Home >Technology peripherals >AI >Microsoft and Nvidia join hands to build the world's most powerful AI supercomputer

Microsoft and Nvidia join hands to build the world's most powerful AI supercomputer

王林
王林forward
2023-04-16 11:13:02913browse

Microsoft is developing new supercomputers hosted on the Microsoft Azure cloud platform for artificial intelligence training and deep learning applications. The partnership with NVIDIA means that Microsoft is one of the first companies to accept the NVIDIA H100 at scale. Prior to this, Microsoft and OpenAI reached a partnership in 2019 and developed the first supercomputer on the Microsoft Azure cloud platform.

H100 is Nvidia’s flagship accelerated GPU for servers, providing higher power consumption and speed, which is 6 times faster than the previous A100 based on the Ampere architecture.

Microsoft and Nvidia join hands to build the world's most powerful AI supercomputer

Geek.com learned that H100 and A100 GPUs will be the core of Microsoft's newly conceived supercomputer, in addition to NVIDIA Quantum-2 400Gb/s InfiniBand network and NVIDIA AI Enterprise software suite. The new business will also leverage Microsoft's cloud computing infrastructure (Azure) and virtual machines (ND and NC Series).

It is reported that through this cooperation, NVIDIA hopes to make greater progress in unsupervised (or semi-supervised) algorithm learning, allowing machines to create content such as text, code, digital images, video or audio. This field is widely known as generative artificial intelligence. Nvidia will achieve this with the Megatron Turing NLG 530B, its answer to OpenAI GPT-3.

Microsoft will address artificial intelligence and deep learning workload optimization through DeepSpeed, an open source library it developed. DeepSpeed ​​can help minimize network infrastructure requirements. The collaboration also ensures that Azure customers will have access to NVIDIA’s cloud-native suite of enterprise-grade artificial intelligence and data analytics tools, software and frameworks, known as the NVIDIA AI Enterprise software suite.

Manuvir Das, vice president of enterprise computing at Nvidia, said: "Our collaboration with Microsoft will provide researchers and enterprises with the most advanced artificial intelligence infrastructure and software to harness the transformative power of artificial intelligence."

Nvidia currently owns a Selene supercomputer, which was built during the COVID-19 pandemic. It is based on A100, with a peak AI performance of 2.8 exaflops and petaFLOPS on HPL. It is used in machine learning, artificial intelligence data analysis and high-performance computing (HPC), and for training the artificial intelligence model GauGAN2. Like OpenAI’s GLIDE and DALL-E, GauGAN2 can synthesize sketches and text into photo-realistic images.

In addition, NVIDIA also has NVIDIA Eos built for advanced climate science research, digital biology and the future of artificial intelligence. It has 576 DGX H100 systems and 4608 DGX H100 GPUs and will provide 18.4 exaflops. Artificial intelligence computing performance and general scientific computing performance (HPL) of 275 petaflops are 4 times faster than Japan’s Fugaku (currently ranked second on the Top500 list).

However, Nvidia is not optimistic about Selene (No. 9 on the Top500 list) and the Eos generative artificial intelligence supercomputer under development. Scott Guthrie, executive vice president of Microsoft's cloud computing and artificial intelligence business, explained: "Our cooperation with NVIDIA will build the world's most scalable supercomputer platform and provide the most advanced artificial intelligence to every user on the Microsoft Azure cloud platform. Functionality.”

Essentially, the two companies’ collaboration aims to enable scalability in generative AI supercomputing, rather than just pure capacity enhancement.

Manuvir Das, vice president of enterprise computing at Nvidia, said: "Customers can deploy thousands of GPUs in a single cluster to train the largest language models, build the most complex recommendation systems at scale, and deliver Enabling generative artificial intelligence."

NVIDIA is also expanding 3D content, design and simulation through the Omniverse Cloud suite of tools and services. Using Omniverse Cloud, tools for developing 3D content can even run on traditional computers without GeForce or NVIDIARTX hardware or any other high-performance capabilities.

Das emphasized, “The advancement of artificial intelligence technology and industry adoption are accelerating, and breakthroughs in basic models have triggered a wave of research, cultivated new startups, and developed new enterprise applications.”

The above is the detailed content of Microsoft and Nvidia join hands to build the world's most powerful AI supercomputer. For more information, please follow other related articles on the PHP Chinese website!

Statement:
This article is reproduced at:51cto.com. If there is any infringement, please contact admin@php.cn delete