US EV pure play Tesla will invest $500mn over the next five Years to develop its new Dojo supercomputer at its Riverbend site in Buffalo, NY, the New York state governor Kathy Hochul has confirmed.
Dojo will have “the power to process millions of terabytes of data from Tesla’s EVs in the Riverbend gigafactory”, the governor says. “That data will allow the company to improve the safety and engineering of its full self-driving and advanced driver assisted vehicles.
“Tesla’s decision was informed by New York’s reliable power supply, strong talent pipeline and availability of usable space for the project,” Hochul continues.
Sign up for EV inFocus
EV inFocus: Hard news for the corporate e-mobility sector
No spam. Unsubscribe anytime.
“A lot of our progress in self-driving is training-limited,” Tesla CEO Elon Musk said on the company’s Q4 results call in late January. “Something that is important with training, it is much like a human. The more effort you put into training, the less effort you need in inference.
“Just like a person, if you train in a subject — [the] classic 10,000 hours — the less mental effort it takes to do something,” Musk continues. “If you remember when you first started to drive, how much of your mental capacity it took to drive.
“You had to be focused completely on driving. Then, after you have been driving for many years, it only takes a little bit of your mind to drive and you can think about other things and still drive safely. The more training you do, the more efficient it is at the inference level. So we do need a lot of training.”
Dual approach
And Dojo is part of a twin strategy that Tesla is pursuing. “We are obviously hedging our bets here with significant orders of Nvidia GPUs,” Musk explains, before clarifying that GPU is “the wrong word” — given that the chips in question will not output graphics. “Neural net processing unit”, may be a better phrase, he suggests.
“We are pursuing the dual path of Nvidia and Dojo. But I would think of Dojo as a long shot,” Musk says.
“It is a long shot worth taking because the payoff is potentially very high. But it is not something that is a high probability; it is not like a sure thing at all. It is a high-risk, high-payoff programme.”
Musk stresses that Dojo “is working”, as it is doing training jobs, and Tesla is still looking to scale it up. “We have plans for Dojo 1.5, Dojo 2, Dojo 3. I think it has got potential. But I cannot emphasise enough — high risk, high payoff,” he continues.
“It is a very interesting programme; it has the potential for something special,” Musk says. “There is also our inference hardware in the car. We are now on what is called Hardware 4, but it is actually version 2 of the Tesla-designed AI inference chip.
Big leap forward
“We are about to complete design of Hardware 5, which is actually version 3 of the Tesla-designed chip. Because version 1 was Mobileye, version 2 was Nvidia, and then version 3 was Tesla. And we are making gigantic improvements from Hardware 3 to Hardware 4 to Hardware 5," the Tesla chief continues.
And there have been, in his view, unforeseen benefits from this step-by-step approach. "I think Tesla is probably the most efficient company in the world for AI inference —out of necessity," he suggests.
"We have actually had to be extremely good at getting the most out of hardware, because Hardware 3 at this point is several years old. I think we are quite far ahead of any other company in the world in terms of AI and inference efficiency, which is going to be a very important metric in the future in many arenas."
And, unsurprisingly, Musk is already thinking ahead on future possibilities. “There is a potentially interesting play where when cars are not in use in the future that the in-car computer can do generalized AI tasks, can run a sort of GPT-4 or something like that," he speculates.
"If you have got tens of millions of vehicles out there — even in a robotaxi scenario where they are in heavy use, maybe they are used 50 out of 168 hours — that still leaves well over 100 hours of time available of compute hours.
“It is possible, with the right architectural decisions, that Tesla may in the future have more compute than everyone else combined,” Musk continues.
Peter boasts 20+ years’ experience in energy/commodity journalism & analysis.
He is convinced the next few decades will be all about energy transition. Insider Focus is a front-row seat for the ride.
