As the tech giant looks to pioneer the creation of a virtual space like no other, the company will need the supporting computing power and systems to facilitate simultaneous connection in immersive digital worlds.
Enter the development of a new AI Research SuperCluster that is destined to become the fastest AI supercomputer in the world.
What is an AI Research SuperCluster (RSC)?
AI supercomputers are engineered through the combination of multiple GPUs into compute nodes which are then connected by a high performance network fabric that allows extra fast communication between those GPUs. RSC comprises of a total of 760 NVIDIA DGX A100 systems as its compute nodes, for a total of 6,080 GPUs.
Each DGX communicates via a NVIDIA Quantum 1600Gb/s InfinitiBand two-level Clos fabric that has no oversubscription.
RSC will be what helps Meta’s AI team build new and improved AI models that are able to learn from trillions of examples; work across hundreds of different languages; develop new AR tools and seamlessly analyse text, video and images together.
An example of the AI systems that Meta hope RSC will help them build are systems that can power real-time voice translations to large groups of people, each speaking a different language, so they can seamlessly collaborate on a research project or play an AR game together.
In a nutshell, the work done with RSC will revolutionise what we know about building technologies for the next major computing platform known as the Metaverse.
Accordng to early benchmarks on RSC, in comparison to that of Meta’s legacy production and research infrastructure, RSC:
- Runs computer vision workflows up to 20 times faster
- Runs NVIDIA Collective Communication Library more than 9 times faster
- Trains large-scale NLP models 3 times faster
This means that with RSC, a model with tens of billions of parameters can finish training in just 3 weeks; 6 weeks faster than before.
How Long Will the Metaverse Take to Develop?
Meta specifically notes that the Metaverse will take years to develop. The tech giant also goes into detail about how any company that pitches itself as ‘Metaverse ready’ is not being honest due to the massive collaboration between platforms required and need for superior computing power.
Eventually, Meta’s RSC will be provide the tech giant with significant advantages in the development of new systems to engineer entirely new user experiences that will change the way people interact and experience digital. Although this will take some time, it will be interesting to observe Meta’s progress and evolution.
