Meta is one of the many companies interested in the development of AI, and in recent years it has taken very important steps. Among many, he announced that he is in the process of manufacturing an “artificial intelligence supercomputer” that, once finished, would be the fastest of its kind. The brutal team would have several uses and would be key to the development of the metaverse that Mark Zuckerberg has promoted so much since Facebook changed its name.
The supercomputer in question is called the AI Research SuperCluster, but is more easily identified by the acronym RSC. According to Meta, its intention is to use the equipment to train machine learning systems; as well as allowing researchers to create artificial intelligence models “that can learn from billions of examples.”
In fact, from the company they assure that the equipment is already being used to train natural language processing models. But that’s just the beginning. One of the main uses that this artificial intelligence supercomputer would aim for would be to power moderation algorithms.
This could serve as a response to the persistent struggle of Facebook moderators, who not only denounce disastrous working conditions, but also suffer strong psychological damage from being continuously exposed to harmful content. The social network promised to implement artificial intelligence in 2020, but knowing that it was not precise enough to no longer depend on human intervention. If RSC is as good as Meta promises, it could become the answer to a historically contentious issue within Facebook.
RSC, Meta’s multipurpose artificial intelligence supercomputer
As we said at the beginning, Meta’s intentions in the development of this artificial intelligence supercomputer are varied. And the potential applications would not only aim to solve some ever-present problems, such as automatically moderating inappropriate content on Facebook, but would also allow us to step into new technological spaces. Among them, of course, two of the company’s most recent interests: augmented reality and the metaverse.
We hope RSC will help us build entirely new AI systems that can, for example, power real-time voice translations for large groups of people, each speaking a different language, so they can seamlessly collaborate on a research project or play an augmented reality game together. Ultimately, the work done with RSC will pave the way towards creating technologies for the next great computing platform: the metaverse, where AI-powered applications and products will play a major role.
Meta, about his artificial intelligence supercomputer
Creating a true beast
Meta explains that the design and manufacture of its new artificial intelligence supercomputer was done from scratch, over the past year and a half. But beyond the technical challenges that a project of this magnitude represents, there was also an extra complication: the COVID-19 pandemic. The first RSC drafts were done completely remotely and as a collaborative document between team members. This is how the first parameters were outlined until they could get down to work in person.
Yet another hurdle to overcome was the global shortage of chips. Mark Zuckerberg’s firm, like other large technology companies, had to face the lack of key components; but that was not all, since he even had problems with the provision of construction materials. “We had to establish new rules around our data center designs, including cooling, power, rack layout, cabling and networking,” they said.
As reported from Meta, the first stage of the artificial intelligence supercomputer has concluded. The team currently has 760 NVIDIA DGX A100 systems being used as compute nodes, with 6,080 graphics cards. However, the company is confident of raising the number of GPUs to 16,000 when RSC phase 2 is complete; the equipment is expected to be completed this year.
