Nvidia Omniverse Poised to Power Avatars Joining Physical, Virtual Worlds

Nvidia Omniverse Enterprise, a new AI-based platform connecting physical and virtual worlds with 3-D digital twins, is now available.

Jensen Huang, Nvidia’s CEO, emphasized on Wednesday the near-term opportunity for the company and the growing Omniverse ecosystem.

“Omniverse has real, immediate applications,” Huang told financial analysts during Nvidia’s quarterly earnings call. As Nvidia raised its forecast for the current quarter, analysts wanted to know more about the potential for Omniverse.

Huang demonstrated an array of new capabilities during last week’s Nvidia GTC event. Among them is Omniverse Avatar, a platform that lets developers create digital twins as conversational AI-based assistants. Nvidia also introduced Omniverse Replicator, a workflow engine for producing physically simulated synthetic data models for training deep neural networks.

Nvidia Omniverse Enterprise comes in a variety of subscription options. It provides virtual reality environments that allow global 3-D design teams to collaborate in real-time using different vendors’ software.

For example, Sony Pictures Animation artists in different locations tested Omniverse to collaborate on creating specific scenes. Other early evaluators include architecture, engineering and consulting firm CannonDesign; product visualization provider Epigraph; and Ericsson.

“In the world of remote working, Omniverse’s collaboration capabilities are going to be really appreciated, and that should happen right away,” Huang said. “We would like to see that that happen in very near term. And that drives, of course, more PC sales, more GPU sales, more workstation sales [and] more server sales.”

Omniverse Avatar

The demonstration of Omniverse Avatar at Nvidia GTC showed the potential for 3-D virtual assistants in mainstream collaboration and videoconferencing. The simulation and collaboration platform for 3-D workflows is now available as a technology preview.

Nvidia said Omniverse Avatar brings together speech AI, computer vision, natural language understanding, recommendation engines, facial animation and graphics rendering. It uses Nvidia’s Riva speech recognition SDK, which understands multiple languages. The natural language understanding is based on Nvidia’s Megatron 530B large language modeling and training platform. And the recommendation engine is based on the Nvidia Merlin framework.

Omniverse Avatar includes a reference architecture called Project Tokkio. In Huang’s demo, he showed a 3-D digital replica of himself conversing with scientists about climate change.

Another demonstration of Project Tokkio showed an avatar appearing in a kiosk at a quick service restaurant conversing with a customer.

“The ability for us to connect all of this amazing technology together on a server and have it respond in about two seconds is really quite remarkable,” Huang said. “You can have a quite an interactive conversation with it.”

Omniverse Avatar includes Project Maxine, an SDK accelerated by GPUs to let developers enable avatars in video conferences and livestreams. Daniel Newman, principal analyst at Futurum Research, said Nvidia demonstrated that it sees viable use cases for so-called “metaverse” environments.

Futurum Research’s Daniel Newman

“Nvidia may end up being more important to the future of whatever these digital and physical world creations are than Facebook-Meta, but not necessarily will they get the credit for it,” Newman said during a podcast.

Synthetic Replication Engine

Huang demonstrated two implementations of Omniverse Replicator, the new engine that creates simulated synthetic data to train deep neural networks. The first, Nvidia Drive Sim, is a simulation tool for training neural networks for the perception systems for autonomous vehicles. The other, Isaac Sim, generates synthetic training data for robots.

Omniverse Replicator automates the tedious and costly process of generating data for AV and robotics developers. Nvidia claims it can provide more complete and accurate data that is not practical or possible for individuals to create.

Beating Forecasts

While GTC showcased what lies ahead for Nvidia and its partners, analysts are paying attention to its most recent performance. Nvidia’s stock price this year is up more than 125%.

For the quarter that ended Oct. 31, revenues of $7.1 billion beat the forecast $6.82 billion. Meantime, Nvidia raised its revenue projection for the current quarter from $6.86 billion to $7.4 billion. Fueling the company’s growth were revenues from its data center GPUs. Data center sales of $2.9 billion increased 55% year-over-year, driven by demand for its processors needed to expand cloud-based AI capability.