A “multi-year cooperation” between Microsoft and Nvidia is underway to create a supercomputer for artificial intelligence (AI) that businesses can use to train, operate, and grow AI models. The Nvidia AI enterprise software suite and the full complement of computing, networking, and storage capabilities from Microsoft Azure will be used to power the AI supercomputer, which will also run on some of the most potent Nvidia A100 and H100 GPUs from the company’s data centres.
According to Nvidia, it will make use of the scalable virtual machine instances optimised for AI training and inference that are part of Azure’s supercomputer architecture. It will employ them to speed up research in generative AI, which creates artificial material from text, photos, and videos using machine learning (ML). A generative AI application is deepfake.
Nvidia announced that it will work with Microsoft to improve DeepSpeed, a software suite for deep learning optimization. Layers of neural networks that are designed after the human brain make up the deep learning subset of machine learning. In order to speed up transformer-based models used in massive language models and generative AI models, Nvidia will provide its H100 Transformer Engine for DeepSpeed.
“Both industry adoption and AI technological advancements are accelerating. The invention of foundation models, according to Manuvir Das, vice president of corporate computing at Nvidia, “has sparked a tidal wave of research, supported new startups, and enabled new enterprise applications.”
According to Scott Guthrie, executive vice president of Microsoft’s Cloud and AI Group, “our cooperation with Nvidia unlocks the world’s most scalable supercomputer platform, which enables state-of-the-art AI capabilities for every enterprise on Microsoft Azure.”
Nvidia is engaged in a number of supercomputer-related projects. Nvidia’s GPUs are used by many of the most potent supercomputers in the world, including HPE Perlmutter, IBM’s Summit, and Sierra. On an AI supercomputer, Nvidia is also collaborating with Facebook parent company Meta. One of the world’s fastest supercomputers, developed by the research team at Meta, was revealed to have over 6,000 Nvidia GPUs in January.
