ComputeLabs Research
NVIDIA Vera Rubin NVL72 achieved up to twice Blackwell Ultra’s performance in the MLPerf Training v6.1 benchmark suite.
· ComputeLabs Research · from the November 16, 2026 edition
NVIDIA’s developer announcement reports up to 2× Blackwell Ultra performance for Vera Rubin NVL72 in MLPerf Training v6.1. The qualifier “up to” is essential: the source does not state that every benchmark or workload doubled in performance.
MLCommons released the suite’s new results on November 16, with two additional benchmarks focused on recommender systems and post-training for agentic systems. Its announcement also reports record participation and new opportunities for head-to-head comparisons.
NVIDIA describes this as Vera Rubin’s MLPerf debut and says it led every Training v6.1 benchmark. Separately, it reports more than 10× generational throughput-per-watt gains for agentic AI in MLPerf Endpoints; that is an inference-efficiency claim from a different benchmark and must not be substituted for the Training v6.1 comparison.
Sources
- NVIDIA Developer Blog: NVIDIA Vera Rubin NVL72 Achieves up to 2x Blackwell Ultra Performance in MLPerf Training v6.1(developer.nvidia.com)
- HPCwire: MLCommons Releases MLPerf Training v6.1 Results, Spotlighting Recent Advances in AI Training(hpcwire.com)
- NVIDIA Blog: NVIDIA Sweeps MLPerf Training and Endpoints as Vera Rubin Debuts With Leading Performance(blogs.nvidia.com)
- NVIDIA Vera Rubin NVL72
- Blackwell Ultra
- MLPerf Training v6.1

