Meta: Facebook owner wants to build ‘the most powerful AI supercomputer in the world’

Facebook-Meta-FTC (Copyright 2021 The Associated Press. All rights reserved.)

Meta says it wants to build the most powerful artificial intelligence supercomputer in the world.

The Facebook owner has already designed and built what it calls the AI Research SuperCluster, or RSC, which it says is among the fastest AI supercomputers in the world.

It hopes to top that league by mid-2022, it said, in what would be a major step towards increasing its artificial intelligence capabilities.

That is partly focused on the metaverse, which Meta has staked its future on. With that new technology, “AI-driven applications and products will play an important role”, it said in its announcement.

The RSC is being used to create models that are being used for natural-language processing as well as computer vision, Meta said. It hopes to use it for speech recognition and other technologies in the future, it said.

But it also hopes that it can build entirely new AI systems, it said. Those could include the ability to provide real-time voice translations for large groups of people – so that a big team, each speaking a different language, could work or play together, it said.

That will require a host of technology that is not yet fully realised. It would need to be able to analyse large amounts of data, in loud scenarios such as concerts, and understand it in different languages, dialects and accents.

Meta has been working heavily on artificial intelligence since 2013, when it built the Facebook AI research lab. It has built computing technologies for that work before – but the RSC is the largest and most spectacular yet.

The plan to build a new computing infrastructure to power that work, based on a totally clean slate, began in early 2020, Facebook said. The aim was to allow it to train models that could analyse data sets as big as an exabyte – equivalent to 36,000 years of high-quality video.

RSC is already working today. But Meta says it will continue to grow it – building from just over 6,000 GPUs today to 16,000, and in so doing increasing its AI training performance by 2.5 times.

That will allow it to improve the artificial intelligence in its existing services, “but also to enable completely new user experiences, especially in the metaverse”, Meta said in a blog.