Meta has unveiled the AI Research SuperCluster (RSC), a new supercomputer that’s among the fastest in the world. And it’ll only get faster – by the end of the year it should rank number one, with computing power on the exascale.
The company formerly known as Facebook has had its fingers in the AI pie for a few years now, and it’s not hard to see why. Through Facebook, Instagram and Whatsapp et al, the conglomerate generates far more data than any mere mortal could possibly process – and there’s obscene amounts of money to be made in sifting through it all.
Meta’s RSC will be up to the task, using this data and immense computing power to train AI algorithms to better recognize objects in images and spoken words in audio, quickly translate between languages, and identify harmful content and misinformation that shouldn’t be on social media. And of course, it’ll probably end up powering the metaverse, which the company is trying to convince the world it needs.
In its current form, the supercomputer consists of 6,080 Nvidia A100 GPUs, linked through a Quantum InfiniBand network that can transmit data at 200 Gigabits per second. Storage-wise, the system packs an astounding 175 petabytes (PB) of bulk storage, 46 PB of cache storage and a further 10 PB of NFS storage. For reference, one petabyte is equal to a million gigabytes.
In benchmarking tests, RSC was able to train natural language processing models three times faster than Meta’s previous AI supercomputer, and was 20 times faster at computer vision tasks.
Importantly, Meta says RSC has a number of security and privacy measures in place. For one, it’s not connected to the internet at large, with data only being fed in from from the company’s own data centers. That data is encrypted all the way through the system, only being decrypted when it’s about to be used to train models.
And it’s just getting started. By mid-2022, RSC will have been upgraded with 10,000 more GPUs, a faster network that can deliver 16 terabytes of data per second, and house a grand total of 1 exabyte of storage – that’s 1,000 PB.
At that point, Meta says that RSC will have a peak performance of an insane 5 exaflops. The top supercomputer in the world at the moment – Japan’s Fugaku – peaks at “only” 2 exaflops, meaning Meta would have a mammoth lead. However, it’s worth noting that not only is RSC not fully operational yet, but it hasn’t been tested on the same benchmarking processes as Fugaku, so the numbers might not be directly comparable.
Either way, RSC will sit comfortably among the top 10 supercomputers in the world when it debuts later this year.