Thursday, March 28

Meta says it will have the world’s most powerful AI supercomputer by mid-2022: 5 exaFLOPS to power the metaverse


Meta wants to take an important step in the field of artificial intelligence (AI). The parent company of companies such as Facebook, Instagram, Messenger, Oculus and WhatsApp has announced that by the middle of this year the supercomputer for the world’s most powerful AI research.

This is Research SuperCluster (RSC), a research system designed to train complex artificial intelligence models. It is already working in phase one, but in phase two it will greatly increase its power. “We believe it is among the fastest AI supercomputers running today,” the company says in a statement. blog post.

RSC, a supercomputer with Nvidia’s heart

RSC works at the moment with 6.080 GPUs de Nvidia DGX A100. Nodes are connected by NVIDIA Quantum-2’s high-performance InfiniBand networking fabric, which delivers 200 GB/s of bandwidth. As for storage, it has 175 petabytes of Pure Storage FlashArray, 46 petabytes of cache on Penguin Computing Altus systems, and 10 petabytes of Pure Storage FlashBlade.

The Menlo Park tech giant says high-performance computing is a key component to its AI research teams. In this sense, in 2017 it launched a system with 22.000 GPUs Nvidia V100 Tensor Core in a single cluster that currently performs about 35,000 training tasks per day. However, the new RSC system is much faster right now, in its first stage.

According to the company, RSC is 20 times faster at running machine vision workflows, nine times faster at running the NVIDIA Collective Communication Library (NCCL2), and three times faster at training NPL models a big scale. All of these values, they promise, will be vastly improved in stage two, aimed at “developing the next generation of advanced AI.”

Facebook Artificial Intelligence Supercomputer

With the goal of achieving AI training performance 2.5 times faster than the initial stage, the final configuration of the supercomputer will feature 16.000 GPUs Nvidia DGX A100. “We think it will be the fastest AI supercomputer in the world, with a performance of almost 5 exaFLOPS of mixed-precision computation“says the tech giant.

What will 5 exaFLOPS of performance be used for?

Meta says the Research SuperCluster will help create more accurate AI models for its existing services, but there’s more. Those of Menlo Park believe that this performance will open the possibility of creating “entirely new user experiences, especially in the metaverse (…) where applications and products powered by artificial intelligence will play an important role.”

Facebook Artificial Intelligence Supercomputer

The truth is that Meta is not alone in this world of high-performance computers for the development of artificial intelligence. Microsoft, in collaboration with OpenAI, launched in 2020 “one of the top five supercomputers” in this field. Nvidia, for its part, announced in 2021 the start of operations of “the most powerful supercomputer in the world” dedicated to AI.

Pictures | Goal (1 Y 2)



www.xataka.com