Automotive

Nvidia's Drive PX 2 is the world's first in-car AI supercomputer

Nvidia's Drive PX 2 is the wor...
The new computer uses deep learning for advanced object detection and path planning
The new computer uses deep learning for advanced object detection and path planning
View 3 Images
The new computer uses deep learning for advanced object detection and path planning
1/3
The new computer uses deep learning for advanced object detection and path planning
The new in-car computer features two next-gen Tegra CPUs and two next-gen GPUs
2/3
The new in-car computer features two next-gen Tegra CPUs and two next-gen GPUs
The capable-looking computer will put put through its paces by Volvo, who will use it in a fleet of 100 XC90 SUVs next year
3/3
The capable-looking computer will put put through its paces by Volvo, who will use it in a fleet of 100 XC90 SUVs next year

Nvidia has announced the successor tolast year's Drive PX in-car computer. The new model, known as theDrive PX 2, is liquid cooled and seriously powerful, sporting 12 CPUcores and 8 teraflops of processing power. It's designed to allow vehicles to accurately sense their surroundings and navigate autonomously.

The computer is designed to provideautonomous vehicles with the latest Nvidia GPU tech, allowing for360-degree situational awareness. According to the company, in termsof raw power, it's roughly equivalent to 150 MacBook Pros. Nvidia believes that the level of power provided by the new computerwill be necessary if self-driving cars are ever to hit themainstream.

With autonomous vehicles using an array of sensors, the Drive PX 2 is naturally able to processinformation from a wide range of sources. It can handle inputs from12 video cameras, ultrasonic sensors, and both radar and LiDAR. Allthat data is combined and analyzed by Nvidia's DriveWorks softwaretool set, allowing the vehicle to picture the world around it,determine its position, pinpoint obstacles, and plot the safestpossible route.

It's made up of two Tegra CPUs pairedwith two discrete GPUs based on the company's Pascal architecture.That hardware provides 8 teraflops of processing power, allowing forup to 24 trillion deep learning operations per second.

According to Nvidia, the deep learning system is better ataddressing certain issues than traditional computer visiontechniques. In particular, it's more capable of identifying anddealing with difficult lighting scenarios, such as sunrise andsunset, and adverse weather conditions like snow or heavyrain.

The capable-looking computer will be put through its paces by Volvo, who will use it in a fleet of 100XC90 SUVs. Those vehicles will hit the road next year as part of thecompany's Drive Me autonomous car pilot program. It's a big part ofVolvo's long term plan, with the company believing that such systemswill play a significant role in road safety.

"Our vision is that no-one should bekilled or seriously injured in a new Volvo by the year 2020," saidVolvo program director Marcus Rothoff. "Nvidia's high-performanceand responsive automotive platform is an important step towards ourvision and perfect for our autonomous drive program and the Drive Meproject."

Source: Nvidia

2 comments
Stephen N Russell
Lisc for other auto makers to use for models,
SamB
"allowing for up to 24 trillion deep learning operations per second." The battle for the straightforward communication of ideas is surely being lost to the marketers.