Meta began designing computing infrastructure “from scratch” in early 2020.
What you need to know
- Meta has announced the AI Research SuperCluster (RSC), which it claims is one of the fastest AI supercomputers in the world.
- RSC will help accelerate the company’s AI research and help build better AI models for Metaverse.
- It is expected to be fully completed by mid-2022.
Facebook’s parent company Meta on January 24 announced the AI Research SuperCluster (RSC), which it says is one of the fastest AI supercomputers on the planet right now. Once it’s fully constructed by mid-2022, Meta claims it will be the fastest in the world.
RSC will help Meta’s AI researchers develop better AI models capable of learning from trillions of examples, working across hundreds of languages, creating new AR tools, and more. More importantly, RSC will allow the company to make significant strides in building Metaverse’s AI-driven applications.
With RSC, we can faster train models that use multimodal signals to determine whether an action, sound, or image is harmful or benign. This research will help ensure that people use our services not only today, but in the future as we build for the metaverse.
RSC uses a total of 760 NVIDIA DGX 100 systems as compute nodes, with a total of 6,080 GPUs. RSC’s storage tier features 175 PB of Pure Storage FlashArray, 46 PB of cache storage, and 10 PB of Pure Storage FlashBlade.
Meta plans to increase the number of GPUs to 16,000 by the end of this year, which will improve AI training performance by more than 2.5 times. The storage system, on the other hand, will have a target delivery bandwidth of 16TB/s and exabyte-scale capacity.
Early benchmarks show that RSC runs computer vision workflows 20 times faster than Meta’s previous systems. On the company’s previous system, a model with tens of billions of parameters took 9 weeks to train, and now the same model can be trained in 3 weeks on RSC.
Meta said that “larger and more complex models” need to be trained to fully realize the benefits of advanced AI for use cases, such as identifying harmful content on the various social media platforms it owns. RSC claims to be able to train models with exabyte-scale datasets, which is equivalent to 36,000 years of high-quality videos.