your current location is：Home > TechnologyHomeTechnology

　　On the evening of September 20, local time, Yingwei released the in-vehicle chip Thor at the 2022 autumn GTC conference, with a single computing power of 2000TFLOPS. Previously, NVIDIA released the self-driving chip Altan, with a single computing power of 1000TFLOPS, which was originally scheduled to be launched in 2024. After the launch of Thor, Altan will be replaced and will no longer be listed.

　　At present, NIO, Xiaopeng, Ideal and other new car manufacturers are using NVIDIA Orin X chips in their new cars, with a single computing power of 254TOPS, which is already the chip with the highest computing power among mass-produced cars. The force is 8 times higher than that of Orin. Jikr has announced that it will be equipped with Thor for the next generation of smart electric vehicles, and production will begin in early 2025. At present, the self-driving chip equipped with the mass-produced model Jikr 001 is Mobileye EyeQ5, with a single computing power of 24TOPS.

　　Some autonomous driving practitioners once told the first financial reporter that the computing power of about 500TOPS can already support the computing power requirements of L3~L4 autonomous driving. The recently launched Ideal L9 and Xiaopeng G9 models are claimed to have point-to-point advanced For driving assistance capabilities, both products use two Orin X chips, and the total computing power is 508TOPS.

　　Do smart cars need 2000TFLOPS of top computing power? According to reports, Nvidia founder Huang Renxun made it clear that this SoC chip was born for the central computing architecture of the car. Using this chip to create a controller can simultaneously provide automatic parking, intelligent driving, car machines, and dashboards. , driver monitoring and other systems to provide computing power.

　　According to reports, Thor can use all of its 2000TOPS for autonomous driving workflows, or it can be allocated for use, part of which is used for smart cockpits and part for assisted driving. In other words, Thor is not only an upgraded version of the Orin X autopilot chip, but also can replace the work of Qualcomm 8155, 8295 and other cockpit chips.

　　At present, the electronic and electrical architecture is changing from the traditional distributed to the domain-centralized architecture, and is developing towards the third-generation central electronic and electrical architecture. The industry generally believes that in the past, hundreds of thousands of low-computing MCUs with traditional distributed architectures could not meet the needs of smart cars. At present, they have been replaced by several large-computing ECUs and become a domain-centralized architecture. However, the electrical and electronic architecture will eventually will enter the era of central computing.

　　In mid-September, Aptiv announced that it has launched the world's first vehicle central computing platform, which processes the communication signals of hundreds of components in the vehicle and provides the hardware platform and software platform foundation of the vehicle for software-defined vehicles. In addition, the upcoming Xiaopeng G9 has also announced that it will adopt the integrated hardware architecture of "central supercomputing + regional control".

　　Some analysts believe that NVIDIA's launch of Thor is not only to refresh the "ceiling" of the computing power of autonomous driving chips, but also to lay a first-mover advantage for the subsequent launch of the central computing architecture.

　　However, some car companies are still insisting on self-developed chips. Tesla Model 3, Model Y and other models have adopted self-developed FSD chips. At the same time, Tesla also launched a self-developed FSD chip on AI Day in 2021. Dojo made by researching the D1 chip. According to recent reports, Cruise, an autonomous driving company owned by General Motors, is developing four chips at the same time.

　　The head of the research and development of an autonomous driving company told reporters that the computing power of chips is very important, but on general platforms such as Nvidia and Mobileye, the relevant OEMs and autonomous driving companies do not necessarily have the technical level to make the hardware play to 100%. Self-developed chips can better integrate and adjust software and hardware in the research and development stage, and give full play to the potential of software and hardware.

　　"Like Apple, it uses self-developed A-series chips and matches the IOS system. Compared with many other mobile phones, the software fluency and hardware energy consumption of Apple mobile phones are relatively better. Tesla has opened up in the United States. The point-to-point driving assistance function, but the computing power of FSD is 144TPOS, which is much lower than that of many domestic products with heap computing power.” said the person in charge.