Tesla Next Generation Dojo Superchips
Tsmc has reported that Tesla’s next-generation Dojo superchips are currently in production.
But whilst there are huge improvement of the previous generation, the new, version coming in two years time will be 40 times more powerful.
That is utterly mind-blowing. Tesla’s supplier, Taiwan Semiconductor Manufacturing Company Limited, or TSMC, has revealed that Tesla’s Dojo Supercomputer is about to get a lot better.
Compute Power at Tesla
Tesla has been heavily investing in its AI training, compute power. It’s no longer compute constrained.
According to Elon Musk, Tesla has been compute constrained for the entire history of the organization.
What this has meant is that they’ve had to select which data they would use to train Dojo.
Tesla’s Strategies and Partnerships
They no longer have to do that all the, apparently, billions of miles that Tesla drivers are driving with full self-driving turned on, Tesla can actually use all the data in that now because their super computer is so massive that it can handle this amount of data.
I find that hard to believe, but that’s what Tesla say. Tesla, though, has been heavily investing in AI training compute power, both through buying NVIDIA hardware and buying its own for its Dojo program.
First Generation Dojo Supercomputing Platform
The first generation of its Dojo supercomputing platform, it’s actually one of the fastest, most powerful super computers in the world, went into operation last summer.
Shortly after that, it was reported that Tesla had expanded its partnership with TSMC, a large semiconductor company that manufactures the Dojo chip with the automaker.
TSMC Contributions to Tesla’s Dojo Chip
Now, apparently, Tesla was using a chip called the D1 chip, but now they are putting 25 of these chips in an area the size of a dinner plate.
It’s about this big, 25 chips in this area. They’re supplying 15 kilowatts of power to these 25 chips, and Apparently, therefore, Tesla has assembled a fully working system that is better than what it previously had.
Next-Generation Dojo Chip
Tsmc has confirmed that Tesla’s next generation Dojo chip has entered production, and their work on technology that could deliver much greater power to Dojo in 2027.
Here’s a quote from them at TSMC’s North American Technology Symposium, the company detailed both its semiconductor technology and chip packaging technology roadmaps.
The former is key keeping the traditional part of Mohs law going.
The latter could accelerate a trend toward processes made from more and more silicon, leading quickly to systems the size of a full silicon wafer.
Production Milestones
Such a system, Tesla’s next-generation dojo training tile, apparently is in production, says TSMC.
So this is Tesla’s proprietary-owned chip that they use for their full self-driving. In 2027, the foundry plans to offer technology for more complex wafer scale systems than Teslas that could deliver 40 times as much computing power as today’s.
What is Tesla doing with this? What are they actually? What’s the point? Well, Tesla’s new planned $500 million dojo cluster is in New York.
This is what Tesla is using to basically create full self-driving to make it work, and also to power their robotics. For example, the Tesla bot.
This is important for both of those areas, it actually is a really good crossover. Tesla is also building a new 100 megawatt data center to train its self-driving AI at the Gigafactory in Texas.
Applications of Dojo Supercomputers
This system, though, apparently, we use NVIDIA hardware. It’ll be in a different system it’s worth pointing out that the NVIDIA chips are an entirely separate system to Tesla’s main Dojo Supercomputer.
But a 100 megawatt data center is massive. And will this train Tesla’s AI in terms of its AI for full self-driving and its AI for Tesla bots? I think it’ll do both of those.
But it’s Tesla a little bit secretive about some of this stuff, as you imagine you would be when you’re spending this money.
Tesla Financial Commitment to AI Development
How much money is Tesla spending? Apparently, by the end of this year, they will have spent $10 billion.
You can see where Tesla is firing staff that they believe it’s not essential because Tesla are going, You know what? We’re all in on full self-traffing.
Strategic Priorities
We’re all in on robotics, on having robots. And the reason is because I think primarily, people don’t realize this, but the robotics industry is far more lucrative than the automotive industry, far more, by a factor of 20.
I mean, the global labor industry is worth trillions and trillions and trillions of dollars every single year.
If Tesla can get just a small little slice the global labor industry, that is worth far more than Tesla’s entire business today.
You can see why this artificial intelligence is now Tesla’s number one priority. I think they’re prioritizing it over everything.