Meta has announced that it is building a huge AI supercomputer that will be the largest AI system in the world when completed around the middle of this year.
The social networking giant decided in early 2020 that the best way to accelerate progress in AI requires a new computing infrastructure and set about designing the new AI Research SuperCluster (RSC).
AI that is 20 times faster
As reported by trade publication Datacenter Dynamics, Meta's current supercomputer was built in 2017 by the Facebook AI Research lab with 22,000 Nvidia V100 Tensor Core GPUs in a single cluster. It currently serves as the company’s main AI system, performing some 35,000 training jobs a day.
In a blog post, Meta noted that RSC will be 20 times faster than its current clusters for production AI, and nine times faster than Meta AI’s research clusters.
The RSC system will incorporate 760 Nvidia DGX systems, each of which includes eight A100 GPUs and two CPUs, or a total of 6,080 A100 GPUs.
In comparison, Thailand’s National Science and Technology Development Agency's (NSTDA) upcoming supercomputer – touted as the largest public high-performance computing system in Southeast Asia will sport 704 A100 GPUs in total.
Other interesting technical specifications of RSC would be its 175 petabytes of flash storage from Pure Storage and 46 petabytes of cache from Penguin Computing.
RSC will help Meta build better AI models that can learn from trillions of examples, work across hundreds of languages, analyze text, images, and video to determine if the content is harmful, and develop new augmented reality tools.
In addition, work with RSC will also pave the way toward building technologies for the next major initiative at Meta – the metaverse.
“Since 2013, we have been making significant strides in AI, including self-supervised learning, where algorithms can learn from vast numbers of unlabeled examples and transformers, which allow AI models to reason more effectively by focusing on certain areas of their input,” said Meta.
“To fully realize the benefits of advanced AI, various domains, whether vision, speech or language, will require training increasingly large and complex models, especially for critical use cases like identifying harmful content.”
Additional information about RSC can be found here.
Image credit: iStockphoto/metamorworks