Tesla Dojo: A Computational Marvel - podcast episode cover

Tesla Dojo: A Computational Marvel

Sep 13, 20236 min
--:--
--:--
Download Metacast podcast app
Listen to this episode in Metacast mobile app
Don't just listen to podcasts. Learn from them with transcripts, summaries, and chapters for every episode. Skim, search, and bookmark insights. Learn more

Episode description

Tesla's Dojo is a supercomputer purpose-built to process and recognize computer vision video data. Going live in July 2023, Dojo serves as the training ground for Tesla's machine learning models aimed at bolstering its Full Self-Driving (FSD) system. Unlike conventional supercomputers, Dojo is architected to handle millions of terabytes of real-life video data from over 4 million Tesla vehicles on the road. This unique task called for an architecture that broke away from traditional supercomputer designs, focusing on efficiency and scalability.

Transcript

Hi everybody and welcome back to the Elon Musk podcast where we discuss all things technology, space, entrepreneurship and more. And today we're talking about something that's definitely caught the attention of Wall Street Tesla shareholders and a I enthusiasts alike.

Tesla stock is climbing higher is something to do with the very own supercomputer at Tesla Dojo. Recently, Morgan Stanley upgraded Tesla stock from equal weight to overweight and hiked the price target from $200.00 to $400.00. But this optimistic outlook largely stems from their view of Dojo's immense potential, not just in advancing Tesla's A I capabilities, but also as a service that could be offered on

its own. Now, Dojo's goal is to effectively process millions of terabytes of video data captured from real life driving situation from Tesla's plus 4 million cars. Now this goal led to a considerably different architecture than conventional supercomputer designs and going live in July 2023, Dojo serves as the training ground for Tesla's machine learning models aimed at bolstering its full self driving system or FSD.

And unlike conventional supercomputers, Dojo is architected to handle millions of terabytes of real life video data from over 4 million Tesla vehicles on the road right now. Now this unique task called for an architecture that broke away from traditional supercomputer designs, focusing on efficiency and scalability. Now, before Dojo, Tesla was already a formidable name in computing.

The company had several parallel computing clusters, one of which was, by some metrics, among the top supercomputers globally. However, there were debates about its actual performance due to unclear metrics. These test the GPU clusters were pivotal in training neural networks to better understand real world driving conditions, and with MOJO, Tesla aims to supersede its own lofty standards.

Mentioned first in 2019 during Tesla's Autonomy Investors Day, MOJO faced some blaze due to power and thermal challenges. Nevertheless, it officially came into the limelight in 2021 and moved into production use in 2023, equipped with 10,000 NVIDIA H100GP U's to enhance his performance. Now Dojo's reception among analysts is mixed, yet mostly

positive. And while some say it's impressive but won't change the supercomputing game, others call it an inhouse developed game changer that can massively speed up the development of autonomous vehicles. Now the endorsement from Morgan Stanley is particularly significant, suggesting that Dojo could be a catalyst for an unprecedented increase in Tesla's market value.

The idea is that Tesla's focus on custom designed a I chips for video training could afford a formidable advantage in the autonomous vehicle sector and beyond. At Tesla's A I day in 2021, the introduced Dojo, but at the time they were just getting started. They had a first chip and some training tiles, but we're still in the process of constructing a full Dojo cabinet in a cluster. Fast forward to a I day 22.

They made considerable headway, including developing a full system tray, and they're aiming to ramp it up to a 100 exaflop capability by the end of 2024. This is particularly significant as Tesla's ambitions becoming increasingly computer constrained, especially in selfdriving endeavors. Moreover, Tesla's feature doesn't just rest on car sales. Supercharging network with partnerships like the one they've received established with Hilton Hotels presents another potential revenue stream.

Adam Jonas believes that the advantages and the advancements and chip technology will add to this by possibly extending the self driving licensing and operating system licensing. Now let's talk numbers for those who love them. Morgan Stanley's price target shift from 2:50 to 400 isn't trivial. The analyst broke down the value and factored in Dojo's impact on the overall valuation. This was a pivotal enough shift to lift Tesla stock by nearly

10% following the notes release. And get this, aside from Tesla, Morgan Stanley also sees the possibility of non Tesla fleet licensing revenue, which means the reach of Tesla's technological advancements could go beyond its own lineup of electric vehicles. And while the core business of Tesla has always been tied to electric vehicles, this new focus on Dojo is a revenue generator isn't really a wild shift. In fact, we've mentioned this

before. The Dojo could potentially be offered as a service, argumenting its utility and contributing to Tesla's overall value. So the concept isn't new. But what is notable is how mainstream financial analysts are now seeing Dojo not just as a little side project for Tesla for their own vehicles, but as a key aspect of Tesla's future growth and the value. Now. One more thing before we go There are other market forces at play.

Currently, there's a looming potential of a massive UAW strike that can negatively affect Tesla's competitors. This labor strike in competing companies could inadvertently give Tesla a huge edge, further fueling optimism around Tesla stock and future prospects. That wraps up the show today. Hope you found it informative. If you enjoyed the show, don't forget to hit the subscribe or the follow button on your

preferred podcast platform. It's absolutely free, takes a second, and helps us out tremendously, and this way you'll never miss out on any Elon Musk news. O thanks for tuning in and please take care of yourselves and each other and I'll see you in the next one.

Transcript source: Provided by creator in RSS feed: download file
For the best experience, listen in Metacast app for iOS or Android