We are pleased to announce the launch of the first cloud private preview of the Azure ND GB200 V6 VM series, based on the NVIDIA accelerating computing platform. This latest VM, powered by the NVIDIA GB200 Grace Blackwell Superchip, features NVIDIA Grace CPUs and NVIDIA Blackwell GPUs with exceptional AI supercomputing capabilities for training state-of-the-art frontier models as well as accelerating generative inferencing.
Video: Microsoft brings new NVIDIA Blackwell platform to the cloud
The Azure ND GB200 V6 VM series is built on Microsoft’s custom server with NVIDIA Blackwell, featuring two GB200 Grace Blackwell Superchips. Each GB200 Superchip connects two high-performance Blackwell GPUs and a Grace CPU via the NVIDIA NVLink-C2C interface. With NVLink-C2C, applications gain high-speed, coherent access to a unified memory space, simplifying programming and supporting the high-speed memory needs of next-generation trillion-parameter large language models (LLMs).
Microsoft’s ND GB200 V6 VMs can scale up to 18 compute servers through NVIDIA NVLink Switch trays, enabling up to 72 Blackwell GPUs in a single NVLink domain. Additionally, these VMs are connected by the latest NVIDIA Quantum InfiniBand, allowing them to scale out to tens of thousands of GPUs for unprecedented AI supercomputing performance.
Each Microsoft server with NVIDIA Blackwell will feature the latest version of Azure Boost, a purpose-built solution that enhances the server virtualization stack for greater resilience, manageability, and security. Azure Boost supports 200 Gbps network speeds and accelerates storage performance, delivering maximum IO performance for both CPU and GPU.
Security is a top priority, and each new Microsoft server will be equipped with an Azure Integrated HSM (hardware security module). This module enhances key management functions by encrypting and ensuring signing keys always remain within the hardware security module, without compromising performance or security. It is designed to meet the requirements of FIPS (Federal Information Processing Standards) 140-3 level 3 certification.
For our customers, these capabilities provide exceptional value through enhanced performance, scalability, and reliability. With the ability to train and deploy cutting-edge AI models more quickly and efficiently, commercial enterprises can accelerate business results and stay on the cutting edge. Whether developing complex neural networks or leveraging existing models with unique datasets to make them more business-relevant, the exceptional compute power offered by the advanced architecture of the latest Azure VM series with NVIDIA GB200 Superchips, combined with Microsoft’s optimized AI software stack, means that our customers can tackle their most ambitious AI projects with confidence.
Azure ND GB200 V6 VMs will be available in a limited private preview for select partners, allowing for co-validation and co-optimization.
Microsoft
