Don’t ever refer to the metaverse as marketing hype or just a product of science fiction.
Nvidia CEO Jensen Huang quickly dispelled any such notion in a call with reporters on Wednesday.
“Metaverse is not a hype,” Huang said in response to a question. During his keynote address on Tuesday at Nvidia’s GTC event and with reporters, he gave example upon example of how virtual worlds “are indistinguishable from the real world” even to the machines and humans operating within them. Sensors inside of digital twins—everything from lidars to radars to cameras—are trained to obey laws of physics in keeping with this definition of reality.
One example of a virtual world Nvidia presented at GTC is used at Siemens Energy, which is deploying the Nvidia Triton Inference Server for power plant inspections and autonomous management of operations. Part of the Siemens operation relies on AI to handle the massive input of data from millions of sensors and onsite cameras at power plants. All told, Huang said 25,000 companies rely on various Nvidia AI products.
Huang also explained Nvidia’s ambitious vision to create a digital twin of Earth relying on the vast majority of data coming from satellites. Nvidia is developing its own Earth twin on its own supercomputers at its own cost. Huang said he has hasn’t decided, yet, what government or entity will benefit from it.
But there also may be many other Earth digital twins. “There will be thousands of Earth digital twins and each will study and monitor one aspect of Earth,” he said. “We’re interested in climate science.”
Currently, the work is to map the entire globe down to 10 meter-sized pieces, which is a big effort with mapping capabilities now at 10 kilometers. That mapping will include low clouds, so the data will need to measure volumes of air, not just the surface. “The amount of computation we need is billion times more” than what is now available, he said. “We now have the breakthrough of physics-informed ML.”
In his keynote address, Huang further laid out the value of a digital Earth twin for climate science. “We can create a digital twin of the Earth that runs continuously to predict the future, calibrating and improving its predictions with observed data-- and predict again,” he said. “Hopefully in a couple of years, data will stream into a digital twin of Earth running in Omniverse. An ensemble of physics ML models will predict the climate.” (Huang’s entire GTC keynote is available online.)
Among its many announcements at GTC, Nvidia included Nvidia Modulus, a physics ML neural simulation framework for developing trained ML physics models for use in digital twins like the Earth twin.
In the fast-developing electric vehicle and autonomous vehicle space, Huang told reporters that Nvidia Omniverse will help companies build entire digital twins of fleets of vehicles used for delivery. Or the digital twins could be used to help vehicle OEMs create a view of their entire fleet or a particular car brand.
One theme throughout GTC was Nvidia’s continuing emergence into a company offering both hardware based on chip acceleration alongside software used by 3 million developers in various industries. Huang noted that Nvidia now supports 150 software development kits in this “full stack” of capabilities.
“The future of computing is a full stack challenge and you can’t solve challenges without domain, algorithms, computer systems and, of course, the chips under that,” Huang said.