Quick take:
- Meta Platforms has unleashed a powerful machine that will be the fastest AI supercomputer when completed in mid-2022.
- The company said its supercomputer is already the fifth-fastest AI supercomputer in the world.
- The AI Research SuperCluster (RSC) is a supercomputer that will perform the hardest tasks and help set a platform for the metaverse.
Facebook’s parent company Meta Platforms has revealed its gigantic AI supercomputer that will be the fastest in the world when completed in mid-2022. The machine called AI Research SuperCluster (RSC) is already the fifth-fastest AI supercomputer in the world.
It is built to perform the hardest tasks whilst also helping set a platform for the metaverse. RSC has 6,080 graphics processing units packaged into 760 Nvidia A100 modules. Meta CEO Mark Zuckerberg said it is the fastest machine built for AI tasks. Meta is planning to boost the performance by a factor of 2.5, expanding to 16,000 GPUs this year.
The company wants to use the machine on several research projects including Multimodal, an AI that derives conclusions from a combination of imagery, sound, and actions, instead of relying on a single type of input data.
CNET wrote that by combining several data points, Meta could help Facebook deal with one of its most niggling problems, the identification of harmful content. The company is also looking to use the RSC machine to build the metaverse, an immersive 3D universe that has gained explosive popularity.
Experts think the amount of power in the RSC machine could be enough to simultaneously translate speech for a large group of individuals speaking different languages.
Commenting on the RSC reveal, Zuckerberg said the metaverse requires a huge amount of computing power, adding that “RSC will enable new AI models that can learn from trillions of examples, understand hundreds of languages and more.”
Researchers at Meta are already using the RSC to train large models in natural language processing, helping to build new and better AI models that can learn from trillions of examples, create regimented reality tools, whilst also working across hundreds of different languages.
Meta’s Technical Program Manager, Kevin Lee and Software Engineer, Shubho Sengupta said RSC is about 20 times faster than its earlier 2017-era Nvidia machine. “For decoding human speech, it’s about three times faster,” they wrote in a blog on Monday.
The company also thinks its RSC machine could help AI models improve on the challenge of self-learning by trining the AI models with large data sets.
Stay up to date:
