Blog

Meta claims to bring out the fastest AI Supercomputer

On Monday, Meta introduced latest Artificial intelligence (AI) supercomputer, the AI Research SuperCluster (RSC), which it says will be the world’s fastest AI supercomputer once fully built up in mid-2022. Artificial intelligence (AI) has been central to tech giants, including Meta.

In currently, Artificial intelligence can perform tasks like translating text between languages and helping identify potentially harmful content, but developing for the next generation of Artificial intelligence will require powerful supercomputers capable for quintillions of operations per second. RSC will help Meta’s Artificial intelligence researchers build better Artificial intelligence models that can learn from trillions of instance; work across hundreds of different languages such as seamlessly analyse text, images and video together; develop new augmented reality tools and much more.

Supply chain blues

The ongoing chip supply shortage has affected for the countless infrastructure projects, and the Research SuperCluster (RSC) was no exception. Research SuperCluster began to completely remote project that the team took from a simple shared document to a functioning cluster in about a year and a half,” said Lee and Sengupta.

COVID-19 or industry-wide wafer supply constraints also brought supply chain issues that made for difficult to get everything from chips to components such a optics and GPUs, and even construction materials all of which had to be transported in accordance with new safety protocols. Bullish on the metaverse, the supercomputer’s work is meant to pave a way for experiences in the metaverse. Ultimately, the work done with Research SuperCluster 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.

Research SuperCluster to play a vital role in Metaverse

In order to understand the full benefits of self-supervised learning or transformer-based models, it requires training increasingly large, complex, and adaptable models. Speech recognition has to work effectively even in challenging scenarios that come with a lot of background noise. NLP has to understand more languages or dialects.

Meta said that the Research SuperCluster can train models that use for the multimodal signals to determine whether an action, sound and image are harmful or benign more quickly. It added that when Research SuperCluster moves to the next phase, it will get even bigger with enhanced capabilities as the groundwork for the metaverse is built. Meta’s researchers have already started by using Research SuperCluster for training large models in NLP and computer vision. 

Privacy and security

Meta said that Research SuperCluster has been built keeping privacy and security as prime focus areas. 

• Research SuperCluster is isolated from the larger internet. It has no direct inbound and outbound connections with traffic flowing only from Meta’s production data centres.
• Before the data is imported to Research SuperCluster, it goes through the privacy review process to confirm it has been correctly anonymised. After that, it’s encrypted before it finds its usage in training AI models. The decryption keys are deleted regularly so that older data isn’t still accessible.
• The entire data path from the storage systems to the GPUs is end-to-end encrypted. It has the necessary tools or processes to verify that these requirements are met for every time, Meta claims

Research infrastructure from NVIDIA

Meta has collaborated with the NVIDIA to build the AI Research Supercomputer. It uses 760 NVIDIA DGX A100 systems as its compute nodes. It comes with the 6,080 NVIDIA A100 GPUs linked on the NVIDIA Quantum 200Gb/s InfiniBand network to give 1,895 petaflops of TF32 performance. Penguin Computing is the NVIDIA Partner Network delivery partner for Research SuperCluster (RSC).

Penguin also provided for managed services and AI-optimized infrastructure for Meta consisting of 46 petabytes of cache storage with its Altus systems. Pure Storage FlashBlade and FlashArray//C provide for the scalable all-flash storage capabilities needed to boost the Research SuperCluster.

For the second time NVIDIA has been the chosen partner of Meta as its base to provide research infrastructure. Since 2017, Meta had built the first generation of infrastructure for AI research with 22,000 NVIDIA V100 Tensor Core GPUs. It had the capabilities of handling 35,000 AI training jobs in these days.

As soon as benchmarks of Meta have shown that Research SuperCluster can train large NLP models three times faster and run computer vision jobs twenty times faster than the previous system. Previously, the second phase, Research SuperCluster will expand to 16,000 GPUs. Meta thinks it will deliver five exaflops of mixed precision of AI performance.