Newsbite analysis · AI, Business, Cloud, Data, Hardware · New Product, Software Release
NVIDIA Newsroom
NVIDIA Vera Rubin NVL72 Delivers Leading Performance in MLPerf Inference v6.1 Debut
NVIDIA announced that its Vera Rubin NVL72 system achieved leading performance in the MLPerf Inference v6.1 benchmark. This marks an important innovation in AI inference capabilities, with tests indicating up to 3.7 times higher throughput compared to previous models. The system's ability to scale efficiently—99% scaling efficiency with 288 GPUs—highlights its robust architecture designed for maximum AI inference productivity. Continuous software optimizations are expected to enhance these results, benefitting organizations involved in AI infrastructure and deployment.
MonitorConfidence: HighShort term horizonStory: First report
EvidenceOfficial announcementCompany supplied dataCompany only
Published 16 Sept 2026, 16:40 UTC6 min read
Lead image — NVIDIA Newsroom
Brief
NewsBite reading:NVIDIA's Vera Rubin NVL72 Sets New Performance Standards in MLPerf Inference v6.1
TL;DR
- Vera Rubin NVL72 leads in MLPerf Inference v6.1 benchmarks.
- Achieved 99% scaling efficiency with 288 GPUs across four racks.
- Improved throughput by up to 1.6x over the previous v6.0 version.
- Demonstrated 30x better AI agent performance compared to earlier models.
- NVIDIA's ecosystem partners, including ASUS and Azure, showcased effective performance on multi-node systems.
Interesting facts
- NVIDIA's architecture uses enhanced Tensor Cores to achieve higher efficiency.
- The NVL72 platform allows running any model across workloads seamlessly.
- New benchmarking metrics will emerge for agentic inference, showcasing industry shifts.