(Reuters) – Facebook parent Meta Platforms said on Monday that its research team has built a new AI supercomputer that it thinks will be the fastest in the world when completed in mid-2022.
Meta said in a blog post that its new AI Research SuperCluster (RSC) would help the company build better AI models that can learn from trillions of examples, work across hundreds of languages, and analyze text, images and video together to determine if content was harmful.
“This research will not only help keep people safe on our services today, but also in the future, as we build for the metaverse,” the company said in a blog post.
The social media company changed its name in October to Meta to reflect its focus on the metaverse, which it thinks will be the successor to the mobile internet.
The metaverse, a broad term which has generated a lot of Silicon Valley buzz in recent months, refers to the idea of shared virtual environments which people can access through different devices and where they can work, play and socialize.
“The experiences we’re building for the metaverse require enormous compute power (quintillions of operations/second!) and RSC will enable new AI models that can learn from trillions of examples, understand hundreds of languages, and more,” Meta chief executive officer Mark Zuckerberg wrote in a Facebook post on Monday.
The company said it believed the supercomputer was currently among the fastest AI supercomputers running.
Meta said the RSC currently comprises 760 Nvidia DGX A100 systems as its compute nodes (which means 6,080 GPUs), all of which communicate via an Nvidia Quantum 1600 Gbps InfiniBand two-level Clos fabric. The storage tier has 175 petabytes of Pure Storage FlashArray, 46 petabytes of cache storage in Penguin Computing Altus systems, and 10 petabytes of Pure Storage FlashBlade.
Meta added the AI supercomputer will continue to evolve throughout 2022: “We’ll work to increase the number of GPUs from 6,080 to 16,000, which will increase AI training performance by more than 2.5x. The InfiniBand fabric will expand to support 16,000 ports in a two-layer topology with no oversubscription. The storage system will have a target delivery bandwidth of 16 TB/s and exabyte-scale capacity to meet increased demand.”
(By Elizabeth Culliford; Reporting by Elizabeth Culliford in New York and Tiyashi Datta in Bengaluru; Editing by Bernard Orr)