The name of the base system (platform): | Nvidia DGX Supercomputers |
Developers: | Meta (formerly Facebook) |
Date of the premiere of the system: | January 2022 |
Technology: | Supercomputer |
2022: Development of a supercomputer with a planned capacity of 5 exaflops
At the end of January 2022, Meta announced the development of an artificial intelligence (AI) supercomputer. It is called Research SuperCluster (RSC) and is positioned as the fastest in the world. True, for the system to officially acquire such a status, it needs refinement, which the developers promise to carry out by mid-2022.
Meta noted that developing the next generation of advanced AI will require powerful new computers capable of performing quintillions of operations per second.
Meta's RSC cluster is based on hundreds of Nvidia DGX systems networked using Nvidia Quantum InfiniBand switches to speed up research teams.
The new AI supercomputer uses 760 Nvidia DGX A100 systems as computing nodes. In total, they contain 6080 Nvidia A100 GPUs networked using the Nvidia Quantum 200 Gb/s InfiniBand, which provides TF32 performance in 1895 petaflops.
Meta's preliminary tests showed that the RSC can train large NLP models 3 times faster and perform computer vision tasks 20 times faster than the 2017 system.
In the second stage, RSC will expand to 16 thousand GPUs, which, according to Meta, will provide performance in AI tasks of 5 exaflops in operations with mixed accuracy. Meta also plans to expand the RSC storage system to an exabyte at 16 TB per second.
We hope the RSC cluster will help us build completely new AI systems that can, for example, provide real-time voice translation to large groups of people who speak different languages so they can collaborate on research projects or play augmented reality games, Meta said in a statement.[1] |