Tesla recently unveiled a custom chip designed to train artificial intelligence (AI) networks in data centers. “We should have Dojo operational next year,” CEO Elon Musk said.
With the chips, Tesla vehicle cameras can collect video feeds which can be used to train models for recognizing a broad range of objects. A model training program requires a great deal of computing work. One employee of Tesla said the new chip would provide the fastest training computer for artificial intelligence.
Tesla began producing vehicles that use AI chips its engineers built in-house two years ago. In response to what is happening on the road, the chip helps the car’s onboard software make very fast decisions.
Ganesh Venkataramanan, senior director of Autopilot hardware, said, “The D1 chip, part of Tesla’s Dojo supercomputer system, uses a 7-nanometer manufacturing process, with 362 teraflops of processing power. 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. We are assembling our first cabinets pretty soon.”
He further 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.”
Tesla currently offers something called “Full Self-Driving Capability” as an add-on for their vehicles. By upgrading the $10,000 package, the driver can program the car to change lanes, navigate on highways, enter parking spots, and emerge from spots automatically. It has previously been promised by Tesla that this feature would be available in 2019; however, the Tesla website now says the Tesla will be able to e-steer on city streets later this year.
