Nvidia can't rely on the "Thor" chip to realize the "mechanical seance" of autonomous driving

time:2022-09-27 06:41:18 source:scripttoolbox.com author:Power Supply
Nvidia can't rely on the "Thor" chip to realize the "mechanical seance" of autonomous driving

Finally, the scolding RTX 40 series graphics cards are here. As NVIDIA's latest masterpiece, its high performance has caused dissatisfaction among consumers. However, in the comment area of ​​this article, please don't scold Huang Renxun in a hurry, because today's protagonist is not the graphics card, but the on-board chip - Thor. NVIDIA entered the field of intelligent driving very early, and previously launched Orin with a computing power of up to 275TFLOPS. According to NVIDIA's original plan, Altan, an iterative product with 1000 TFLOPS of computing power, will be mass-produced from 2024 to 2025, but now NVIDIA can't wait, launching Thor, and Altan has been cut off. Source: pixabay In terms of parameters, Thor is very strong, with a full 77 billion transistors and a computing power of up to 2000 TFOPS. If you don’t know what this computing power represents, Xiaolei, for example, Tesla’s Model S, which can run 3A games, has an on-board AMD chip with a computing power of 10TFLOPS, a gap of 200 times. It is not that strong computing power means stronger performance, but a 200-fold gap is enough to smooth out many problems in software optimization and adaptation. Of course, such a powerful computing power is not only used to support system operation, nor is it only used for intelligent driving. Thor is a multi-functional vehicle chip that can simultaneously shoulder the performance requirements of intelligent driving, entertainment and other aspects, and can also intelligently adjust the computing power. force distribution. NVIDIA's idea is very good. One chip solves multiple problems and reduces the difficulty of car design and integration, but can this solution really be realized?

All functions in one chip, is it possible?

In order to keep various functions from interfering with each other, many functions of the car are not responsible for the same chip. For example, the automatic driving and entertainment functions are two different chips, and the radar detection function required for automatic driving also requires A separate chip is done. Due to the large size of the car, the requirements for interior space are not high, and car companies will not care about too many chips. At most, it is more troublesome to design. NVIDIA's solution is different. Many functions need to be integrated into one chip, such as Thor, which can be used to run the vehicle system, support functions such as entertainment and navigation, and also have the functions of automatic driving. The question that arises from this is, is it possible in the future to integrate all the functions required by a car with a chip? Source: pixabay In recent years, many chip manufacturers have been engaged in functional integration and increased chip integration. For example, Desay SV is trying to integrate the driving assistance and automatic parking functions into the same chip, and many car companies choose this solution. Semiconductor company Infineon published an article saying that the continuous improvement of the integration of automotive computer control modules (ECUs) can make the circuit panel volume unchanged or reduce, reduce the number of chipsets, and reduce the cost of automotive chips. However, it is not an easy task to integrate all functional operations into one chip. How to connect and process various sensors in the car is a headache for chip companies. These sensors are not produced by the same company. Therefore, from a technical point of view, functional integration can be achieved, and related semiconductor companies are working in this direction, but there is still a long way to go to really achieve this. At present, most semiconductor companies, like NVIDIA, can only integrate one or two functions of computing into one chip. Image source: pixabayNVIDIA is a special one, because the integrated functions of Thor are the most important car-machine systems and autonomous driving of future cars, and Thor's computing power has reached 2000TFOPS. According to the data of the Yiou think tank, the computing power needs to increase tenfold for each level of autonomous driving. In the future, the mainstream L4 autonomous driving will require about 400 TFOPS computing power. Thor can divide a part of the performance under the premise of satisfying the L4 autonomous driving. Used to maintain other functions of the car. As of today, Thor is the chip with the highest computing power in the automotive industry. Coupled with the influence of NVIDIA, it is foreseeable that Thor will become a sweet pastry in the automotive industry after mass production. Main selling point promotion.

Thor, let the sparrows fly to the branches and become phoenixes?

From the perspective of computing power, everyone guessed that the price of Thor will never be cheap. Xiaolei inquired about it. The 275TFOPS computing power chip Orin launched by NVIDIA before, the purchase cost of the car company is about 500 US dollars (about 3,500 yuan) per piece, and the price of Thor is definitely more exaggerated. The status quo of new energy vehicles is that the old car companies have a difficult turnaround situation, and the market value of new car manufacturers with little product sales is approaching the old car companies. Although the foundation of these emerging car companies is not stable, almost all of them have to enter the high-end market in order to build brand awareness and improve brand quality. Source: pixabay will naturally not be so easy to hit the high-end, but unlike the old car companies, many executives of the new car manufacturers come from Internet companies, or some Internet companies have injected capital, and marketing is their best. The old car companies also have publicity, but it is difficult for ordinary people to understand the obscure car knowledge. If you go to the street and ask a person who does not understand cars what is the difference between turbocharging and natural aspirating, few people can explain clearly. In contrast, the car chip is a very simple and clear publicity stunt. After all, I go to the street to find someone to ask him which chip to choose between 2000TFOPS and 10TFOPS computing power without considering the price. I believe most people will choose it. 2000TFOPS computing power chip. In fact, car companies have already started marketing the concept of "running points". The direction of their marketing is computing power. 1TFOPS can run 1 trillion times per second. However, the imagination is very full and the display is very skinny. The blessing that Thor can bring to car companies is very limited. There are three reasons. First, the brand image of the car companies themselves is insufficient. Cars can rush to the high-end market without high-end configuration. Brand value needs long-term operation to stabilize. Source: pixabay Second, car companies and consumers need to consider costs. If the car uses high-end chips, the price will naturally not be low. If the brand value is not recognized by consumers, the sales of the car will definitely not be high, and the loss will be more serious. Third, the integration capacity is insufficient. Chips can solve the problem of computing power, but it is impossible for an ordinary car company to instantly have excellent autonomous driving and assisted driving algorithms, and it is even more impossible for ordinary car companies to immediately provide a driving experience comparable to that of luxury cars. Integration is a very complicated thing, and not all car companies can handle it. There are many new car manufacturers that have hit the high-end market in recent years, but only a few car companies such as Weilai, Ideal, and Xiaopeng have achieved good results. Byton Motors, which was once favored by many investors, made a full 8.4 billion yuan go to waste. Another car company with a strong momentum is Celis. With Huawei's help, the industry has performed well, but it is still unknown how far it will go in the future. For car companies that have already entered the high-end grade, Thor is the key to further product strength, but for car companies that have not been able to rush to the high-end, it is the right way to continue to pursue cost-effectiveness.

Hash rate should not be a gimmick for car companies

Digital products such as mobile phones and computers are often used in performance competitions. Use various benchmarking software, or run some games to compare frame rate and stability. Different cars, autonomous driving technology has not been fully popularized, and the automotive software ecosystem is relatively simple, which is not enough to support the comparison of competing products of car companies. Therefore, the concept of computing power, which is often mentioned by virtual currency miners, has become a new gimmick in the mouth of car companies. On some automotive media platforms, Xiaolei has seen a lot of comparison articles on car computing power, as well as things like car computing power rankings. According to data released by AutoLab, the current car with the highest computing power is the Wei brand Mocha DHT-PHEV lidar version, followed by the Weilai ES7. It is true that the increase in computing power will greatly upgrade the performance of a car, but for autonomous driving and user experience, computing power is not the most critical, just enough, the most important thing should be ecology. When it comes to ecology, we have to blow Tesla. In addition to the traditional in-vehicle entertainment functions, Tesla even wanted to move the game console to the car. Model S already supported running some games earlier, and later Tesla even planned to move the Steam platform to the car. Apple CEO Cook once said that the car is the ultimate form of mobile devices, and the computing power of the chip alone is definitely not enough for this ultimate device, and ecology is definitely a more important factor. Blindly competing for chip computing power will only increase costs and product prices, a situation that consumers absolutely do not want to see. Perfecting the ecology is what car companies need to pay attention to. In fact, there are already car companies working hard in this regard, for example, Weilai is also “moving” games to cars. The current automobile ecology is still weak, and the new forces that can build cars have always had a close relationship with Internet companies. The perfection of the automobile ecology may be faster than we imagined. At that time, we really need high computing power chips.

(Responsible editor:Monitor)

Related content