This is a true superlative festival involving the Meta group, formerly Facebook. The Mark Zuckerberg company has just announced the construction of a massive supercomputer dedicated to artificial intelligence. The infrastructure, called the AI Research SuperCluster (RCS), brings together 6,080 Nvidia A100 GPU computer cores connected by a 200 Gb/s InfiniBand network and equipped with 231 petabytes of Flash storage. And it’s not over yet. By mid-2022, Meta expects to reach 16,000 processing cores, which will make it “world’s fastest AI supercomputer”, with computing power “about 5 exaflops in mixed precision”.
Such a computer sample should beat a priori on calculations of this kind “Fugaku”, the current number one of the TOP500, the ranking of the world’s largest supercomputers. This computer from the Japanese Riken Research Institute features a maximum computing power of 442010 teraflops in high precision (64 bits), but can exceed exaflops in mixed precision (32 or 16 bits).
Ability to digest excess data
RCS will allow Meta to use a dizzying amount of data to calculate its artificial intelligence model. As part of a machine learning algorithm, it will be able to use ex-grip size data sets, which represent “equivalent to 36,000 years of high-definition video”.
“We hope that RSC will help us build completely new AI systems that can, for example, provide real-time voice translation to large groups of people who each speak a different language, so they can work together in meaningful ways. . transparency to research projects or play augmented reality games together », read blog posts from Meta.
This supercomputer will also be one of Meta’s main tools for building the metaverse of the future, the virtual universe that – according to the company – the current and future internet. “where AI-driven apps and products will play a critical role”. Thanks to the incredible computing power of RCS, Meta hopes to make “the basic technology that will make the metaverse work”.
–
“Fan friendly food. Very charming explorer. Zombie nerds. Unapologetic web fans. Often fall.”
—-
–