Another Tesla boss has left.
Ganesh Venkataramanan, the head of its proud supercomputer Dojo project, has resigned.
Advertisement
Venkataramanan’s title is Tesla’s senior director of autonomous driving hardware and can report directly to Musk.
There are currently different opinions on the reasons for his resignation, but what is certain is that on Wednesday, Venkataramanan’s name could no longer be found in Tesla’s internal organizational directory.
What’s even more interesting is that Musk and Tesla have not responded to the resignation of such a high-level figure.
Indian technology executive
Advertisement
Venkataramanan may have been fired.
According to a report by Bloomberg, his resignation was revealed last month, but it has only been made public now.
According to another overseas media, Electrek, as early as half a month ago, insiders broke the news that Venkataramanan’s dismissal from Tesla was related to the next-generation supercomputer project, but the authenticity could not be determined at the time.
Now that Venkataramana’s resignation has become a fact, Electrek reported that insiders revealed to them that the specific reason for the executive’s departure from Tesla was that the second-generation “Dojo 2” chip did not meet expectations, which affected his job.
The media also stated that there are some major problems with the Dojo project, which is why Tesla will make major personnel adjustments at the top and give up talents like Venkataramanan, and he is not the only one who has been implicated in the second-generation chip problem. people.
However, the above statement has not yet received an official response.
Venkataramanan is well-known in the industry for his first appearance at Tesla. When the first-generation Dojo supercomputer was released on AI Day in 2021, it was he who walked into the venue holding the first-generation Dojo D1 chip.
He is one of the few Indian engineers at Tesla. He studied electrical engineering at the University of Bombay and obtained a bachelor’s degree. He later went on to pursue a postgraduate course in integrated electronics at the Indian Institute of Technology Delhi.
He joined Tesla in 2016 and initially served as director of autonomous driving hardware. In fact, before he joined Tesla, he was already a well-known technology expert in the industry.
Before joining Tesla, Venkataramanan worked in AMD’s CPU division and worked for the company for nearly 15 years. He initially joined as a Senior Design Engineer and worked his way up to Senior Design Engineering Director, managing a team of more than 200 highly skilled engineers.
He was also a core member of the team that developed several generations of x86 and ARM cores. The industry’s first x86-64 chip, the first dual-core x86 and Zen core were completed by him and his team. Prior to this, Venkataramanan also worked with Analog Devices and Hexaware Technologies.
Later, he joined Tesla in March 2016. It was also that year that Tesla established an artificial intelligence hardware and chip team. As of his resignation, he had worked at Tesla for 7 years and 10 months.
In 2018, Musk first proposed the concept of supercomputer Dojo internally. At that time, Venkataramanan was still serving as the director of autonomous driving hardware. Due to his excellent technology, he was promoted by Musk to senior director of autonomous driving hardware that year and began to lead Dojo. project.
Facts have proven that no matter how good your resume is, if it fails to meet Boss Ma’s expectations, you will be dismissed.
Former Apple executive takes over
Tesla has now found a successor for the Dojo supercomputing project.
Peter Bannon is also considered an old man at Tesla. He joined Tesla after leaving Apple in 2015. Peter has been working at Tesla for nearly 8 years, one year longer than Venkataramanan.
Before taking over from Venkataramana in charge of the supercomputing project, Peter Bannon had been Tesla’s vice president of low-voltage electronics & silicon engineering, leading the team to design a fully autonomous driving on-board computer, which has been used in Tesla cars.
Someone once asked Peter why Tesla, a car manufacturing company, wants to develop a supercomputing cluster for deep neural network training?
Peter believes this question fundamentally misunderstands the nature of Tesla. He said that Tesla is essentially a hardcore technology company. “People across the company are working hard in science and engineering to advance fundamental understanding and methods we can use to build cars, energy solutions, robots and anything else we can use to improve the human condition around the world.”
Peter graduated from the University of Massachusetts Amherst in 1984 with a bachelor’s degree in electrical and computer engineering.
Before joining Tesla, Peter was the chief architect of the first 32b ARM processor on the iPhone 5, and formed the team that built the first 64b ARM processor on the iPhone 5s. In addition, he was also the architect of the Apple A5 chip. Head of R&D, and stayed at Apple until the A9 era.
Before joining Apple, Pete was the Vice Chairman of the Architecture and Verification Department at PA Semi.
With many years of work experience and rich working experience in major manufacturers, it is natural for Peter Bannon to take over the Dojo supercomputing project.
The Dojo supercomputer project has always been part of Tesla’s plan to advance autonomous driving. Dojo can be seen as a powerful computing center used to handle a large number of artificial intelligence tasks, especially in Tesla’s Autopilot and fully autonomous driving. (FSD) system, and also provides computing power support for Tesla’s humanoid robot Optimus.
In 2019, Musk shared the supercomputer plan for the first time. The D1 chip designed by Venkataramanan, Bannon and others to drive Dojo was officially released at Tesla AI Day in 2021.
Musk has said Dojo is designed to handle large amounts of data, including videos of Tesla cars driving, needed to create self-driving software.
In other words, the Dojo project is a supercomputer platform customized by Tesla for artificial intelligence machine learning and helping to train its autonomous driving vision technology.
Dojo is responsible for processing the data captured by the vehicle, which is then transferred and used to learn the vehicle’s behavior and tendencies, essentially training the vehicle to be ready to navigate in any situation. At present, Dojo has deployed 18,000 nodes, with a total computing power of more than 1.8 exaFLOPS. The goal for 24Q4 is that Dojo’s computing power reaches 100EFlops (approximately 91 clusters).
Tesla’s plan is to use neural networks to store data and then build the world’s safest and most accurate self-driving cars.
Tesla also plans to invest more than $1 billion in Dojo supercomputing by 2024.
Although Tesla has yet to respond to the change of the leader of the Dojo supercomputing project, many people have already thought of the departure of Andrej Karpathy, director of Tesla’s artificial intelligence (AI) and autonomous driving department last year.
These big bosses have resigned one after another, which means that Tesla’s work in autonomous driving, artificial intelligence and other aspects is not so easy to advance.
The foundation of an AI empire
Before Dojo was launched, Tesla used a large supercomputer based on NVIDIA GPUs. This supercomputer was already one of the most powerful supercomputing clusters in the world. Tesla said it was ranked fifth in the world at the time. supercomputer.
But as Tesla’s data volume increased exponentially, the number of GPUs it required also increased rapidly. Since it did not have its own chips, Tesla also frantically purchased Nvidia’s GPUs during that period.
Nvidia pays hundreds of millions of dollars in royalties every year, and even Musk himself publicly complained on Twitter (X) that Nvidia’s GPUs are too expensive.
Tesla Dojo is the cornerstone of Musk’s “AI empire”. It can not only help train Tesla’s autonomous driving functions and robots, but also get rid of its own dependence on Nvidia GPUs.
In July this year, Tesla began production of the Dojo supercomputer, which is used to train artificial intelligence (AI) models for self-driving cars.
However, Dojo also needs continuous iteration. According to Musk, the Dojo V1 version is highly optimized through large-scale video training and general AI, but Dojo V2 will solve such limitations.
This also means that the V2 version of Dojo can be optimized independently to a certain extent and become an AI for training AI. This idea is indeed crazy.
At the critical moment from V1 to V2, although Musk decided to change coaches on the spot, the Dojo supercomputing project must continue to operate.
Tesla hopes that Dojo’s computing power will be among the top five in the world by 2024. To achieve this goal, Musk will even invest US$1 billion.
At Tesla AI Day last year, Tesla also demonstrated the latest achievements of the Dojo supercomputer, which consumed so much power that the local substation tripped.
In September this year, Morgan Stanley raised Tesla’s share price to $400 in a report on the Dojo supercomputing platform. He also believes that if Dojo can promote the popularity of self-driving taxis and software services, it will increase Tesla’s market value by nearly $600 billion. This report also caused Tesla’s market value to surge by US$79.9 billion (approximately RMB 580 billion) overnight.
Whether from a capital perspective or a strategic level, Dojo is too important to Tesla. So there is nothing wrong with Musk being stricter. It just depends on whether the new person in charge can promote progress according to Lao Ma’s wishes.