Nvidia’s new computer gives AI brains to robots

Nvidia CEO Jensen Huang sees a future where billions of robots serve humans, bringing in trillions of dollars in revenue for the company.

To meet that goal, Nvidia on Monday unveiled a new computing device that will go into high-performing robots that could then try to replicate human behavior.

The Jetson Thor robotics computers are capable of AI intelligence. The devices will help robots hear, detect patterns, make decisions and act. Many of the AI capabilities arise from generative AI (genAI) technologies available today.

“Thor…is the ultimate platform for physical AI and robotics built for the age of reasoning AI running at the edge — low power, high performance, real time,” Deepu Talla, vice president of robotics and edge AI at Nvidia, said during a news briefing.

The boards have Nvidia’s flagship Blackwell graphics processor, which power the latest genAI models. Users currently have to access those GPUs in the cloud, which is why the robotic implementation of Blackwell in Thor is different. Robots make real-time decisions, and the GPUs can run AI on the device without cloud access or internet connections.

“…Typically, your actuation is in fractions of a second…,” Talla said. “If you go to the cloud, typically the latency is hundreds of milliseconds…. That’s why you want to do as much as possible at the edge.”

The Thor developer board delivers roughly 2,070 teraflops of low-precision FP4 data performance; that pales in comparison to Blackwell datacenter GPUs, which deliver 20 petaflops.  While 10 times faster than the robotics GPUs, they also consume significantly more power.

The robotics computers draw between 40 and 130 watts of power. Given that robots will operate on batteries, that adds up to two times the power draw of standard work laptops, which draw between 30 watts to 60 watts of power.

Robots with Thor are expected to be used in heavy industries with physical labor such as manufacturing, oil and gas, and mining. “Our vision for robotics is not just humanoid,” Talla said. “We think general purpose robotics is extremely important so that they reason out about different tasks that they encounter in the real world.”

Nvidia has recently made a variety of announcements involving AI robots. Earlier this month, the company announced Cosmos Reason, a vision AI model that helps robots make better decisions by evaluating their surroundings. The model can process pixels in video input to process information.

Thor is the hardware where these AI enhancements will come to life, Talla said. “Because of reasoning we are able to tackle any model — adding reasoning to it allows it to generalize even more, which enables us to solve problems that [were] previously unsolvable,” Talla said.

The $3,499 AGX Thor computer system has 14 ARM CPU cores, 128GB of RAM, and Gigabit Ethernet ports. Nvidia also has the T5000 and T4000 production boards on which developers can build robots. 

The T5000 board draws 130 watts of power, while the T4000 draws 70 watts, making it suitable for robots with smaller batteries.

More Nvidia news and insights:

Nvidia turns to software to speed up its data center networking hardware for AI

Nvidia: ‘Graphics 3.0’ will drive physical AI productivity

Nvidia launches Blackwell-powered RTX Pro GPUs for compact AI workstations

Nvidia’s new genAI model helps robots think like humans

Nvidia patches critical Triton server bugs that threaten AI model securityNvidia’s new computer gives AI brains to robots – ComputerworldRead More