This system is designed to train the next generation of machine learning systems. It is expected to be completed by the middle of this year, and will consist of about 16,000 graphics processors.
The Meta Technology conglomerate is the latest technology giant to build a “supercomputer UI” – a supercomputer designed specifically for training machine learning systems. The company claims that the new AI Research SuperCluster, or RSC, is already among the fastest devices of its kind and will be the fastest in the world after its completion in mid-2022.
“Meta has developed what we consider to be the world’s fastest AI supercomputerMeta CEO said in a statement, Mark Zuckerberg. “We called it RSC (AI Research SuperCluster) and it will be completed later this year,” he added.
The news shows how important AI research is to companies like Meta. Competitors such as Microsoft and Nvidia have already announced their “artificial intelligence supercomputers” that differ slightly from traditional supercomputers. The RSC system will be used to train different systems in Meta companies: from content-modifying algorithms used to detect hate speech on Facebook and Instagram, to augmented reality features that will one day be available in the company’s future AR devices. On top of all that, Meta says that RSC will be used to create an experiment for metaviruses.
“RSC will help AI researchers at Meta create new and better AI models that can learn from trillions of examples, work in hundreds of different languages, easily analyze text, images and video, develop new augmented reality tools and much more.The company’s engineers wrote, Kevin Lee in a Chubu Senguptain the blog post in which they provide the news.
“We hope that RSC will help us build completely new artificial intelligence systems that, for example, will enable real-time voice translation of large groups of people, each speaking a different language, so that they can participate in a research project or play. AR game together.‘, they add.
How developed is the Meta supercomputer?
Work at RSC began a year and a half ago, when Meta engineers completely designed the various machine systems – cooling, power, connectors and wiring. The first phase of the RSC project is already in the works and consists of a 760 Nvidia GGX A100 system with 6080 graphics processors attached (a series of processors that are particularly good at solving machine learning problems). Meta says it already provides up to 20 times better performance on standard machine vision search tasks.
The second phase of the project will be completed before the end of 2022. At that time, the supercomputer will contain about 16,000 graphics processors and will be able to train artificial intelligence systems “with more than a trillion parameters on data sets as large as exabytes.” (This initial number of GPUs only provides a narrow measure of overall system performance, but for comparison, Microsoft’s artificial intelligence supercomputer, built by OpenAI Research Lab, consists of 10,000 GPUs.)