Tesla has unveiled its own computer chips that will train its automated driving system.
Revealed at the automaker’s AI Day event, Tesla’s custom chip will showcase the company’s ongoing pursuit of vertical integration, according to senior director of Autopilot hardware at Tesla, Ganesh Venkataramanan.
The D1 chip, part of Tesla’s Dojo supercomputer system, uses a 7nm manufacturing process, with 362 teraflops of processing power, Venkataramanan explained. Tesla places 25 of these chips on a single “training tile,” and 120 of these tiles converge across several server cabinets, amounting to more than an exaflop of power, he added.
He said Tesla technology will be the fastest AI training computer. Chipmaker Intel, graphics card maker Nvidia and start-up Graphcore are among the companies making chips that allow companies to train AI models.
Venkataramanan added that the chips could help train models to recognize a variety of items from video feeds collected by cameras in Tesla vehicles. Model training requires a lot of math.
The work comes two years after Tesla began producing vehicles with AI chips it built in-house. Such chips help the car’s on-board software to make quick decisions in response to what’s happening on the road.
According to Tesla CEO Elon Musk, the Dojo chip should be operational next year.
Speaking at the AI Day event, Musk also said the automaker will likely launch a “Tesla Bot” robot prototype next year, designed for dangerous, repetitive or boring work that people don’t like doing.
The entrepreneur said the robot, which is about 1.76 m high, can handle tasks from fastening bolts to cars with a key or picking up groceries from stores.
According to Musk, the robot would have “major consequences for the economy”, addressing a labor shortage. He also said it was important not to make the machine “super expensive”.
The AI Day event took place amid growing research into the safety and capabilities of Tesla’s ‘Full Self-Driving’ advanced driver assistance system.
Musk didn’t comment on that study, but said he was confident he could become fully self-driving with a higher level of safety than people using current in-car cameras and computers.
Earlier this week, US safety authorities opened an investigation into Tesla’s driver assistance system over accidents in which Tesla cars collided with stationary police cars and fire engines.