Tesla’s dojo, time line Techcrunch

Rate this post


Elon Musk does not want Tesla to be just a car manufacturer. He wants Tesla to be an AI company that figures out how to get cars to move.

It is crucial for this mission Dogo, a custom -made Tesla Super Computer Designed to train its full neural driving networks (FSD). FSD is not really a fully self -governing self; He may perform some automated driving tasks, but still requires a careful person behind the wheel. But Tesla considers more data, more skillful power and more training, it can pass the threshold from almost self-driving to full self-driving.

And that’s where Dogo enters.

Musk has been annoyed for some time, but the CEO has been discussions about the supercomputer in 2024. Now that we are in 2025, another supercomputer called Cortex has entered the chat, but the meaning of DOJO for Tesla may be existential – with EV Sales Rolling, investors want assurances that Tesla can achieve autonomy. Below is a Dogo time line also mentions promises.

2019

First references to dojo

April 22 – At Tesla Autonomy DayThe carmaker had his own AI team on stage to talk about autopilot and complete self -government, and AI feeds them both. The company shares information about Tesla’s custom chips, which are designed specifically for neural networks and self -driving car management.

During the event, Musk annoyed Dojo, revealing that he was a supercomputer for training AI. He also notes that all Tesla cars that are manufactured at the time would have all the hardware needed for full self -driving and only need to update the software.

2020

Musk begins dojo roadshow

February 2 – Musk He says Tesla will soon have more than a million connected vehicles worldwide with sensors and calculations needed for full self-driving-and will be associated with DOJO’s capabilities.

“Dojo, our supercomputer training, will be able to process huge amounts of video training data and effectively launch hyper -reference arrays with a huge number of parameters, lots of memory and ultra high bandwidth between the nuclei. More about this later. “

August 14 Musk repeats Tesla’s plan to develop a neural network training computer called Dojo “to process truly huge amounts of video data”, calling it “beast”. He also says the first version of dojo is “About a year,” Which would put the date of launch sometime around August 2021.

December 31 Says Elon Dojo is not neededBut this will make self-driving better. “It is not enough to be more favorable than human drivers, in the end the autopilot should be more than 10 times more favored than human drivers.”

2021

Tesla makes dojo official

August 19 – The manufacturer formally announces dojo of Tesla’s first AI dayAn event designed to attract engineers to Tesla’s AI team. Tesla also presents its D1 chip, which the carmaker says will use – along with the NVIDIA graphics processor – to power the DOJO supercomputer. Tesla notes that its AI cluster will have 3000 D1 chips.

October 12 – Tesla versions a Dojo Technology Whitepaper“Tesla and arithmetic floating points of floating points.” Whitepaper outlines a technical standard for a new type of binary arithmetic with floating comma that is used in neural networks in deep training and can be realized “entirely in software, entirely in hardware or hardware in any combination of software and hardware. “

2022

Tesla reveals dojo progress

August 12 – Musk says Tesla will ”phase in dojo. You will not need to buy such gradual graphics processors next year. “

September 30 – In Tesla’s Second day of AIThe company reveals that it has installed the first DOJO cabinet, testing 2.2 megawatts testing. Tesla says it has built a tile a day (which is made up of 25 D1 chips). Tesla Demos Dojo on stage, performing a stable diffusion model to create AI generated image of Cybertruck on Mars.

The important thing is that the company sets the target date of the full cluster of Exapod, which will be completed by Q1 2023, and says it plans to build a total of seven exapods in Palo Alto.

2023

“A long shot bet

April 19 – Musk tells investors during Tesla’s first quarter revenue The fact that Dojo “has the potential for improving the order in the cost of training” and also “has the potential to become a selling service that we would offer to other companies in the same way that Amazon Web Services offers web services.”

Musk also notes that “he will look at Dogo as a long shot,” but “a bet worth it to do.”

June 21st Tesla Ai X Account Posts that the company’s neural networks are already in customer vehicles. The thread includes a timetable schedule of Tesla’s current and projected computing power, which marks the beginning of Dogio production on July 2023, although it is not clear whether this is about the chips D1 or the supercomputer itself. Says Musk On the same day, when DOJO was already online and tasked in Tesla data centers.

The company also designs that Tesla’s calculation will be top five in the world until about February 2024 (no indications that this was successful) and that Tesla will reach 100 Exaflops by October 2024.

July 19 – Tesla notes In his second quarter profit report, he began production of dojo. Musk also says Tesla plans to spend more than $ 1 billion on DOJO by 2024.

September 6 – Muscular publications on x This Tesla is restricted by AI Training Compute, but this NVIDIA and DOJO will repair it. He says data management of approximately 160 billion video frames that Tesla receives from its cars a day is extremely difficult.

2024

Scaling plans

January 24 – During the winning of the fourth and year -round profits of Tesla, Musk again admits that DOJO is a high -risk high -prize project. He also says that Tesla has pursued the “double road of Nvidia and Dojo”, that Dogio works and “does the job.” He notes that Tesla scales it and has “plans for dojo 1.5, dojo 2, dojo 3 and more.”

January 26 – Tesla has announced plans to spend $ 500 million to build a Dojo supercomputer in BuffaloS Then Musk downplayed the investment somewhat, Post The fact that while $ 500 million is a large sum, it is “only equivalent to a 10K H100 system from NVIDIA. Tesla will spend more on Nvidia’s hardware this year. The bets on the AI ​​competition table are at least several billion dollars a year at this time. “

April 30 – At the North America Technology Symposium of TSMC, the company says the next generation-D2 training tile, which puts the whole dojo tile on a silicone waffle instead of connecting 25 chips to make a tile in production, according to IEEE spectrumS

May 20 – Musk notes That the back of the extension of the Giga Texas factory will include the construction of a “super thick, water -cooled supercomputer group”.

June 4 – A CNBC report It reveals that Musk has diverted thousands of NVIDIA chips reserved for Tesla to X and XAI. Once it was initially said that the report was incorrect, Musk publications on x That Tesla had no place to send NVIDIA chips to include them due to the prolonged construction of the southern giga expansion of the Texas, “So they would just sit in a warehouse.” He noted that the extension would “accommodate 50k H100 for training of FSD “.

He too Posts:

“Out of approximately $ 10 billion in AI-related costs, I said that Tesla will make this year, about half is internal, mainly on Tesla-designed computer and sensors present in all our cars, plus dojo. To build Superclusters for AI training, NVIDIA hardware is about 2/3 of the cost. My current best assumption about NVIDIA’s purchases from Tesla is 3B to $ 4b this year. “

July 1 – Musk reveals on x These current Tesla vehicles may not have the right hardware for the next -generation AI model of the company. He says that the approximately 5-fold increase in the number of AI parameters of the next generation is “very difficult to achieve without upgrading the computer for the conclusion of the vehicle.”

NVIDIA Delivery Challenges

July 23 – During the call of Tesla’s second quarter profit, Musk says the demand for NVIDIA hardware is “so high that it is often difficult to get graphic processors.”

“I think this requires us to make a lot more efforts on Dojo to ensure that we have the training opportunities we need,” Musk says. “And we see a way to be competitive with Nvidia with dojo.”

The Tesla investor deck graphics predicts that the Tesla Ai Capital Capital will invade approximately 90,000 H100 equivalent graphics processors by the end of 2024, compared to about 40,000 in June. Later the same day of X, Muscular publications This dojo 1 will have “approximately 8K H100 equivalent of online training by the end of the year.” It also publishes photos On the supercomputer that looks use the same stainless steel exterior, resembling a refrigerator as Tesla cyber.

From dojo to bark

July 30 – AI5 is a distance of ~ 18 months from large volume production, says Musk in a Response On a publication from someone who claims to start a club from “Tesla HW4/AI4 owners, angry to stay behind when AI5 comes out.”

August 3 – Musk publications on x That he took a walk on the Tesla’s “Supercomput Cluster in Giga Texas (also known as Cortex)”. He notes that this will be made of approximately 100,000 H100/H200 NVIDIA GPU with “Massive Storage for Video Training of FSD & Optimus”.

August 26 – Musk publications on x A video of Cortex, which he calls “Giant New Ai Training Supercluster, which is built in Tesla HQ in Austin to allow AI in the real world.”

2025

No DOJO updates in 2025

January 29 – Call the Tesla Q4 and a year -round profit of 2024 Included does not mention Dojo. Cortex, Tesla’s new Ai Supercluster Austin Gigafactory, has appeared. Tesla noted in his Shareholder deck that it has completed the deployment of Cortex, which consists of approximately 50,000 H100 NVIDIA GPU.

“Cortex helped to activate the FSD V13 (Supervisory), which boasts great improvements in safety and comfort thanks to a 4.2x data increase, video inputs with higher resolution … among other improvements,” according to the letter, ” S

During the call, Finance Officer Vaibhav Taneja noted that Tesla accelerated Cortex to accelerate the deployment of the FSD V13. He said the accumulated capital expenditure associated with AI, including infrastructure, “has been approximately $ 5 billion so far.” In 2025, Tanezha said he expected Capex to be flat as it was associated with AI.

This story originally published on August 10, 2024 and we will update it with the development of the new information.



 
Report

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *