Tesla unveils chip to train AI models
Image credit: Dreamstime
Tesla has unveiled its own computer chips that will train its automated driving system.
Revealed during the automaker’s AI Day event, Tesla’s custom chip will show the firm's continuing 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 onto a single “training tile”, and 120 of these tiles come together across several server cabinets, amounting to over an exaflop of power, he added.
He said the Tesla technology will be the fastest AI-training computer. Chipmaker Intel, graphics card maker Nvidia, and start-up Graphcore are among the companies that make chips that companies can use to train AI models.
Venkataramanan added the chips can help train models for recognising a variety of items from video feeds collected by cameras inside Tesla vehicles. Model training requires extensive computing work.
The work comes two years after Tesla began producing vehicles containing AI chips it built in-house. Such chips help the car’s on-board software decide quickly in response to what’s happening on the road.
According to Tesla’s CEO Elon Musk, the Dojo chip should be operational by next year.
Speaking at the AI Day event, Musk also said the automaker will probably launch a ‘Tesla Bot’ robot prototype next year, designed for dangerous, repetitive, or boring work that people don’t like to do.
The entrepreneur said the robot, which stands around 1.76m tall, could handle jobs from attaching bolts to cars with a wrench or picking up groceries at stores.
According to Musk, the robot would have “profound implications for the economy”, addressing a labour shortage. He also said it was important to make the machine not “super-expensive”.
The AI Day event came amid growing scrutiny over the safety and capability of Tesla’s 'Full Self-Driving' advanced driver assistant system.
Musk didn’t comment on that scrutiny, but said that he was confident of achieving full self-driving with higher safety than humans using current in-car cameras and computers.
Earlier this week, US safety regulators opened up an investigation into Tesla’s driver assistant system because of accidents where Tesla cars crashed into stationary police cars and fire trucks.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.