Spurned by Tesla, NVIDIA’s new Orin self-driving processor ups the game by 7x

NVIDIA has unveiled what they call the “world’s most advanced processor” for use in autonomous vehicles and robots. The new NVIDIA DRIVE AGX Orin chip can perform 200 trillion operations per second, which is almost seven times as many as NVIDIA’s previous Xavier chip (30 trillion operations) and more than Tesla’s FSD Computer (144 trillion).

NVIDIA has been focusing on self-driving hardware lately, seeing it as a big area for potential future growth in their business. Self-driving cars need tremendous amounts of data processing power to take in images of the world around them and properly process what they’re seeing, as quickly as possible, so they can act as potentially safer drivers than humans.

NVIDIA’s Xavier chip was used in a multi-processor configuration and paired with GPUs in their DRIVE PX Pegasus self-driving computer, which NVIDIA claimed at the time would be able to offer level 5 autonomous driving.

The Orin chip will be capable of scaling from level 2 to level 5 autonomous driving. It will be available to automakers for the 2022 model year. There will be a range of configurations available for manufacturers to choose from.

NVIDIA is also offering open-source access to its AI models. This includes systems involving traffic light and pedestrian recognition, path perception, and gaze detection.

Previously, Tesla used NVIDIA’s DRIVE PX 2, the predecessor to the PX Pegasus, for its Autopilot AP 2 and 2.5 systems. In April of this year, Tesla moved to start using its own HW3 FSD computer. Tesla still uses NVIDIA hardware in some data centers for image processing.

Tesla and NVIDIA traded some light blows earlier this year, with Tesla claiming their FSD Computer was more capable than NVIDIA’s Xavier chip. This recent unveiling puts NVIDIA more in line with Tesla’s FSD computer’s capability.

One of the problems Tesla saw with NVIDIA’s chip was in power consumption, which is critical for electric cars. We don’t know what power consumption the final configuration of NVIDIA’s next self-driving computer will have, but it’s estimated that the new Orin chip will consume 60-70 watts on its own.

If configured into a multi-processor computer like the Xavier was in the DRIVE PX Pegasus, this could end up using similar amounts of power as the Pegasus, which could tax EV batteries significantly. Not as much as the vehicle’s drive system, mind you, but any drain on the battery is worth considering.

There are more issues worth considering, including memory bandwidth and final configuration of the chips, so these systems can’t be compared directly, especially when we don’t know all the specs of NVIDIA’s Orin yet (or, indeed, what Tesla’s computers will look like in 2022). Also, all of this processing power depends on the data and algorithms that are being fed through the computers.

But clearly both of these companies are not content to stand still, and want to keep offering more powerful computers for eventual self-driving applications.