LATEST

← Back to GPN · All articles

NVIDIA's Vera Rubin Sweeps MLPerf Training in Benchmark Debut

NVIDIA's Vera Rubin Sweeps MLPerf Training in Benchmark Debut

OFFICIAL Written by GPN · Nov 16, 2026, 12:06 PM ET · NVIDIA Vera Rubin · Tech

NVIDIA's Vera Rubin platform has made its MLPerf debut with a clean sweep, posting results on every MLPerf Training v6.1 benchmark — the only platform to do so this round. The benchmarks cover large language models, image generation, recommender systems and reinforcement learning post-training, and faster training means smarter agents reach users sooner.

In its debut, Vera Rubin NVL72 delivered up to twice the performance per GPU of the highly optimized GB300 NVL72 system. On the DeepSeek-V3 benchmark, Vera Rubin scaled from 256 to 1,024 GPUs at about 97% efficiency, cutting training time from 16 minutes to about four. NVIDIA also ran at the largest scale in the round, 10,240 GPUs on GB300 systems, a scale Microsoft Azure matched with 10,240 GPUs on GB200 systems.

On the inference side, Vera Rubin's first MLPerf Endpoints submission delivered more than 10x the generational gains in throughput per watt for agentic AI. On the Kimi K3 agentic benchmark, Vera Rubin NVL72 delivered more than 10x the throughput per watt of GB300 NVL72 at the same response speed. NVIDIA notes the Kimi K3 result is provisional, pending verification by MLCommons.

Partners submitted extensively, with entries from 20 organizations including ASUS, Azure, Cisco, CoreWeave, Dell Technologies, Google Cloud, HPE, Oracle and Supermicro. NVIDIA says Vera Rubin also leads on independent benchmarks including SemiAnalysis AgentX and Artificial Analysis AgentPerf.

Facts gathered from

More on this