How a once-tiny research lab helped Nvidia become a $4 trillion-dollar company

When Bill Dali joined the Nvidia Research Lab in the 21st, it was appointed about a dozen people and concentrated on a rendering technique used in computer graphics.

This once small research lab has now appointed more than 5 people, who helped convert NVDia to NVDia from a video game in the nineties to fuel artificial intelligence boom in $ 4 trillion.

Now, the company’s research lab has a sights for the development of robotics and the technology needed to get AI. And some of those lab works are already being displayed on products. The company unveiled Monday New Set World AI ModelLibrary and other infrastructure for robotics developers.

Dali, now the chief scientist of Nvidia, started consulting Nvidia while working at Stanford on the 21st. When he was ready to resign from the Department of Computer Science Department of Stanford a few years later, he planned to take subtle. Nvidia had a different idea.

At that time, David Kirk and Nvidia CEO Jensen Huang thought that the research lab had a better idea of a better position in the research lab. Dali TechCrunch had a “full court press” on why the pair had to join the NVEDIA Research Lab and finally confirmed him.

“It’s been a perfect fit for my interest and my talents,” said Dali. ” “I think everyone is always searching for the place where they can make the biggest, you know, contribute to the world. And I think to me, it’s definitely nvidia.”

When Dali accepted the lab in the 21st, the expansion was the first and foremost. Researchers started working in the outside regions of Ray Tracing now with circuit design and VLSI or very large -sized integration, a process that combines several million transistors in a single chip.

The research lab has never stopped expanding.

“We try to understand what the most positive differences will make for the organization because we are constantly seeing new zones, but some of them, you know, they do great if we have trouble telling us [we’ll be] It is wildly successful in it, ”said Dali.

For a while it was making better GPU for artificial intelligence. Nvidia was early in the future AI Boom and began to shake with the concept of AI GPU in the 21st – more than a decade before the current AI madness.

“We said it was amazing, it could completely change the world,” said Dali. “We need to start twice about it and Jensen believed that when I told him. We started specializing our GPUs for it and started developing a lot of software to support it, it was clearly relevant to researchers around the world who were doing it.”

Physical AI focus

Now, because of the commanding leadership of the NVDia AI GPU market, the technology company has begun to look for new fields of demand beyond AI data centers. This search has taken Nvidia to Physical AI and Robotics.

“I think the robots are finally going to be a huge player in the world and we basically want to create all the robots brains,” said Dali. “We need to start it, you know, to develop the original technologies.”

It is there that Sanza Fidler, Vice President of AI Research in Nvidia has arrived. Fidler joined the Nvidia Research Lab in 2018. At that time he was already working on simulation models for robots with a team of students in MIT. He was interested when he informed Huang about what was working at a researchers’ reception.

“I couldn’t resist against joining,” Fidler told TechCrunch in an interview. “That’s exactly as if you know, it’s just such a big deal fit and at the same time so great culture fit you you know, Jensen told me, work with me, not for us, not for us, you know?”

He joined Nvidia and began creating a research lab called Nvidia Platform in Toronto, which concentrated on creating simulations for physical AI.

The first challenge to create this simulated Worlds was to look for 3D data, feedler. These may find the appropriate volume of potential images to use the technology needed to convert these images into 3D presenters and use simulators.

“We have invested in this technology called differential rendering, which basically makes rendering rendered to AI, isn’t it?” Fidler Dr. “You go [from] Rendering means 3D to figure or video, isn’t it? And we want it to go the other way. “

World model

Omonvers have published the first version of its model that converts images into 3D models, Ganvers3dIn 2021. Then it can work to determine the same process for the video. Fidler said they used videos of robots and self-driving cars to create these 3D models and simulations Neurral neural reconstruction engineWhich the company first announced in 2022.

He added these technologies were the backbone of the company Cosmos family of world AI models That was announced at CES in January.

Now, the lab is focusing on to speed up these models. When you play a video game or simulation, you were able to respond to the technology in real time, feedler said they were working to faster the response time for robots.

Fidler said, “Robot does not need to see the world at the same time, as the world works,” said feedler. “It may seem like a quick 100x so so if we can make this model significantly faster than today, they are going to be extremely effective for robotic or physical AI applications.”

The company has continued to progress in this goal. Nvidia has declared a fleet New World AI Model Robots are designed to create synthetic data that can be used to train robots at the Siggraph Computer Graphics Conference on Monday. New libraries and infrastructure software have also been announced at the Nvidia Robotics developers.

Despite the progress – and the current hype about the robots, especially the Humanoids – the NVIDIA research team remains realistic.

Both Dali and Feedler said that the industry has at least a few years of vacation from having humanoids in your home, comparing the hype and timeline related to feedler autonomous vehicles.

“We are making huge progress and I think you know that AI was truly capable of here,” Dali said. “Starting with Visual AI for robot realization, and then you know the generator AI, it is extremely valuable for task and speed plans and manipulation.