NVIDIA

Drive Thor, the company's new AI chip for self-driving cars that will be mass produced in 2024

Nvidia has unveiled a new chip called Drive Thor that is scheduled to go into mass production in 2024 and is set to revolutionize self-driving cars (pictured above).

According to Danny Shapiro, vice president of Nvidia, Thor processors will start in vehicles hitting the roads in 2025, starting with the 001 EV from Chinese automaker Zeekr. They are based on the new Hopper graphics architecture and have a lot of AI features, which is critical for self-driving cars. “These processors are capable of achieving full autonomy,” Shapiro said, referring to level 4 or 5 autonomy, in which cars can be driven without human attention or even human presence.

Nvidia had planned to release a processor called Atlan by 2024, but canceled it in favor of Thor, whose AI software runs at 2 quadrillion operations per second — twice as fast as Atlan and eight times as fast as the existing Orin processor. Thor supports one of Hopper's key features: the ability to speed up the work of AI algorithms using special transformation engines. Nvidia also plans to release cheaper versions of the Thor processor for less demanding driving tasks like lane control and automatic emergency braking.

The automotive processor market is constantly growing as automakers demand ever newer and more powerful processors and other semiconductor chips for driver assistance systems, infotainment systems and electronic control units that control everything from engine to GPS navigation. For example, the Porsche Taycan contains over 8,000 different semiconductor elements.

The growing market brings fabulous profits for the development companies. Nvidia has pre-orders for automotive processors and components worth $11 billion, while its main competitor, Qualcomm, has more than $19 billion.

With 77 billion transistors, Thor will be one of the most complex processors on the market. In addition to using Hopper GPUs, it borrows CPU cores from the NVIDIA Grace processor for general computing tasks. It also uses technologies from the latest gaming GPU technology, the Ada Lovelace architecture. All of this, according to Nvidia, will allow automakers to replace small, more expensive, power-hungry chips with a powerful system-on-a-chip and will make the process of updating automotive software over the air much easier. Tesla has had a great technological advantage for many years precisely because of its use of this technology.

In addition, the Thor processor is planned to be used in robots and medical equipment. It will be able to simultaneously run three operating systems – Linux, QNX and Android – for various parts of the vehicle's computing environment. This processor-splitting technology ensures that lower priority tasks, such as operating the infotainment system, do not conflict with security-related priority tasks.

Fully autonomous cars have been promised for years but are still in the testing phase. The new Nvidia processors should accelerate the transition from testing to mainstream use.

“Technically, this turned out to be a much more difficult task than initially thought,” Shapiro said of self-driving cars. “And because safety is paramount, no one is ready for widespread use of such vehicles until full confidence in the reliable operation of computer control systems is achieved.”

Mundoquatrorodas