Microsoft Azure announced the new NDv6 GB300 VM series, featuring a supercomputing-scale production cluster of NVIDIA GB300 NVL72 systems for demanding AI inference workloads. The cluster includes over 4,600 NVIDIA Blackwell Ultra GPUs connected via NVIDIA Quantum-X800 InfiniBand. This achievement is the result of deep partnership between NVIDIA and Microsoft for AI infrastructure.

The liquid-cooled, rack-scale NVIDIA GB300 NVL72 system powers Azure’s new NDv6 GB300 VM series with 72 GPUs, 36 CPUs, 37 terabytes of memory, and 1.44 exaflops of FP4 Tensor Core performance per VM. The system supports training and inference for AI models, utilizing NVIDIA’s full-stack AI platform for breakthrough performance in reasoning AI.

Microsoft Azure’s cluster connects over 4,600 Blackwell Ultra GPUs using a two-tiered NVIDIA networking architecture with NVLink Switch fabric within each rack and Quantum-X800 InfiniBand platform for scaling beyond the rack. This setup ensures seamless communication and high-performance operation for large-scale training and inference, driving the future of AI infrastructure.

Delivering the world’s first production NVIDIA GB300 NVL72 cluster at scale required a reimagination of Microsoft’s data center layers, from custom liquid cooling to reengineered software stack. This milestone marks a significant advancement in unlocking the future of AI infrastructure. As Azure aims to deploy hundreds of thousands of Blackwell Ultra GPUs, more innovations are expected to emerge.

Read more at NVIDIA: Microsoft Azure Unveils World’s First NVIDIA GB300 NVL72 Supercomputing Cluster for OpenAI