Azure Trains Llama 3.1 405B in 7 Minutes on 8,192 GB200 GPUs

Azure's MLPerf Training submission used 8,192 NVIDIA GB200 NVL72 GPUs to hit the Llama 3.1 405B training target in 7.07 minutes, one of the largest Blackwell-based submissions recorded. The result reflects full-stack co-optimization across silicon, networking, and software between Microsoft and NVIDIA.