News Categories

Tesla reportedly working with AMD on custom AI chip for self-driving cars

By Michael Low & Koh Wanzi - on 21 Sep 2017, 10:00pm

Tesla reportedly working with AMD on custom AI chip for self-driving cars

Image source: Tesla.

Tesla is reportedly developing its own custom-built AI chip for fully autonomous self-driving cars, and it’s working with AMD to do that.

Tesla CEO Elon Musk has promised that a fleet of such vehicles would be ready for the road by 2019, and it recently hired ex-AMD chip designer Jim Keller, who helmed the development of a number of chips at AMD and also helped Apple build its own mobile processors.

News of the collaboration came from GlobalFoundries CEO Sanjay Jha, according to a CNBC report, but a company spokesperson claims that the CEO was misquoted, as GlobalFoundries does not comment on customers or potential customers.

CNBC later clarified that Jha mentioned Tesla as an example of a company working with chip fabricators, but did not actually say that it was a GlobalFoundries customer.

The Abu Dhabi-owned GlobalFoundries fabricates chips, and it has a wafer supply agreement with AMD that runs through 2020.

That said, CNBC also cited an unnamed source as saying that Tesla has already received back samples of the first implementation of the AI processor, and is now running tests on it.

Under Musk, Tesla has focused on vertical integration, so this report is in line with its push to make everything itself and reduce reliance on other companies.

More than 50 people are said to be working on the chip under Keller at Tesla, which has since recruited other AMD engineers, such as Bill McGee, Dan Bailey, and Ganesh Venkataramanan.

Existing Tesla vehicles use NVIDIA GPUs as part of the former’s Autopilot 2.0 update, so the latest news, if confirmed, is a clear win for AMD.

A fully autonomous vehicle would require extremely capable chips to handle the data and telemetry generated by the multitude of cameras and sensors, so many of the world’s leading chipmakers have begun carving out a slice of the pie for themselves.

For instance, Intel just announced a closer collaboration with Waymo, even though its hardware has been powering Google’s cars since 2009.

Today’s GPUs are suitable for parallel processing and various AI workloads, but a custom chip would be able to have its performance tweaked to handle just a few narrow tasks more efficiently.

Source: CNBC.