Custom Tesla AI chip(s)

[Pale_Rider]

Not much info here but confirmation at least that they are working on one:

Elon Musk confirms that Tesla is working on its own new AI chip led by Jim Keller

 

[ggies07]

I'm trying to find a video of it and this is all I've found so far:

 

[Tiger]

Tesla is not producing a self-driving AI chip with AMD, despite reports

 

[Pale_Rider]

Tesla is not producing a self-driving AI chip with AMD, despite reports

Older article and technically they aren't doing it WITH AMD, they took AMD's guy and are developing it in-house...

 

[Techbwoys]

Tesla designing there own AI Chip is a sign that they are forward looking. I think companies need to invest and development there core technologies. Looking at the problem differently and seeing what you can innovate on. Short sight people would say Tesla is doing to much but all they are doing is finding a path to stay ahead. Apple does with the there chips like the W1,W2, T1,T2 chips. AirPods pairing is amazing. Apple also did 64 bit leaving the industry was like no way taking competitors be surprise. Google does the say buy have custom chips as well. Lets not forget the billionaires club is a small club. Vertical integration will keep Tesla ahead.

 

[Lasairfion]

A friend of mine who works in programming and did a lot of the open source work for code used on the space station, in mobile phones etc said to me, last year, that he fully expected programming to turn towards being hardware specific; with tailor-made hardware components for doing specific roles becoming the norm.

Rather like how computers never used to have graphics cards, but now GPUs are essential for gaming; future projects will utilise customised hardware solutions to get greater percentage benefits with regards to a specific task.

Tesla has simply understood this and implemented it.

 

[MarkS22]

A friend of mine who works in programming and did a lot of the open source work for code used on the space station, in mobile phones etc said to me, last year, that he fully expected programming to turn towards being hardware specific; with tailor-made hardware components for doing specific roles becoming the norm.

Rather like how computers never used to have graphics cards, but now GPUs are essential for gaming; future projects will utilise customised hardware solutions to get greater percentage benefits with regards to a specific task.

Tesla has simply understood this and implemented it.

They were doing this from the beginning, though, using the EyeQ3. It’s purely designed for computer vision processing. The original DrivePX1 even had a dedicated EyeQ3 on it.

I suspect the urgency to create their own came when Tesla and Mobileye parted ways, leaving them with a DrivePX2 that dropped the dedicated computer vision hardware support.

 

[Tezlanian]

A friend of mine who works in programming and did a lot of the open source work for code used on the space station, in mobile phones etc said to me, last year, that he fully expected programming to turn towards being hardware specific; with tailor-made hardware components for doing specific roles becoming the norm.

Rather like how computers never used to have graphics cards, but now GPUs are essential for gaming; future projects will utilise customised hardware solutions to get greater percentage benefits with regards to a specific task.

Tesla has simply understood this and implemented it.

This will happen because of the death of of Moore's law - Wikipedia coming very soon without a materials science miracle. The only options left are parallelization of tasks or physical manifestations of the algorithms themselves in custom hardware. Mileage varies on which is better for different scenarios. Probably some more usage of Field-programmable gate array - Wikipedia as well.

There's also significant energy (heat) savings in ASICs.