Microsoft introduces the ND H100 v5 instance family for its Azure cloud platform. The instances are suitable for AI workloads and revolve around the Nvidia H100 GPU.

Microsoft introduces a new family of instances for Azure: the ND H100 v5 series. As the name suggests, these instances are built around the Nvidia H100 GPU. This powerful chip, based on the Hopper architecture, makes the solution ideal for customers who need the necessary performance for AI-related workloads.

The ND H100 v5 series is aimed at users with heavy workloads. Each instance ships with eight H100 GPUs. They are internally connected via NVLink technology, which in this case is more powerful than PCIe.

AI supercomputer

Thanks to the Nvidia Quantum-2 InfiniBand connection, you can chain multiple ND H100 v5 servers together. For example, you can create an HPC cluster in Azure with theoretically thousands of H100. The computing power comes from scalable fourth-generation Intel Xeon processors. Although AMD’s chips are more powerful and economical, Microsoft is ignoring them for this new offering for now.

With the ND H100 v5 instances, Microsoft gives companies the opportunity to build their own exascale AI supercomputer in the cloud. This is useful and even necessary for training advanced AI models. Microsoft is now making the new instances available in the preview. Over time, they will become a standard part of the Azure offering.