Meta, the parent company of Facebook, announced that it has built an artificial intelligence supercomputer that will be used in various applications within the conglomerate’s companies, including the Mataverse.
The supercomputer, itself a series of high-powered computers connected together, goes by the name of the AI Research SuperCluster (RSC), and according to Facebook, “it is among the fastest Artificial Intelligence supercomputers, and will be the fastest of its kind when completed in mid-2022.”
It’s a statement, Meta explained that the RSC supercomputer will help the conglomerate’s researchers in tasks such as the analysis of texts, images and videos, as well as in the development of augmented reality tools. “Ultimately, the work done with RSC will pave the way for the design of the technology that will be the largest computing platform, the killverse, in which artificial intelligence applications will play a key role,” Meta pointed out.
But what is the Meta RSC supercomputer made of? The conglomerate details that the platform is made up of 760 Nvidia DGX A100 systems, with a total of 6,080 GPUs (specialized graphics computers). Storage consists of 175 petabytes of Pure Storage FlashArray, 46 petabytes of cache on Penguin Computing Atlus systems, and 10 petabytes of Pure Storage FlashBlade.
Meta added that when RSC is completed it will integrate 16,000 GPUs, and that its data flow capacity used for training AI systems will be 16 terabytes per second, “and we plan to scale it to 1 exabyte.”
Once the supercomputer has its full potential, it will have a performance of “approximately 5 exaflops of mixed-precision computation.”