Tesla is at the forefront of AI technology. The company recently presented its D1 Dojo custom application-specific integrated circuit (ASIC) for training artificial intelligence software. It could be a game-changer in speeding up workloads around the world. The D1 chip is part of the Tesla Dojo supercomputer chip.
There is a great deal of competition in the AI hardware market. From countless start-ups to big players like Amazon, Baidu and Intel are building ASICs for AI workloads. But not everybody gets it right or can satisfy every need perfectly – that’s why Tesla chose to develop its chip just for training purposes.
According to Tesla Official, Tesla’s new D1 chip has been designed to be one of the most powerful chips ever with 362 teraflops processing power. The company places 25 of these chips on a single “training tile,” and 120 tiles come together across several server cabinets (a total of more than an exaflop).
The D1 chip is a product of TSMC’s commitment to innovation and forward-thinking designs. The 7nm semiconductor node leaves the processor with immense die size, packing over 50 billion transistors along with the chip boasts a huge die size of 645mm^2.
The chips can help train models for recognizing a variety of items from video feeds collected by cameras inside Tesla vehicles. This could significantly reduce the need to train a model on other data sources, often costly and time-consuming.
Tesla AI Day: https://www.youtube.com/watch?v=j0z4FweCy4M&t=6750s