News/Hardware
Worth readingHardware·MeaningfulOfficialbreakingUpdated Sep 16·Updated 9×·First seen Sep 16
NVIDIA Vera Rubin nvl72 Debuts in Mlperf Inference Benchmarks
Benchmark
Read up
Context that changes how you build, even if there's nothing to install.
On September 16, 2026, NVIDIA published preview performance results for its Vera Rubin NVL72 architecture in the MLPerf Inference v6.1 benchmark suite.
It defines the performance baseline for major enterprise AI training and high-speed data center inference for the coming lifecycle.
AILookup take
NVIDIA's explicit focus on optimization for the prefill phase proves they are engineering hardware directly to handle complex, long-context reasoning loops. This ensures their stranglehold on cloud providers remains intact for now.
Who cares
cloud data center providershigh-scale model training infrastructure teamshardware procurement managers
Watch next
Watch for AMD's next competitive chip release to see if they can close the 3.7x inference throughput performance gap.
Details
- Sets the performance baseline for the next two years of AI infrastructure.
- The hardware is optimized for the 'prefill' phase of LLM inference, which is a critical bottleneck.
- Indicates the continuing hardware lead NVIDIA holds over competitors like AMD and Intel.
Related articles (2)
NVIDIA Vera Rubin NVL72 Delivers Leading Performance in MLPerf Inference v6.1 DebutNVIDIA Blog· 2 stories
More in Hardware
Agility Robotics Digit Adds Safety-Certified Human Avoidance Behaviors1 source · Sep 15D-Matrix Adopts NVIDIA Nvlink Fusion for Raptor Xpu Deployment1 source · Sep 10Apple Announces Foldable iPhone Duo and Hardware-Level Image Authenticity7 sources · Sep 9Meta Announces Mtia 300 Chip with Integrated Nics4 sources · Sep 3NVIDIA Dlss 5 Launches with Real-Time Generative Video Filtering2 sources · Sep 1