Nvidia Is Building an AI Supercomputer With Microsoft

Nvidia and Microsoft are teaming up to build “one of the most powerful AI supercomputers in the world.”

Microsoft is set to provide the infrastructure using Azure, and Nvidia will of course provide the performance through its data center GPUs, networking, and AI software. The end result is expected to produce AI that can be deployed at scale by enterprises(Opens in a new window), while allowing them to run “state-of-the-art models.”

Nvidia intends to utilize Azure’s scalable virtual machines, running on a combination of Nvidia A100 and H100 GPUs linked together using Nvidia Quantum-2(Opens in a new window) 400Gb/s InfiniBand networking. That’s tens of thousands of GPUs working together, with a key goal of this new supercomputer being to “accelerate advances in generative AI.”

Customers will be able to deploy as many GPUs as they need to act as a single training cluster for their AI workload. Microsoft’s DeepSpeed(Opens in a new window) deep learning optimization software and Nvidia’s H100 Transformer Engine will be available to accelerate modeling, with a key focus being in the areas of large language models, writing computer code, and the aforementioned generative (self-learning) AI.

Some hardware upgrades will be required before Nvidia and Microsoft can claim to have one of the most powerful AI supercomputers. At the moment, Azure instances run on A100 GPUs and 200Gb/s InfiniBand networking. The two companies are planning a future upgrade to take advantage of 400Gb/s InfiniBand.

Microsoft Studio 2+ Amps Up the Power in a Trusted All-In-One Design