerf’s first post-training benchmark
Sources
- NVIDIARead ↗NVIDIA Sweeps MLPerf Training and Endpoints as Vera Rubin Debuts With Leading Performance
- Traders UnionRead ↗NVIDIA sweeps MLPerf benchmarks as Vera Rubin posts AI performance gains
- Business InsiderRead ↗MLCommons Releases MLPerf Training v6.1 Results, Spotlighting Recent Advances in AI Training
- MLQ.aiRead ↗Nvidia unveils Rubin HPC, a science GPU due in 2028, and posts Vera Rubin's first MLPerf results
- Tech TimesRead ↗MLPerf Training v6.1 Adds Post-Training Test; NVIDIA, Google, CoreWeave Results Not Comparable
Details
MLPerf Training v6.1 included the first results for NVIDIA’s Vera Rubin GPUs and Google’s Ironwood TPU, as well as a new reinforcement-learning post-training test. Tech Times reports that MLCommons published results from 24 organizations, covering 153 systems and 250 results, but does not name winners. NVIDIA says its platform was the only one with submissions across every benchmark and had the fastest time on six of seven.
NVIDIA reports up to twice the per-GPU performance of an optimized GB300 NVL72 system. It also claims more than 10 times the throughput per watt of GB300 on an agentic inference test using Kimi K3. NVIDIA labels the Vera Rubin result provisional and says it is awaiting MLCommons verification; the reported Rubin and GB300 runs used different accelerator counts and query concurrency.
Written from coverage as of
Images
- NVIDIA
- Tech Times
- NVIDIA
- NVIDIA
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh