Nvidia issued "Thor's Hammer" Smart Motor Central Calculation Times Breaking Dawn

21st Century Business Herald reporter Zhao Yunfan Shanghai report

On September 21, at the Nvidia GTC 2022 press conference, the company's founder Huang Renxun released the central computing chip Drive Thor, saying that it was a chip born for autonomous driving.

It is reported that Huang Renxun said at the press conference that Thor said that the chip will be mainly applied to the central computing architecture and use the chip to create a controller to simultaneously use multiple systems such as ADSD, autonomous driving, motor, cockpit and other systems at the same time. Domain provides computing power.

In addition, the company also revealed that the "Thunder" is expected to mass -produced vehicles in 2025, and the first car brand equipped with the chip will be a Geely Motor's Gee Wan brand model.

In this regard, some analysts told the 21st Century Business Herald reporter that Nvidia's "Thor" chip further proves the trend of the evolution of the vehicle's electronic architecture towards the domain, but the appearance of this chip is unlikely to change the value of the autonomous driving industry in the short term. pattern.

The first smart car central computing chip

Gaming chips, smart AI computing chips, and smart cars, Nvidia's new architecture chips are constantly landing in new business scenarios. The GTC in 2022 was even considered the highest conference of Nvidia's "car content".

According to Nvidia, the "Thor" single chip has a built -in 77 billion transistor, and the single -core computing power can reach 2000TOPS. Last year, the single computing power of Altan chip released by Nvidia was 1,000 times, and the single -driving chip Orinx, which has been put in large amount, was around 250TOPS. The computing power of "Thor" was 2 times and 8 times that of the first two.

And because the computing power of this chip is too strong, Nvidia even gave up the landing of Altan chips released last year in commercial scenarios -even though the chip has increased by more than 4 times the increase in the commercialized ORINX series chips.

In addition to the strong computing power, it has made the industry's attention to its cross -domain capabilities. Huang Renxun introduced that the new Thor chip can not only use all the computing power for autonomous driving workflows; some can also be used for cockpit AI and information and entertainment systems, and the other part is used for autonomous driving.

This means that the central computing architecture of the Smart Motors first appears fist products in the chip link for the first time.

Chen Meng, the manager of the Western Lide Fund, was an engineer of the Shanghai R & D Center of Nvidia Semiconductor, told the 21st Century Business Herald reporter: "Thor is positioned by the" central computing brain ', that is, the later form of the development of the E/E architecture. The "autonomous driving chip" of several generations of products is different. Nvida clarifies that Thor is constructed for the central computing architecture, for the "domain fusion" -The is mainly the fusion of the driver and the cockpit field. For L2+/L3/L4, the ORINX scheme is currently sufficient, and some computing power should be divided into smart cockpit fields at the same time, which will have higher requirements for the computing power of single chips. "

In terms of technological implementation, Chen Meng introduced to reporters that Thor can separate some computing power for smart cockpit. After virtualization, Thor can run Linux (smart driving/cockpit environment), QNX (instrument panel environment), android (IVI environment), which is equivalent to isolating different operating systems. The class function realizes the fusion of the cockpit area and the driver's region.

"In fact, due to architectural reasons, the chip utilization rate is not high. The bottleneck depends on the memory call speed. Although the theoretical peak is becoming larger, the computing unit of an instantaneous response may not reach its peak. It also involves software matching and adjustment. The issue of excellent and proportion. In the future, it is necessary to effectively use more computing power to improve comprehensive improvement, and it also needs to be improved for power consumption issues. "Chen Meng said.

It takes time to land

In terms of hosting manufacturers, Geely Automobile's extremely unexpectedly became the first model of this Thor chip -even two years before Thor's mass production.

It is worth noting that some Nvidia's AI chips were restricted to China by the US authorities before. Some analysts believed that the acquisition of vehicle smart chips with high computing power architecture will also face tests.

Getting the thunder setting point to a certain extent relieves the doubts of the car chip acquisition. Huang Renxun also stated in the press conference that the chip model that restricted exports will be partially adjusted to make its architecture conveniently export.

On the other hand, a car analyst at a state -owned investment company told reporters that although it has not yet been made public, the cost cost of the system equipped with Thor products must be very high: "Considering that the price of Orinx is already above $ 400, car host manufacturers have used it to use it. The chip still needs to consider the ability to bear terminal consumption. "

Earlier, the Jiyang 001 model, which has already been listed, has its autonomous driving computing power. ; January this year, Ji Yan announced deepening cooperation with Mobileye, focusing on the L4 -level autonomous driving system products released in 2024.

In addition to the extreme, Nvidia has completed the commercialized ORINX chip, and has used the domestic electronic integrated manufacturer Desesiwei (002920.SZ) to integrate the autonomous driver's domain controller, and it is equipped with Weilai ET7, ideal L9, Xiaopeng G9 G9 , High -end models such as Smart L7 and other domestic vehicles. And because high -level autonomous driving is indispensable for computing power, connected, and perceived high -end facilities, the market volume of these models can often be not too high. In addition, some people in the industry told reporters that the rhythm of core chip controller development is often very slow, so the release of a new chip in the short term will not affect the current industrial structure.

Chen Meng also told reporters: "Nvidia's two -area fusion chip released this time is expected to be produced in the second half of 2024 and began mass production in 2025. At the same time, it is expected that Qualcomm will also release the two -field fusion chips to catch up. It will be completed in 24 years. Based on this, the high -calculation of the fusion chip is aimed at the models of at least 25 years. In the future, the car manufacturers may be able to see further landing. High computing chips will at least target models above L3 or above in the future. Considering the factors of cost, mass production, and product positioning, it is expected to be the flagship model of the car company's head, but the number is not large, but the value will increase. "

"For the autonomous driving industry, in the long run, the entire E/E architecture has evolved in the direction of the domain. The chip platform solves the problem of high computing power demand and will promote the development of the industry to a certain extent. Of course, the autonomous driving industry The development is not determined by a single factor. It also involves other factors such as laws and regulations, the integration of V2X industry, and the division of labor and game of the vehicle industry chain. "Chen Meng said.

Desaisiwei may still cooperate with Nvidia

Nvidia chips have been released. Who will become a partner of Nvidia's car central computing platform?

It is worth noting that Deyesiwei released the first car central computing architecture platform ICP Aurora earlier this year. The company pointed out that ICP Aurora revolves around the central computing concept and builds a multi -dimensional technology core for future evolution. The software integrates the core functional domains such as smart cockpits, smart driving, and networking services. Use to increase business processing density.

Earlier, the Dedsexi IPU03, IPU04 autonomous driver domain controller has been equipped with Nvidia's two -car chip Xavier and Orinx, respectively. The company was once the only domain controller developer in China that uses Nvidia Orin platforms.

What will happen when "Thor" encounters "Aurora"?

After Nvidia released Thor, the Science and Technology Team of CICC was the first to believe that there was a greater probability of cooperation with the Nvideda God of Desesi.

CICC pointed out: "Deyesiwei's cooperation with Nvidia is deep and is expected to be the starting supplier. Desesixway's IPU03 domain controller based on Nvidia's Xavier chip has been equipped with Xiaopeng P7/P5 and achieved sales. Nvidia Orin's IPU04 domain controller is officially mass -produced with the listing of the ideal L9 this year, and it is expected to contribute to the company's considerable incremental income to the company with the successive release of Xiaopeng G9. THOR chip continues to deepen cooperation.

- END -