In 2020, Meta (Previously Facebook) envisioned the world’s largest, fastest, and super webbed next-gen AI supercomputer. This Year, they have launched AI Research SuperCluster (RSC) aimed to be fully built by mid-2022. As per the latest post by Meta, this AI powdered supercomputer will be able to operate quintillions of operations per second.
In the recent blog post by Shubho Sengupta and Ken Lee working as AI Research and Technical Program Manager at Meta respectively, posted
“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 analyze 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.”By Meta Research Team
The RSC will help the AI team to build the new neural networks. It can perform real-time voice translations to a broader number of people speaking in different languages. With this level of complexity, it will be able to benefit various domains like speech, vision, language along with trillions of other parameters. Also, the AI research team is determined to eliminate detrimental contents which can harm their AI development.
Meta SuperComputer Specificities
With the completion of this project, RSC will have 16,000 GPUs, having a capacity of 1 exabyte (1 billion bytes). This will be able to provide 16 terabytes per second of data to the system. These new GPUs are to be from Nvidia’s latest core, the A100 which has passed all the higher expectations of AI systems. It shows a 20-fold speedup in computer vision tasks and a 3-fold speedup in natural language processing.
The predecessor of this model contained 22,000 graphic processing units (GPUs) of Nvidia’s V100 Tensor Core. Despite being designed for playing videos games, it was also a highly efficient tool for training artificial intelligence models.
Extreme care has been taken to make sure that the RSC is more reliable, secure, and more anonymous. The AI Research Team will ensure that there is zero chance of data leakage.