Meta, the parent umbrella of Facebook, has announced that it is building what it believes will soon be the world’s fastest supercomputer. Announcing the new feat, Meta chief Mark Zuckerberg stated that they’re calling it AI Research SuperCluster, which stands for RSC. It’s still being built, and once finished by mid-2022, Meta believes that it will help the company build the next generation of AI products that can help you feel safer on its platforms, and — unsurprisingly — build its metaverse.
In a blog post announcing the AI RSC supercomputer, the company said, “RSC will help Meta’s AI researchers build better AI models that can learn from trillions of examples; work across hundreds of different languages; seamlessly analyse text, images and video together; develop new augmented reality tools and more. Ultimately, the work done with RSC will pave the way toward building technologies for the next major computing platform — the metaverse, where AI-driven applications and products will play an important role.”
With this move, Meta becomes the latest company to join a growing list of big technology corporations with claims of having built among the world’s fastest supercomputers. At its Build 2020 conference, Microsoft announced its Azure-hosted supercomputer built collaboratively with OpenAI. The latter, at the time, was reportedly among the five most powerful and publicly disclosed supercomputers in the world, and was destined to be used by Microsoft in training extremely large artificial intelligence and machine learning models.
A few months later, in August 2020, Google announced that it broke the fastest performance mark on Nvidia’s MLPerf benchmarking platform, which is an industry-accepted standard denoting the ability of a machine learning training rig to perform at its peak. The supercomputer at Google used 4,096 tensor processing units (TPUs) to hit the reported performance level. TPUs are custom circuits developed by Google specifically for machine learning workloads.
At the moment, it’s not entirely clear what performance levels can the Meta AI RSC supercomputer reach. One of its key use cases, according to Meta, will be to analyse text, images and other content faster than what the company’s systems can do today. “This research will not only help keep people safe on our services today but also in the future, as we build for the metaverse,” the company statement said.
The crop of supercomputers mentioned above fall under classical computing, which is expected to be succeeded in the coming years with quantum computing. IBM, one of the biggest movers so far in the quantum computing space, said at the unveiling of its 127-qubit ‘Eagle’ quantum processor that it expects to soon go beyond the performance levels that classical supercomputers — akin to Meta’s AI RSC rig — can reach.