Elon Musk does not want Tesla to be just a car manufacturer. He wants Tesla to be Amnesty International, which has discovered how to make cars lead itself.
It is important for that task, the super -computer design that is designed from Tesla designed to train full self -driving nervous networks (FSD). FSD is not actually self -driving completely; It can perform some automatic driving tasks, but a person still requires vigilance behind the steering wheel. But Tesla is considering more data, more arithmetic energy and more training, can cross the threshold from self -driving almost to full self -driving.
This is where Dogo comes.
Musk was harassing Dojo for some time, but the executive authority has stayed on the super computer throughout 2024. And and after that in 2025, I entered the other super computer called Cortex chat, but the importance of Dojo to Setla may still exist – with EV – with EV. Sales decline, investors want assurances that Tesla can achieve independence. Below is a schedule for Dojo and promises.
2019
The first Dogo is mentioned
April 22 – On Tesla’s autonomy day, the Automobile Manufacturer had the Amnesty International team on stage to talk about the automated pilot and the full self -driving, and the AI that it works in both. The company shares information about the specially designed Tesla chips that are specially designed for nervous networks and self -driving cars.
During this event, Musk Dojo raises, reveals a super computer to train artificial intelligence. It also indicates that all Tesla cars that are produced at the time will get all the devices needed for full self -guidance and only need to update the program.
2020
Musk starts to display Dojo
February 2 – Musk He says Tesla will soon get more than a million connected cars all over the world with a required sensors and account for full self-driving-and accumulate dojo capabilities.
“Dojo, our super -training computer will be able to process huge amounts of video training data and operate the superior spacecraft efficiently with a large number of parameters, a lot of memory and the superior domain show between core. More than this later.”
August 14 – Repeat musk The Tesla Plan for the Development of a Neurological Network Training Computer called Dojo “to tackle huge quantities of video data”, called “a monster”. It also says that the first version of Dojo is “About a year,” Which would put the date of launch somewhere in August 2021.
December 31 – Eleon says Dojo is not necessaryBut it will make self -driving better. “It is not enough to be safer than human drivers, the automated pilot must eventually be more than 10 times than human drivers.”
2021
Tesla makes Dogo responsible
August 19 – DOJO is officially announced on the day of artificial intelligence in Tesla, an event aimed at attracting engineers to the Tesla’s Ai team. Tesla also offers its D1 chip, which the auto industry says it will use – along with the NVIDIA graphics processing unit – to operate the dojo superCant. Tesla notes Amnesty International Group that will include 3000 D1 chips.
October 12 – Tesla versions A Dojo Technology Whiteper“A guide for TESLA’s floating floating point formats.” The white paper defines a technical standard for a new type of floating dual account that is used in deep educational nervous networks and can be implemented “completely in programs, completely in devices, or in any set of programs and devices.”
2022
Tesla reveals Dogo’s progress
August 12 – Musk Tesla says.A stage in Dogo. You will not need to buy the largest possible number of additional graphics processing units next year. “
September 30 – On the second AI Day in Tesla, the company reveals that it installed the first Dojo cabinet, as it tests 2.2 megawatts of pregnancy test. Tesla says she was building one slab per day (which consists of 25 d1 chips). Tesla Demos Dojo on the stage that works to play a stable spread model to create an artificial intelligence of “Cybertruck on Mars”.
More importantly, the company sets a targeted date for the complete Exapod group to complete it by Q1 2023, and says it plans to build a total of seven Exapods in Palu Alto.
2023
A long bet‘
April 19 – Musk informs investors during Tesla profits in the first quarter of Tesla, which has Dojo “the possibility of an arrangement of size in the cost of training”, and also “has the ability to become a serviceable service that we will provide to other companies in the same way.
Musk also notes that “he looks at Dojo as a kind of long bet”, but it is “a bet that deserves to be made.”
June 21 – Tesla AI X account Pillars The company’s nerve networks are already in customer vehicles. The thread includes a graphic drawing with a timetable for the current and expected Tesla account power, which sets the beginning of the production of Dojo in July 2023, although it is not clear whether this indicates D1 chips or the supercar itself. Says musk On the same day that dojo was already online and operating tasks in Tesla Data Data centers.
The company also scores that the Tesla account will be the first five places in the whole world by around February 2024 (there are no indications that this was successful) and that Tesla will reach 100 Exaflops by October 2024.
July 19 – Timing Notes In the second quarter profit report, the production of Dojo began. Musk also says that Tesla plans to spend more than one billion dollars on Dogo until 2024.
September 6 – Musk functions On x Tesla is limited to the artificial intelligence training account, but NVIDIA and Dojo will fix this. He says data management from the 160 billion video frames of Tesla cars daily is very difficult.
2024
Plan to expand its scope
January 24 – During the Tesla’s profit call in the fourth and entire year, Musk admits again that Dojo is a high -risk and highly rewarding project. He also says that Tesla was following the “double path for NVIDIA and Dojo”, that “Dojo works” and “performs training functions”. It is noted that Tesla is distorting it and has “plans for Dojo 1.5, Dojo 2, Dojo 3, etc.”.
January 26 – Tesla has announced plans to spend $ 500 million to build Dojo Supercycrane in Buffalo. Musk reduces somewhat investment, Publishing on x While $ 500 million is a large amount, it only equals the 10k H100 system from NVIDIA. Tesla will spend more than that on NVIDIA this year. At least at least millions of dollars in artificial intelligence schedule is several billion dollars annually in This stage.
April 30 – At the Technology Symposium in North America in TSMC, the company says that Dojo-D2 training, which completely puts Dojo tiles on one chip of silicone, instead of connecting 25 chips to make one tile-production, according to production, according What he mentioned, according to what I mentioned IEEE SICTRUM.
May 20 – Musk Notes The rear part of the Giga Texas factory extension will include the construction of a “large collection of supermarket with water”.
June 4 – A CNBC report Musk reveals the conversion of thousands of NVIDIA chips for Tesla to X and Xai. After saying at the beginning that the report was a mistake, Musk publications on x Tesla had no site to send NVIDIA chips to run, due to the continued construction of the southern extension of Giga Texas, “So they would sit in a warehouse.” He pointed out that the extension will “include 50k H100s to train FSD.”
It is also Pillars:
“Of nearly $ 10 billion in the expenses related to the prosecution, which said that Tesla will make this year, about half of which are internal, in the first place a computer and sensors of Amnesty International designed from Tesla located in all our cars, in addition to Dojo. To build intelligence training equipment Artificial, Nvidia hardware is about 2/3 of the cost.
July 1 – Musk reveals x The current Tesla compounds may not have the appropriate devices for the company’s artificial intelligence model. He says that the increase of approximately 5X in the number of parameters with artificial intelligence from the next generation “is very difficult to achieve without upgrading a car conclusion.”
NVIDIA provision challenges
July 23 – During the Tesla profit call in the second quarter, Musk says that the demand for NVIDIA devices “is very high so that it is often difficult to obtain graphics processing units.”
“I think this requires that we make more effort on Dojo in order to ensure that we have the training ability we need,” says Musk. “We see a way to compete with NVIDIA with Dojo.”
A graphic fee on the surface of the Tesla investor predicts that the Tesla Ai training capacity will descend to about 90,000 H100 from official -equivalent treatment units by the end of 2024, up from about 40,000 in June. Later that day on X, Musk functions This will this dojo 1 “about 8k H100 will be an online training equation by the end of the year.” It also publishes photo From a super computer, which seems to use the same refrigerator’s stainless steel like Tesla’s Cybletructs.
From Dogo to the shell
July 30 – AI5 is about 18 months from large size production, says Musk in A. to reply To a post from a person who claims to start a club from “Tesla Hw4/AI4 owners angry at them when AI5 comes out.”
August 3 – Musk publications on x He did a wandering in the “Tesla SuperCOCOPTE group in Giga Texas (also known as Cortex). It indicates that it will be manufactured about 100,000 H100/H200 Nvidia GPU with “tremendous video storage from FSD and Optimus.”
August 26 – Musk publications on x A video clip on Cortex, which is referred to as the new giant “Techluster Supercluster that is built in Tesla HQ in Austin to solve artificial intelligence in the real world.”
2025
There are no updates on Dogo in 2025
January 29 – It included calling in the profits in Tesla and the entire profit call to the entire profit. However, Cortex, Tesla New Ai Traaining Supercluster in Austin Gigavakteri, did not appear. I noticed Tesla in it The roof of the shareholders It has completed the spread of dandruff, which consists of about 50,000 H100 Nvidia graphics processing units.
“Cortex helped in enabling the V13 of FSD (under supervision), which features significant improvements in safety and comfort thanks to an increase of 4.2x in data, high -resolution video inputs … among other improvements,” according to the message.
During the call, CFO VAIBHAV TANEJA indicated that Tesla accelerated the creation of Cortex to accelerate the FSD V13 show. He said that the capitalist expenditures associated with Amnesty International, including infrastructure, “so far were about $ 5 billion.” In 2025, Taneja said it is expected to be flat PAPEX in terms of his behavior.
This story was originally published on August 10, 2024, and we will update it while developing new information.