Meta To Build An AI Supercomputer:
Meta announced in January 2022 that it is building an AI supercomputer, the AI Research SuperCluster (RSC).
- Meta considers the RSC as a powerful supercomputer capable of quintillions of operations per second.
- It can perform tasks like translating text between languages and help identify potentially harmful content on Meta’s platform.
- The RSC, compared with Meta’s legacy production and research infrastructure, can run computer vision workflows up to 20 times faster, and train large-scale natural language processing (NLP) models three times faster.
- Meta estimates that a model with billions of parameters can finish training in three weeks, compared to the nine weeks it was before.
- RSC today comprises a total of 760 NVIDIA DGX A100 systems as its compute nodes, for a total of 6,080 GPUs.
- RSC’s storage tier has 175 petabytes of Pure Storage FlashArray, 46 petabytes of cache storage in Penguin Computing Altus systems, and 10 petabytes of Pure Storage FlashBlade.
- A supercomputer can perform high-level processing at a faster rate when compared to a normal computer.
- Supercomputers are made up of hundreds or thousands of powerful machines which use better artificial intelligence (AI) models to improve operations that process huge amounts of data in less time than normal computers.
- Supercomputers require high-speed and specialised chip architectures. The chip performs 660 operations per cycle and thus run up to 230 gigaflops at 350 MHz.
- AI supercomputers are built by combining multiple graphic processing units (GPUs) into compute nodes, which are then connected by a high-performance network fabric to allow fast communication between those GPUs.