Rongchai Wang
Oct 09, 2025 22:13
Microsoft Azure companions with NVIDIA to unveil the world’s first GB300 NVL72 supercomputing cluster, designed to reinforce AI mannequin growth and solidify U.S. management in AI know-how.
Microsoft Azure, in collaboration with NVIDIA, has launched a pioneering supercomputing cluster, the NVIDIA GB300 NVL72, designed to satisfy the rigorous calls for of AI mannequin growth. This modern platform is ready to bolster the US’ standing within the AI sector, based on a latest announcement by Microsoft.
Revolutionizing AI Infrastructure
The newly launched NDv6 GB300 VM collection represents the business’s first supercomputing-scale manufacturing cluster using NVIDIA GB300 NVL72 techniques. This initiative is particularly tailor-made to assist OpenAI’s superior AI inference workloads. The cluster includes over 4,600 NVIDIA Blackwell Extremely GPUs, interconnected by the NVIDIA Quantum-X800 InfiniBand networking platform, guaranteeing excessive inference and coaching throughput for complicated AI fashions.
This growth signifies a milestone within the long-standing partnership between NVIDIA and Microsoft, aimed toward developing AI infrastructure able to dealing with essentially the most demanding workloads. Nidhi Chappell, Company Vice President of Microsoft Azure AI Infrastructure, emphasised the importance of this achievement, highlighting the shared dedication of Microsoft and NVIDIA to optimize trendy AI knowledge facilities.
The Powerhouse: NVIDIA GB300 NVL72
Central to Azure’s new providing is the liquid-cooled, rack-scale NVIDIA GB300 NVL72 system. Every rack integrates 72 NVIDIA Blackwell Extremely GPUs and 36 NVIDIA Grace CPUs, creating a strong unit for accelerating coaching and inference processes in large-scale AI fashions. This method boasts 37 terabytes of quick reminiscence and 1.44 exaflops of FP4 Tensor Core efficiency per VM, important for dealing with reasoning fashions and multimodal generative AI.
The NVIDIA Blackwell Extremely platform, supported by NVIDIA’s full-stack AI platform, excels in each coaching and inference. Current MLPerf Inference v5.1 benchmarks demonstrated record-setting efficiency, showcasing as much as 5 instances increased throughput per GPU on substantial AI fashions in comparison with earlier architectures.
Superior Networking and Scalability
The supercomputing cluster employs a two-tiered NVIDIA networking structure to attach over 4,600 GPUs, guaranteeing each scale-up and scale-out efficiency. Inside every rack, the NVIDIA NVLink Swap material gives 130 TB/s of bandwidth, remodeling the rack right into a unified accelerator with a shared reminiscence pool. For broader scalability, the cluster makes use of the NVIDIA Quantum-X800 InfiniBand platform, providing 800 Gb/s of bandwidth per GPU for seamless communication throughout all the system.
Microsoft Azure’s cluster additionally incorporates NVIDIA’s superior adaptive routing and congestion management capabilities, enhancing the effectivity of large-scale AI coaching and inference operations.
Envisioning the Way forward for AI
The deployment of the world’s first manufacturing NVIDIA GB300 NVL72 cluster marks a big development in AI infrastructure. As Microsoft Azure goals to develop its deployment of NVIDIA Blackwell Extremely GPUs, additional improvements are anticipated, pushed by prospects comparable to OpenAI. This growth is predicted to unlock new potential in AI know-how, paving the best way for future breakthroughs.
For extra data on this announcement, please go to the official weblog of NVIDIA.
Picture supply: Shutterstock
