In late 2017, we learned that Tesla was attempting to build its very first computer chip for self-driving cars, and Elon Musk said in October 2018 that the silicon was a mere six months away. Now — in an unusual example of Elon Musk correctly predicting when a product will launch — Tesla has revealed that the chip is actually here on schedule.
In fact, Musk says it’s been shipping its new Full Self Driving Chip in the Tesla Model S and Model X for over a month now, and has been placing it in the Model 3 for ten days already.
And today, at Tesla’s Autonomy Investor Day in Palo Alto, California, the company gave the world its first, detailed glimpse at what Musk is now calling “the best chip in the world” — a 260 square millimeter piece of silicon, with 6 billion transistors, that the company claims offers 21 times the performance of the Nvidia chips it was using before.
There’s an awful lot of specs associated with this new Tesla-designed, Samsung-manufactured silicon — you can peruse them at your leisure in our gallery above — but the overall message Tesla’s trying to send today is that this hardware is purpose-built to handle all of the data from the car’s sensors far faster and more efficiently than the AI chips it could buy off the shelf. There are plenty of bigger, more powerful processors out in the world, and plenty of smaller ones as well, and Tesla itself points out the general-purpose CPU and GPU components here are nothing particularly special. It isn’t overengineered, Tesla’s basically arguing.
Notably, Tesla says this silicon, with its twin neural network arrays capable of 36 trillion operations per second (each), will only cost the company 80 percent of what it was paying before for that 21x performance gain, and draw little enough additional wattage (72W, vs. 57W) that it can continue to promise the same range out of each car and without impacting the cost.
Each of Tesla’s new Full Self Driving Computer boards actually come with two of the new chips for redundancy, and it’s just one of the many redundant features you’d hope and expect to find in situations where you’ll be trusting your life to a computer — you’ll also find redundant power, and even redundant calculations where the system compares results from both processors before it steers the car. “Any part of this could fail, and the car will keep driving,” says Musk. “The probability of this computer failing is substantially lower than someone losing consciousness — at least an order of magnitude.”
Now, Musk says, every car Tesla is producing will have the hardware it needs for full self-driving capability. “All you need to do is improve the software” from this point on, he says.
You can watch the presentation for yourself right here at YouTube.