So far, Tesla has been relying on the NVIDIA Drive platform. So why adjust now?
Tesla said that by designing the chip itself, the company can focus on its own needs and thus ensure efficiency.
'We know what our neural network looks like, and we know what it will look like in the future,' said Pete Bannon, director of the Hardware 3 project. Bannon pointed out that hardware upgrades will begin next year.
Maske added, 'The key is to run a neural network at the bare metal level. You have to let the circuit do the calculations yourself, not in the emulation mode. This is how the graphics processor and CPU work. You A lot of calculations are needed. '
According to Musk, the end result will be amazing: He said that Tesla computer vision software runs about 200 frames per second on NVIDIA hardware, but Tesla's specially designed chips can achieve 2000 frames per second.
Artificial intelligence analyst James Wang pointed out that this move will give Tesla more control in the future: Apple's custom SoC is the key reason for iPhone performance to stay ahead in the market. The same is true for Tesla's artificial intelligence chips. - This will help Tesla lead other automakers in terms of performance and functionality, all of which need to wait for NVIDIA or Intel's next-generation chips.
With their own chips, Tesla can meet their own needs at their own pace. If they suddenly realize that the hardware is missing, they don't have to wait for others to do research and development. This is definitely not a simple task - but If Tesla can do this successfully and the company is not bankrupt (Muske says the cost of chip development is as much as the existing hardware), this will eventually become a major advantage for the company.
As for how they want to embed the chip in the existing Tesla car, Mask said: 'The replacement of the computer is very easy, so much is done. Take the computer out, then insert a new one. All connections The devices are all adapted. '