Understand this story
NVIDIA Vera Rubin NVL72 made its first appearance in the MLPerf Training v6.1 benchmark tests. According to information released by the NVIDIA Newsroom and NVIDIA Developer Generative AI, the platform achieved the fastest training time in six of the seven benchmarks, with single-GPU performance up to 2 times that of the optimized GB300 NVL72. On the DeepSeek-V3 671B benchmark, Vera Rubin NVL72 completed training in 4.14 minutes using 1,024 GPUs, compared with 8.03 minutes for GB300 NVL72 at the same scale; when scaling from 256 to 1,024 GPUs, it maintained more than 97% scaling efficiency. In the newly added Qwen 3.5 397B GRPO post-training benchmark, Vera Rubin NVL72 led with 10.44 minutes and simultaneously submitted both TensorRT-LLM and vLLM inference backends. The NVIDIA Newsroom also stated that in its first Endpoints submission, Vera Rubin achieved a generational improvement of more than 10 times in agentic AI throughput per watt, and in preliminary Kimi K3 results, throughput per watt exceeded 10 times that of GB300 NVL72 at the same response speed.
AI-generated from public reports
Latest development · Report
NVIDIA Vera Rubin Debuts in MLPerf, Sweeps Training Benchmarks and Announces Endpoints ResultsDevelopments
1 confirmed developments
NVIDIA Vera Rubin NVL72 makes its debut in MLPerf Training v6.1
Source:NVIDIA:Newsroom
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 17
NVIDIA:NewsroomFirst-party
NVIDIA Vera Rubin Debuts in MLPerf, Sweeps Training Benchmarks and Announces Endpoints Results
The NVIDIA Vera Rubin platform swept all benchmarks in Training v6.1 in its first MLPerf benchmark test, and achieved a generational improvement of more than 10x throughput per watt for agentic AI in its first Endpoints submission. On the DeepSeek-V3 benchmark, Vera Rubin NVL72 maintained approximately 97% efficiency when scaling from 256 to 1024 GPUs, reducing training time from 16 minutes to about 4 minutes, with per-GPU performance up to 2x that of GB300 NVL72. In preliminary results for Kimi K3, Vera Rubin NVL72 achieved more than 10x the throughput per watt of GB300 NVL72 at the same response speed.
NVIDIA:NewsroomRelated discussion
NVIDIA:Newsroom joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
NVIDIA Developer · Generative AIFirst-party
NVIDIA Vera Rubin NVL72 in MLPerf Training v6.1 reaches up to…
NVIDIA Vera Rubin NVL72 participated in MLPerf Training v6.1 for the first time, achieving the fastest training time in six of the seven benchmarks, with per-GPU performance up to 2x that of the optimized GB300 NVL72. On the DeepSeek-V3 671B benchmark, it completed training in 4.14 minutes using 1,024 GPUs, compared with 8.03 minutes for GB300 NVL72 at the same scale; scaling from 256 to 1,024 GPUs maintained over 97% scaling efficiency. In the newly added Qwen 3.5 397B GRPO post-training benchmark, Vera Rubin NVL72 led with 10.44 minutes and submitted both TensorRT-LLM and vLLM inference backends.