When Bill Dally joined NVIDIA’s Research Lab in 2009, only used for twelve people and focused on Ray tracking, rendering techniques used in computer graphics used in computer graphics.
A small research lab that can now use more than 400 people, who have helped fix nvidia from the GPU Game Startup at a $ 4 Town Comport freed boom intelligence intelligence.
Currently, the research lab research has a set of scenes for developing technology required for robotic power and AI. And some of these lab work has been shown in the product. The company runs Monday a New new model modelLibraries, and other infrastructure for the robotic developers.
Dally, now the head of Nvidia scientists, start the consultation for Nvidia in 2003 while they work on Stanford. As he was ready to independently from the department department stanford in the next few years, he planned to take the sabbatical. NVIDIA has another idea.
David Kirk, who runs the research lab at the time, and Nvidia CEO JENSEN Huang, thought to be more permanent position in the research lab is a better idea of the Dally tells the techcrunch pairing attached to the “complete court press” why he / she should participate in Nvidia research lab and eventually convinced.
“You can be the kind of suitable for interest and talent,” Dally said. “I think everyone who loves to find a place in life where he can make the biggest, you know, contribute to the world. And I think I’m, of course.”
When a Dally took on the Lab in 2009, the first and most important expansion. Researchers begin to work on the territory around the ray, including circuit and vlsi designs, or excessive scale integrated, a combined millions of transisions in one chip.
The research lab has not stopped since.
TechCrunch Events
San Francisco
|
October 27, 2025
“We try to find out what will make the most positive difference because we continue to see new areas, but some people know, but we have to see successful problems,” Dally said.
Briefly to build a better GPU for artificial intelligence. Nvidia Early AI boom and started tinkering with the idea ofi GPU in 2010 – more than a decade before AI Frenzy Now AI.
“We say this is incredible, this will change fully,” Dally said. “We have to start the dozen on this and Jensen believes that when I told him. We make the software to support, engage in the future researchers.”
Physical Ai focus
Currently, when NVIDIA holds the lead instructions on the AI GPU market, the technology company begins to pursue the desired region in the new area at AI data center. The search has caused Nvidia to AI and the physical robotics.
“I think the end of the robot will be a great player around the world and we want the base to make the brain all robots,” Dally said. “To do that we must start, you know, develop a major technology.”
Here is the Fidgeter’s Sanja, Vice President of AI ai ai in Nvidia, entered. Labor Research Fidler in 2018. At that time, he had been working with a simulated model with a student team in mit. When he told Huang about what was used in researcher researing researchers, she wanted to see.
“I can’t refuse to join,” FIDLER say of techcrroch in the interview. “This only, you know, it’s just a very good topic and at the same time. You know, Jensen told me, not for us, you know?”
He joined Nvidia and had to work for the research lab in Toronto called Omniverse, Nvidia platform, which focused to build simulation to Ai physical.
The first challenge of building the world that simulation finds the required 3D data, the FIDLER says. This includes finding the corresponding potential volumes of potential for use and build technology required to activate the image as a 3D Simulator Reorder.
“We invest in this technology called different rendering, which means to make whispered for AI, right?” Told FIDLER. “You go (from) mean 3D mean to pictures or video, ta? And we want another way.”
Model worldwide
Omniverse launched the first version of the model that was a picture of 3D 3D model, Ganverse3DIn 2021. Then work to find out the same process for the video. FIDLER says he uses the video from the robot and car driving itself to make the 3D model and simulation over Reconstruction Reconstruction Necklace ReconstructionThat’s the company first announced in 2022.
They add the technology as a bone of company Modend World Model Cosmos Family The announced in CES in January.
Now, the lab focuses to make the model faster. If you play video games or simulation you want to react to real, say FIDLER, for robots they make faster reaction time.
“The robot does not have to watch the world at the same time, in the same way of the world,” FIDler said. “Can watch like 100X faster. So, if we can make this model faster than now, they will be useful for the robography application or physical.”
The company continues to make progress on this purpose. NVIDIA announced ARMADA New Ai Models Designed to create synthetic data that can be used to train robots at the Siggraph’s graphic conference on Monday. Nvidia also announced the library and the new infrastructure software that is intended to the developer of robotics as well.
Although the progress – and hype is now about robots, especially humanoids – NVIDIA research team remains realistic.
Both and FIDLER says industries are still at least a few years because they have humanoids at home, with fibles compare with hype and time about autonomous vehicles.
“We make great progress and I know you know AI absolutely been on here,” Dally said. “Start visual AI for a robot understanding, then you know a generative, the valuables for tasks and ways to train network, the robot will grow.”