Electric vehicle giant Tesla’s next-gen Dojo Artificial Intelligence (AI) training tile has finally entered the production line, IEEE Spectrum reports, citing supplier TSMC’s recent announcement.
Next-gen Dojo chip enters production
Tesla supplier Taiwan Semiconductor Manufacturing Company Limited (TSMC) discussed its semiconductor and chip-packaging technology road maps during the North American Technology Symposium on Wednesday.
The company’s chip-packaging tech can accelerate the shift to processors made from more silicon, leading to systems with the size of a full silicon wafer.
Tesla’s version of this tech, the next-gen Dojo training tile, has finally entered production. Interestingly, TSMC also revealed that they are now developing new technology that could enable greater power to Dojo in 2027.
“Tesla’s next generation Dojo training tile is already in production, TSMC says. And in 2027 the foundry plans to offer technology for more complex wafer-scale systems than Tesla’s that could deliver 40 times as much computing power as today’s systems.”
IEEE Spectrum
Tesla’s AI push
Tesla has significantly invested in AI training compute power by acquiring NVIDIA’s hardware and producing its own units under the Dojo program.
Tesla officially launched the operation of its first-gen Dojo supercomputing platform in the summer of last year. The Musk-led company then solidified its partnership with TSMC.
TSMC currently produces Tesla’s Dojo chip to intensify its AI push.
Now, the next-gen AI training tile that recently entered production will likely aid Tesla’s planned $500 million Dojo cluster in New York.
In addition, Tesla is also developing a new 100 MW data hub to train its autonomous driving AI at Giga Texas. However, it is expected to employ NVIDIA’s hardware rather than TSMC’s.
About Dojo
Dojo is a supercomputer that Tesla developed for computer vision, video processing, and recognition.
Tesla plans to use Dojo to train its machine learning models, which could improve its Full Self-Driving (FSD) suite.
Dojo primarily aims to process millions of terabytes of video data accumulated by Tesla’s own fleet. The video data comes from real-life driving conditions, indicating a significant advancement from conventional supercomputer designs.
