Earlier this week, NVIDIA held its GTC 2021 fall event, with Jen-Hsun Huang, CEO of NVIDIA, speaking at length over several new announcements, with AI being at the core of it all. Among the new features, one of them caught our interest.
That feature revolves around NVIDIA’s Omniverse and the new technologies it is introducing, Avatar and Replicator. In the case of Avatar, the technology provides Omniverse users to create their very own online avatar from scratch via a combination of the Megatron 530B language model, Merlin recommender framework, and the GPU brands Metropolis computer vision framework.
To show what could be achieved, Huang also showed off Project Tokkio; an avatar created in his own likeness, just smaller and a lot more animated. And before you ask: yes, there is ray-tracing being used on the avatar. Even more amazing is that, through accelerated computing and training via NVIDIA’s servers, Huang’s avatar is shown to be able to answer some pretty tough questions, albeit looking nervous enough while doing so unscripted, unlike most AI or digital assistants.
To be fair, NVIDIA putting an AI through a set of training models isn’t new, by any measure of the term. However, as its Omniverse matures and progresses over time, Project Tokkio is clearly just one of the latest technologies that the GPU maker hopes will help enrich its ecosystem.
(Source: NVIDIA via YouTube)