Microsoft, Nvidia collaborate to build giant AI supercomputer in the cloud
Nvidia and Microsoft announced Wednesday a multi-year partnership to build an AI supercomputer in the cloud, adding tens of thousands more. Nvidia GPU arrive Microsoft Azure.
The new deal makes Azure the first public cloud to incorporate Nvidia’s full AI stack — its GPU, network, and AI software.
By augmenting Azure’s infrastructure with Nvidia’s full AI suite, many businesses will be able to train, deploy, and scale AI — including the most modern, large models.
“Advances in AI technology and industry adoption are accelerating,” said Manuvir Das, Vice President of Enterprise Computing at Nvidia. “The disruption of platform models has triggered a wave of research, spurred new startups and enabled new enterprise applications.”
Also: 8 ways to reduce AI burnout
Azure instances already have an Nvidia A100 GPU paired with an InfiniBand Nvidia Quantum 200Gb/s network. Future versions will be integrated with the Nvidia H100 GPU and the InfiniBand Nvidia Quantum-2 400Gb/s network.
Meanwhile, Nvidia’s software suite, Nvidia AI Enterprise, is certified and supported on Microsoft Azure instances with the A100 GPU. Support for Azure instances with the H100 GPU will be added in a future software release.
Azure enterprise customers will also have access to Nvidia’s entire AI workflow and software development kit, optimized for Azure. The companies will also work together to optimize Microsoft’s DeepSpeed deep learning optimization software.
The new deal also gives Nvidia the opportunity to use the Azure virtual machine to advance research in general AI.