TSMC has begun production of Tesla's Dojo chips, with a 40-fold increase in computing power in three years.

Tesla's path to autonomous driving requires sufficient computing power to support it. TSMC has confirmed that the production of Tesla's next-generation Dojo supercomputer platform training chips has commenced, with a significant leap in computing power expected by 2027.

Elon Musk's grand plan involves more than just selling cars; software is the key to real profit. To achieve this goal, not only is high-efficiency computation required, but also powerful computing power. In terms of hardware, Tesla has adopted a dual approach, on one hand purchasing tens of thousands of NVIDIA H100 GPUs, and on the other hand, developing its own supercomputing chip, Dojo. TSMC, responsible for producing the supercomputer chips, has now confirmed the start of production.

Advertisement

"Production of Tesla's next-generation Dojo training modules has begun, and by 2027, we will provide more complex wafer-level systems with computing power more than 40 times that of existing systems," detailed TSMC's North American Technology Symposium, which introduced semiconductor technology and advanced packaging, enabling the establishment of systems on a full wafer scale and creating ultra-high computing performance.

Wafer integration can provide 40 times the computing power.

The core of Tesla's designed Dojo supercomputer lies in the training module, which arranges 25 D1 chips in a 5x5 matrix. The chips use a 7-nanometer process, can accommodate 50 billion transistors, and provide 362 TFLOPS of computing power. Most importantly, they have the capability to scale up, allowing for continuous stacking and allocation of computing power and power consumption according to software requirements. This is just the version of Tesla in use.

According to TSMC, the new product for Tesla is different from the wafer-level system supplied to Cerebras. In simple terms, the Dojo training module (5x5 processor matrix) is placed on a carrier wafer, filling all the empty spaces, and then covered with a high-density interconnect using TSMC's fan-out (InFO) technology, which maximizes the data bandwidth between chips, making them operate like a giant chip.

By 2027, TSMC plans to integrate these wafer-level systems with CoWoS advanced packaging technology into wafers (SoIC), at which point the complete wafer will provide 40 times the computing power, exceeding 40 photomask silicon and up to 60 high-bandwidth memory chips.According to IEEE Spectrum, the InFO_SoW technology aims to achieve high-performance connections, allowing 25 chips of Tesla's Dojo to operate as if they were a single processor; at the same time, to maintain consistency in the wafer-level processor, TSMC filled the gaps between the chips with virtual chips.

The Tesla wafer-level Dojo processor actually includes 25 ultra-high-performance processors, which consume a significant amount of power, thus requiring a complex cooling system.

To meet the power supply needs of the Dojo processor, Tesla uses a complex voltage regulation module to provide 18,000 amperes of electricity to the computing plane, generating up to 15,000W of heat, hence necessitating water cooling for heat dissipation.

Tesla has not yet disclosed the performance of its Dojo wafer system — however, considering all the challenges faced during its development, it seems poised to become a very powerful solution for artificial intelligence training.

According to Musk, if NVIDIA provided enough GPUs, Tesla might not need to develop the Dojo on its own. It is preliminarily estimated that this generation of the Dojo supercomputer will become part of Tesla's new Dojo cluster, located in New York, with an investment of at least $500 million.

Despite the substantial computational power invested, Tesla's AI endeavors are still full of challenges. In December last year, two senior engineers in charge of the Dojo project left the company, and now Tesla is continuously laying off staff to save costs. More outstanding talents are needed to dedicate their brains and liver to have a chance to launch the self-driving taxi on schedule and to further enhance the Full Self-Driving (FSD) system.

Tesla's next-generation Dojo computer will be located in New York, while the headquarters in the Texas Gigafactory will build a 100MW-scale data center to train self-driving software, with hardware supplied by NVIDIA. Regardless of the location, these chips all lead to the same end, essentially produced by TSMC, and it is not an exaggeration to call them the AI driving force.Tesla Chairman Urges Shareholders to Reapprove Musk's Massive Pay Package

In a video released on Friday, Tesla's board chair, Robyn Denholm, urged shareholders to reapprove CEO Elon Musk's 2018 compensation package, which a Delaware judge overturned in court. Musk's pay package was once valued at up to $56 billion, but due to Tesla's stock price decline, it is now worth about $47 billion.

Denholm stated in the video that a shareholder vote to reinstate Musk's compensation would be "very important for the company's future."

The chairperson said to investors: "We believe that the opinion of one judge should not invalidate the will of millions of votes by all company shareholders. Therefore, we once again ask you to make your voice heard by voting to approve the 2018 performance award."

Denholm argued in the video that it is fair for Musk to receive this reward. Shareholders set this bonus to motivate the tech billionaire to achieve "ambitious financial and operational goals."

She pointed out that under Musk's leadership, Tesla's revenue grew from $11.8 billion to $96.8 billion, and turned a loss of $2.2 billion into a profit of $15 billion.

These milestone events have increased Tesla's value from $53.7 billion to over $790 billion, she said.

"His success is shared by all shareholders," Denholm said of Musk, adding that investors have seen their stock value increase by 1100%.

TSMC Begins Production of Tesla Dojo Chips, with a 40-Fold Increase in Computing Power in Three Years.

Tesla's path to autonomous driving must be supported by sufficient computing power. TSMC has confirmed that the production of Tesla's next-generation Dojo supercomputer platform training chips has begun, with a significant leap in computing power expected by 2027.Elon Musk's grand plan is not just about selling cars; the real profit lies in software. To achieve this goal, not only is a highly efficient algorithm required, but also powerful computing power. In terms of hardware, Tesla is taking a dual approach by stacking resources. On one hand, they are purchasing tens of thousands of NVIDIA H100 GPUs, and on the other hand, they are developing their own supercomputing chip, Dojo. TSMC, which is responsible for producing supercomputer chips, has confirmed that the product has begun production.

"Tesla's next-generation Dojo training module has started production, and by 2027, we will offer more complex wafer-level systems with a computing power more than 40 times that of existing systems." TSMC's North American Technology Symposium provided a detailed introduction to semiconductor technology and advanced packaging, allowing the construction of systems on the scale of an entire wafer, creating ultra-high computing efficiency.

Wafer integration can provide 40 times the computing power.

The core of Tesla's Dojo supercomputer design lies in the training module, which arranges 25 D1 chips in a 5x5 matrix. The chips use a 7-nanometer process and can accommodate 50 billion transistors, providing 362 TFLOPS of computing power. Most importantly, it has the capability to scale up, allowing for continuous stacking and allocation of computing power and power consumption according to software needs. This is just the version of Tesla's Dojo in use.

According to TSMC, the new product for Tesla is different from the wafer-level system supplied to Cerebras. Simply put, the Dojo training module (5x5 processor matrix) is placed on a carrier wafer, and after filling all the blanks, TSMC's fan-out (InFO) technology is used to add a layer of high-density interconnect, thus maximizing the data bandwidth between chips and making them operate like a giant chip.

By 2027, TSMC plans to integrate these wafer-level systems with CoWoS advanced packaging technology into a wafer (SoIC), at which point the entire wafer will provide 40 times the computing power, more than 40 photomask silicon, and up to 60 high-bandwidth memory chips.

According to IEEE Spectrum, the InFO_SoW technology aims to achieve high-performance connections, allowing Tesla's Dojo's 25 chips to work like one processor; at the same time, to keep the wafer-level processor consistent, TSMC filled the gaps between the chips with virtual chips.

Tesla's wafer-level Dojo processor actually includes 25 ultra-high-performance processors, which consume a lot of power, thus requiring a complex cooling system.

To meet the power supply needs of the Dojo processor, Tesla uses a complex voltage regulation module to provide 18,000 amperes of electricity to the computing plane, generating up to 15,000W of heat, which requires water cooling.Tesla has not yet disclosed the performance of its Dojo chip system — however, considering all the challenges faced during its development, it seems poised to become a very powerful solution for artificial intelligence training.

According to Musk, if NVIDIA provided enough GPUs, Tesla might not have needed to develop the Dojo on its own. It is initially estimated that this generation of the Dojo supercomputer will become part of Tesla's new Dojo cluster, located in New York, with an investment of at least $500 million.

Despite the significant computational power invested, Tesla's AI endeavors are still full of challenges. In December last year, two senior engineers in charge of the Dojo project left the company, and now Tesla is continuously laying off employees to save costs. More top talent is needed to contribute their brains and liver to have a chance of launching the self-driving taxi on schedule and to improve the Full Self-Driving (FSD) system to the next level.

Tesla's next-generation Dojo computer will be located in New York, while the headquarters in Texas will build a 100MW-scale data center to train self-driving software, with hardware supplied by NVIDIA. Regardless of the location, these chips all come from TSMC, and it is not an exaggeration to call them the AI pushers.

Tesla's Chairman Urges Shareholders to Reapprove Musk's High-Priced Compensation Package

In a video released on Friday, Tesla's board chairman, Robyn Denholm, urged shareholders to reapprove CEO Elon Musk's 2018 compensation package, which was overturned by a judge in Delaware. Musk's compensation package was once valued at up to $56 billion, but due to Tesla's stock price decline, it is now worth about $47 billion.

Denholm stated in the video that a shareholder vote to reinstate Musk's compensation would be "very important for the company's future."

The chairman said to investors: "We believe that the opinion of one judge should not invalidate the will of millions of votes by all company shareholders. Therefore, we ask you once again to make your voice heard by voting to approve the 2018 performance award."

Denholm argued in the video that it is fair for Musk to receive this award. Shareholders set this bonus to motivate the tech billionaire to achieve "ambitious financial and operational goals."She pointed out that under Musk's leadership, Tesla's revenue grew from $11.8 billion to $96.8 billion, and turned a loss of $2.2 billion into a profit of $15 billion.

She said that these milestone events increased Tesla's value from $53.7 billion to over $790 billion.

"His success is shared by all shareholders," said Dan Holm when talking about Musk, adding that investors have seen their stock value increase by 1100%.