The Facebook group Meta has put a new supercomputer for artificial intelligence (AI) applications into operation.
The "AI Research Super-Cluster" will be the fastest AI computer in the world when it is fully developed in the middle of the year, the company announced on Monday.
Meta's researchers are already using computers to make digital systems understand what's happening and what's being said in videos.
In the future, the technology could also be used, for example, for simultaneous translations in video conferences or games.
The new applications are based on AI models that learn from examples in order to independently understand languages, for example. Models on the RSC are designed to learn from trillions of examples and work with hundreds of different languages. In this way, text, images and videos should be able to be analyzed together seamlessly. When it comes to the goal set by company founder and CEO Mark Zuckerberg to align the company to a more comprehensive virtual world (“metaverse”), new augmented reality tools should also be created. The computer is intended to make a significant contribution to making the concept of a metaverse a reality for the company.
AI supercomputers are built by combining multiple GPUs into compute nodes, which are then linked together via a high-performance network.
According to Meta, the RSC currently consists of a total of 760 NVIDIA DGX A100 systems as computing nodes.
That's more than 6000 graphics processing units (GPUs) in total.
The storage tier has a capacity of 175 petabytes - that's 175 million gigabytes.
The computer is ready for use, but is to be expanded further.
By the middle of the year, the number of GPUs is expected to grow to 16,000, which is said to increase AI training performance by more than 2.5 times.