<ul><li>The latest MLPerf Inference V5.0 benchmarks show that the NVIDIA Blackwell platform has set records and achieved NVIDIA's first MLPerf submission using the NVIDIA GB200 NVL72 system designed for AI reasoning.</li><li>The complexity of manufacturing intelligence at scale in AI factories requires rapid innovation across technology layers, including silicon, network systems, and software, to maintain high inference throughput and low cost per token.</li><li>The recent updates to MLPerf Inference introduce larger and more challenging open-weight models, reflecting the constraints of production deployments. The NVIDIA Hopper platform also demonstrated exceptional performance with full-stack optimizations.</li><li>NVIDIA's Blackwell architecture, in combination with an optimized software stack, delivers new levels of inference performance, enabling AI factories to provide higher intelligence, increased throughput, and faster token rates.</li></ul>

Speed Demon: NVIDIA Blackwell Takes Pole Position in Latest MLPerf Inference Results

Discover more