Meta AI Research Supercomputer Aims to Be World’s Fastest

The immersive experiences Meta Platforms has planned for the metaverse will require processing power beyond what’s possible today, tipping into the quintillions of operations-per-second. But the company has a fix for that. Meta announced Monday that it has been working on a new artificial intelligence supercomputer called the AI Research SuperCluster, that will be the fastest in the world when fully built this summer. Drawing on resources from partners like Nvidia, Penguin Computing and Pure Storage, the effort took two years and involved several hundred people who collaborated remotely through the COVID-19 pandemic for much of the project.

At present, Meta is “using the supercomputer to train AI models in natural-language processing and computer vision for research,” writes The Wall Street Journal. “The aim is to boost capabilities to one day train models with more than a trillion parameters on data sets as large as an exabyte, which is roughly equivalent to 36,000 years of high-quality video.”

Meta CEO Mark Zuckerberg provided WSJ with a statement that said, “The experiences we’re building for the metaverse require enormous compute power … and RSC will enable new AI models that can learn from trillions of examples, understand hundreds of languages, and more.”

Meta’s AI Research SuperCluster is packed with 6,080 Nvidia GPUs, enough to make it the fifth fastest of the world’s supercomputers, Meta says, noting that by mid-summer it will contain approximately 16,000 GPUs, making it the fastest AI supercomputer in the world, capable of “performing at nearly 5 exaflops of mixed precision compute,” according to a company blog post.

Meta says “RSC will help researchers develop better AI models that can learn from trillions of examples. Among other things, the models will be able to build better augmented reality tools and ‘seamlessly analyze text, images and video together,’” Engadget reports.

Meta vice president of AI Jerome Pesenti says in WSJ it will be a tremendous advantage for the company’s researchers “to train a single model with a lot more compute than anybody else in the world has access to. We believe that this will really unlock an AI that just understands the world around you much, much better.”

Bill Gropp, director of the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, speculates in WSJ that Meta will be looking at “the treasure trove that they have of the data that they get from their users, and that’s a huge data set, too big for most of the artificial intelligence systems that researchers typically use.”

Initially, the AI Research SuperCluster will be used for research purposes, Pesenti confirmed to WSJ, although it is unlikely that products will arise from it for a number of years. The immediate goal “is to ingest troves of data to build AI models that can think like a human brain, with multiple inputs — such as voice and visual recognition — and can deliver contextual understanding of situations” in multiple languages.

“You want to have an artificial intelligence that’s a lot more context aware, that understands all the subtleties and a level of granularity that we can’t match today,” Pesenti explained.

Related:
Meta Is Developing a Record-Breaking Supercomputer to Power the Metaverse, VentureBeat, 1/24/22