NVIDIA Blackwell Ultra Surpasses MLPerf Inference Records
NVIDIA's Blackwell Ultra architecture has redefined AI inference performance, setting unprecedented benchmarks in the MLPerf Inference v5.1 benchmark. The platform's debut showcased its dominance in handling large language models, including the 671-billion parameter DeepSeek-R1 and Llama 3.1 series models.
Blackwell Ultra delivered 1.5x higher peak NVFP4 AI compute and doubled attention-layer capabilities compared to predecessors. Enhanced HBM3e memory capacity further solidified its technical lead. These advancements position Nvidia at the forefront of accelerated computing for generative AI workloads.