Building the biggest and most powerful computer on the planet is now a truly global competition. By the way, Meta (AKA Facebook) is about to enter this market and it is likely to be one of the 10 fastest supercomputers in the world…
Meta’s RSC for AI Research Supercluster
“Meta has developed what we believe to be the world’s fastest AI supercomputer. We call this RSC for AI Research Supercluster And it will be completed by the end of the year.”
Mark Zuckerberg, who broke the news from Facebook’s former official blog, reaffirmed the social media giant’s commitment to artificial intelligence. The supercomputer will enable META’s artificial intelligence researchers to build new AI models:
- develop new augmented reality tools;
- Work in hundreds of different languages
- learn from trillions of examples;
- “Analyze all text, images and videos.
“The work done with RSC will pave the way for building technologies for the next great computing platform: the Metaverse, where AI-based applications and products will play a key role,” says Meta.
RSC will pave the way for Meta’s next major computing platform, the Metaverse, in which AI-based applications and products will play a key role.
Large AI models, of which OpenAI’s GPT-3 is probably the most famous, are not assembled on laptops and desktops; They are the end product of weeks and months of computation backed by high-performance computing systems that eclipse even the most advanced gaming platforms.
The team that founded RSC, Pure Storage, is proud to be successful almost entirely remotely.
It announced that META had selected the company as a storage partner to provide scalable storage capacity to power its new AI Research Supercluster (RSC), an AI supercomputer. Using pure storage solutions such as FlashArray and FlashBlade, RSC will be able to analyze both structured and unstructured data.
Work on the RSC began a year and a half ago and then in full imprisonment. Phase 1, which includes 760 Nvidia GGX A100 systems (universal systems dedicated to AI workloads) with 6,080 GPUs (graphics processing units or graphics processors), is already underway. The use of GPUs is particularly well suited for artificial intelligence applications. The system thus provides 20 times better performance for its traditional computer vision research efforts.
By the end of 2022, Phase 2 of the RSC will be completed. At this stage :
- It will be able to run on around 16,000 GPUs;
- It will be able to train artificial intelligence systems “on data sets with over a trillion parameters as large as an exabyte”. Specific benchmark tests would be required to assess the performance.
Meta said the supercomputer would be the fastest in AI, which changes the method of measurement. Zuckerberg’s real challenge appears to be to train algorithms to moderate content in both the metaverse of the future and the classic social network. All this out of respect for privacy.
Is this important? And if so, to whom? If any, it could be important for the top 500, so I asked if he had any comments on it. But that doesn’t change the fact that the RSC will be one of the fastest computers in the world, perhaps the fastest to be operated by a private company for its own purposes.