Tesla developed its own AI chip for automatic operation, 10 times performance of NVIDIA system and Mask CEO is languages

CEO Earlon Mask revealed that Tesla of the electric vehicle (EV) manufacturer is developing its own AI chip dedicated to the automatic operation system under development. The mask CEO boasts that Tesla's proprietary AI chip, which has been rumored for a long time, boasts more than ten times the computing power of NVIDIA systems adopted by many automobile manufacturers.

Tesla says it's dumping Nvidia chips for a homebrew alternative | Ars Technica
https://arstechnica.com/cars/2018/08/tesla-says-its-dumping-nvidia-chips-for-a-homebrew-alternative/

Tesla is building its own AI chips for self-driving cars | TechCrunch
https://techcrunch.com/2018/08/01/tesla-is-building-its-own-ai-chips-for-self-driving-cars/

Machine learning (AI) system such as image recognition technology is indispensable for automatic driving car technology, and NVIDIA of GPU maker provides " NVIDIA DRIVE " for automobile using PX 200 , and develops automatic driving technology It is adopted by many automobile manufacturers. Tesla has also greatly improved the automatic driving technology "auto pilot" using the NVIDIA system, but Mask CEO officially acknowledged that he has been developing AI chips for a long time in Rwahsa Street.

Mask CEO is a place for reporting the second quarter of 2018, "Although it was a" half stealth "(public secret) mode for the past two or three years, in a world specially designed for autonomous driving "We developed the most advanced computer itself," revealed that we completed the AI ​​chip and computer system dedicated to Tesla vehicles. According to Mask CEO, the proprietary AI chip boasts overwhelming performance compared with NVIDIA hardware. The NVIDIA chip is 200 fps with image recognition performance, while the Tesla chip achieves 2000 fps. Moreover, it states that it ten times more than NVIDIA system with taking in margin to take in the mechanism of redundancy and failover.

The reason why Tesla in-house AI chips is "to obtain higher performance". By developing hardware specialized for Tesla's autopilot system rather than a versatile NVIDIA system, performance beyond conventional can be expected. According to Mask CEO, the key to performance is to run a neural network system at the bare metal level. Tesla believes that data transfer between the GPU and the CPU becomes one of the constraints of the system and it is disadvantageous for the performance improvement in the system that uses the CPU and the GPU and calculates the neural network in some kind of emulation mode.

Also, developing own hardware including AI chips has the advantage that "development speed of NVIDIA does not become a bottleneck". Many car manufacturers rely on NVIDIA for the core of the hardware of automatic driving technology development, and significant performance improvement needs to wait for the appearance of NVIDIA's new system. On the other hand, Tesla will be able to develop and introduce new technologies one after another by independently developing AI chips and computer systems without waiting for the development of NVIDIA.

It is pointed out that Tesla's own software development not only software but also hardware in a computer system is easy to understand in parallel with Apple's way of using smartphones. By independently developing A Series chips, Apple succeeded in bringing out the full performance of software (iOS), maintaining its superiority in performance competition with Android terminals that adopt general-purpose chips. As Tesla is a smartphone and Apple does it, we aim to integrate it as an advanced automatic driving system by developing our own software + hardware even in automobiles.

According to Mask CEO, our own AI chip will be released in 2019. The system including this AI chip maintains backward compatibility with the existing Tesla car's hardware, and the Tesla owner is planning to upgrade the hardware system. Tesla is planning to accelerate the practical application of level 4 or higher automatic driving technology in autopilot by AI chip developed by our company.