Highlights:

  • Microsoft Corporation and Nvidia Corporation have partnered to develop a supercomputer optimized for running artificial intelligence software.
  • The partnership between Nvidia and Microsoft focuses heavily on software. The companies will make their respective AI development tools more accessible on Azure as part of the collaboration.

Microsoft Corporation and Nvidia Corporation have announced their partnership to develop a supercomputer optimized for running artificial intelligence software.

According to the companies, the system will be one of the most powerful AI supercomputers in the world. It will be implemented on the Azure public cloud platform from Microsoft.

Manuvir Das, the vice president of enterprise computing at Nvidia, said, “AI technology advances, as well as industry adoption, are accelerating. The breakthrough of foundation models has triggered a tidal wave of research, fostered new startups, and enabled new enterprise applications. Our collaboration with Microsoft will provide researchers and companies with state-of-the-art AI infrastructure and software to capitalize on the transformative power of AI.”

Microsoft will outfit Azure with tens of thousands of Nvidia graphics processing units as part of the agreement. Supercomputers utilize GPUs extensively due to their potential to accelerate AI and scientific applications. Microsoft will implement the GPUs alongside several other Nvidia technologies, including the chipmaker’s Quantum-2 series of network switch.

Azure already has a significant number of Nvidia GPUs. Customers of Azure have access to multiple compute instances powered by Nvidia’s A100 chip, which was the chipmaker’s flagship data center graphics card at the time of its launch in 2020. Microsoft will introduce Azure cloud instances powered by Nvidia’s current flagship data center GPU, the H100 chip, through the partnership announced.

In March, the H100 made its debut. It features 80 billion transistors and can train AI models up to six times faster than Nvidia’s A100 graphics card from the previous generation. The H100 also includes optimizations that enable it to run Transformer models, a type of advanced neural network commonly employed for tasks such as natural language processing, more efficiently.

Microsoft’s forthcoming H100-powered Azure instances will utilize Nvidia’s Quantum-2 InfiniBand switch series to manage network traffic. The switches can process traffic of 400 gigabits per second per network port, double from Nvidia’s previous-generation hardware capacity.

The partnership between Nvidia and Microsoft focuses heavily on software. The companies will make their respective AI development tools accessible on Azure as part of the collaboration.

DeepSpeed is an open-source toolkit provided by Microsoft for developers to reduce the infrastructure requirements of neural networks. Microsoft claims the toolkit can reduce the hardware required to train and operate neural networks. Microsoft and Nvidia will optimize DeepSpeed for the H100 graphics card.

The focus of the optimization effort will be to help accelerate the development of AI models that employ the popular Transformer neural network architecture. The speed boost will be provided by a feature called the Transformer Engine, which is integrated into Nvidia’s H100 graphics card. According to the chip manufacturer, the Transformer Engine accelerates neural networks by reducing the amount of data they must process to perform calculations.

Nvidia offers a software platform called Nvidia AI Enterprise to facilitate the execution of AI applications on its processors by businesses. The platform lowers the manual labor necessary to construct, implement, and administer neural networks. In addition, it contains a collection of preset neural networks designed for activities such as producing purchasing suggestions.

Nvidia certified Nvidia Enterprise AI to run on Azure instances powered by its A100 chip as part of a previous partnership with Microsoft. The firms will now collaborate to offer support for Nvidia Enterprise AI on forthcoming H100-powered Azure instances from Microsoft.

Scott Guthrie, executive vice president of Microsoft’s Cloud + AI Group, said, “Our collaboration with NVIDIA unlocks the world’s most scalable supercomputer platform, which delivers state-of-the-art AI capabilities for every enterprise on Microsoft Azure.”

As a result of the cooperation, Nvidia intends to extend its usage of Azure. The chipmaker will utilize Azure instances to help its research in the field of generative AI. This is a sophisticated neural network that can generate text, video, and software code, among other things.

Nvidia has already made substantial efforts in this field. In October of last year, the business introduced MT-NLG, a generative AI system characterized at the time as the most potent in its class. MT-NLG is equipped with 530 billion parameters, the configuration variables influencing how a neural network processes input.