Tesla is ditching Nvidia GPUs and developing its own self-driving car AI chips

[Polycount]

Following numerous crashes involving Tesla's Autopilot self-driving car technology, the company recently announced that it's been quietly developing improvements to the hardware behind the tech.

Tesla is moving away from Nvidia's GPU-based offering and moving towards its own proprietary AI chips. This information was revealed in an earnings call with company CEO Elon Musk. During the call, Musk added that Tesla's in-dev chip will most likely be used in its Model 3, Model S, and Model X electric vehicles.

If you're wondering why Tesla would move away from a tried-and-true solution and strike out on its own, the answer lies in performance.

Company director Pete Bannon hopped on the call to say Tesla's new hardware will be able to process up to 2,000 frames per second - a significant bump from the 200 frames per second Nvidia's hardware could handle.

Faster processing speeds are likely not the only reason Tesla wants to develop their AI hardware in-house, though. By building the chip internally, Tesla can move its tech forward at a much quicker rate than its competitors.

As TechCrunch notes, the company won't have to wait for Nvidia to roll out its next set of AI hardware. Instead, Tesla can move forward whenever its ready, and fix problems or improve its hardware as needed. Additionally, it's worth noting that Musk claims Tesla's in-house hardware will cost them about the same as the tech they were buying from Nvidia.

If you're wondering when Tesla plans to implement this hardware in its vehicles, Bannon believes the upgrade process will start as early as next year.

Permalink to story.

https://www.techspot.com/news/75782-tesla-ditching-nvidia-gpus-developing-own-self-driving.html

 

[JaredTheDragon]

Oh wow, now this is impressive! A profitless, government-subsidized nobody group that's never made a single PU can outperform Nvidia's latest and best offerings by 1000% just at the drop of a hat! Wave a magic wand and POW! The best PU ever made, instantly. Just add water.

 

[MonsterZero]

I would buy a Tesla video card if they started making one lol

 

[Theinsanegamer]

I would buy a Tesla video card if they started making one lol

It would cost $10k, have a 75% failure rate, and come without any kind of control panel in the drivers so you cant change any settings.

Oh, and it takes 2 years to be assembled and shipped to you.

 

[JaredTheDragon]

Nvidia already made these anyway. Musk is gonna have a rough time with branding now.

 

[Jackwoz]

Yeah, but will it be able to play Crysis

 

[wiyosaya]

They are probably going to task this with figuring out how to crash pesky autopilot users who cannot keep their hands on the wheel ten times faster than before.

 

[Evernessince]

Oh wow, now this is impressive! A profitless, government-subsidized nobody group that's never made a single PU can outperform Nvidia's latest and best offerings by 1000% just at the drop of a hat! Wave a magic wand and POW! The best PU ever made, instantly. Just add water.

It's possible given their design would be much more focused than Nvidia's offerings. It's the same deal with the bitcoin market, when you make a chip for a specific task it will beat general purpose products. The car market is most certainly big enough for them to make chips specifically designed for it.

 

[JaredTheDragon]

It's possible given their design would be much more focused than Nvidia's offerings. It's the same deal with the bitcoin market, when you make a chip for a specific task it will beat general purpose products. The car market is most certainly big enough for them to make chips specifically designed for it.

If the car market could produce faster, more potent chips than the world's leading manufacturers then they wouldn't be the world's leading manufacturers. Intel, AMD, and Nvidia wouldn't exist. Samsung and Apple and Qualcom wouldn't exist. Money does not equal technological success at all - only real technology becomes successful, because it's only real tech that is useful. It doesn't matter how much money you throw at an idea if the idea is wrong.

Mitsubishi doesn't make graphics cards or processing units, nor Nissan, nor Toyota, nor any American or European company - all of them far, far larger than Tesla in every way. The Nissan Leaf is still the best-selling electric car by far, you know. Tesla doesn't even rank.

 

[Evernessince]

If the car market could produce faster, more potent chips than the world's leading manufacturers then they wouldn't be the world's leading manufacturers. Intel, AMD, and Nvidia wouldn't exist. Samsung and Apple and Qualcom wouldn't exist. Money does not equal technological success at all - only real technology becomes successful, because it's only real tech that is useful. It doesn't matter how much money you throw at an idea if the idea is wrong.

Mitsubishi doesn't make graphics cards or processing units, nor Nissan, nor Toyota, nor any American or European company - all of them far, far larger than Tesla in every way. The Nissan Leaf is still the best-selling electric car by far, you know. Tesla doesn't even rank.

Given that ASIC manufacturers have already created chips faster then AMD and Nvidia cards at mining bitcoin, it isn't hard to believe the same could happen here as well. What you find so hard to believe has been done in the past already. Re-read my first post, this has nothing to do with graphics cards. Tesla isn't going out to make graphics card, they are looking at making an ASIC.

 

[Flebbert]

I think Tesla have stolen a few people from the silicon world to do this, you higher people to do this.

Actually just read the post above.. that pretty much covers it.

 

[Anton Skryaga]

If the car market could produce faster, more potent chips than the world's leading manufacturers then they wouldn't be the world's leading manufacturers. Intel, AMD, and Nvidia wouldn't exist. Samsung and Apple and Qualcom wouldn't exist. Money does not equal technological success at all - only real technology becomes successful, because it's only real tech that is useful. It doesn't matter how much money you throw at an idea if the idea is wrong.

Mitsubishi doesn't make graphics cards or processing units, nor Nissan, nor Toyota, nor any American or European company - all of them far, far larger than Tesla in every way. The Nissan Leaf is still the best-selling electric car by far, you know. Tesla doesn't even rank.

Given that ASIC manufacturers have already created chips faster then AMD and Nvidia cards at mining bitcoin, it isn't hard to believe the same could happen here as well. What you find so hard to believe has been done in the past already. Re-read my first post, this has nothing to do with graphics cards. Tesla isn't going out to make graphics card, they are looking at making an ASIC.

ASIC not 'faster' in general purpose algorithms. Bitcoin miner is a just SHA-256 calculator. It useless outside of it's purpose. Algorithms for modern 'AI' systems is not so simple as for SHA. And ASICs have certain problems with memory-hungry algorithms (that's why no ASICs for ZCash for example).

 

[Evernessince]

ASIC not 'faster' in general purpose algorithms. Bitcoin miner is a just SHA-256 calculator. It useless outside of it's purpose. Algorithms for modern 'AI' systems is not so simple as for SHA. And ASICs have certain problems with memory-hungry algorithms (that's why no ASICs for ZCash for example).

Well it's a good thing this isn't general purpose.

"And ASICs have certain problems with memory-hungry algorithms"

Yes and they also said ethereum would never have an ASIC yet Bitmain was able to make one.