Editor's Note 丨 "T Observation" is a commentary column focusing on the forefront of the automotive industry launched by Sohu Automobile, focusing on the focus issues of electric, intelligent, networked, sharing and other tracks, starting from the perspective of strategy, industry, technology, market, management, user, etc., comparing different cases in various fields, analyzing the operation mode and internal law of the observation object, taking the phenomenon as the point and the law as the line, building an information value network, providing readers with valuable and reference views.
Produced by 丨 Sohu Auto Car Cafe
It has been a while since nio's three new models, the ES7, ET5, and ET7, have been on the shelves. All three models are equipped with NVIDIA large computing chip, with a computing power of 1016TOPS, which is the best among the currently listed models.
Logically, the large computing chip is to prepare for the high-level automatic driving ability to get on the car. However, because the regulations have not yet been opened and Weilai does not have this ability, consumers can only buy "futures".
Although the high-level autonomous driving capabilities of L3 and above cannot be experienced, has the already existing L2 assisted driving capability been improved? So far, we have not seen the conclusions given by WEIlai officials or relevant experiencers. Although there is no necessary connection between large computing power chips and the level of automatic driving capabilities, it is obviously not advisable to only stack computing power and not improve the ability.
In addition to a variety of Models of Weilai, the Ideal L9 is also equipped with NVIDIA's large computing chip, and the computing power reaches 508TOPS. The problems of the two are similar, in order to achieve higher autonomous driving capabilities, the hardware is ready, but the basic capabilities such as software algorithms and regulations have not seen much substantial progress.
Both Weilai and Ideal claim to be full-stack self-research, but in contrast, the autonomous driving capabilities that can be provided are not leading in the industry, and there is no strong planning and rhythmic market arrangement. After all, whether computing power can be transformed into ability cannot rely solely on courage and confidence.
What exactly can big computing power bring? From the perspective of brand building, basically every company that plays a big computing power brand wants to use the scientific and technological attributes of the chip to label itself a technology label and beautify the competitiveness of the product. But looking back at the smart phone industry that likes analogies in the field of smart cars, although Apple, Huawei, Xiaomi and other companies will take out the chip when the product is released, the three brands and products are widely recognized and relished in the system is smooth, good camera effect, and high cost performance. Computing power is the support, but it has not been hung up by consumers.
In practice, high-level automatic driving requires greater computing power, although large computing power can solve the problem, but the practice of "vigorous miracles" is not advisable, and will pass on power consumption and cost to consumers. At present, the practice of joint tuning of software and hardware has become the consensus of the industry, which has also been verified on the self-developed chips of Apple and Tesla. Instead of piling up materials, think about how to make the most of it.
In fact, in the field of smartphones, this phenomenon is also happening. Steve Blank, the founder of the Silicon Valley Lean Startup movement and the most influential serial entrepreneur, once suggested that instead of "how good we can develop chips", we should focus on "what can be done with such a good chip". The problem points directly to the fact that consumers change their phones every year, and the improved performance is not fully utilized. We are now facing "overperformance", not "crushing performance".
At present, the thinking of chip companies on the horizon is similar to Tiff Blank's view of how to make good use of chips, not emphasizing the large computing power alone, but focusing on the fast calculation.
In addition to the above obvious problems, what are the less obvious motivations of vehicle companies? At a forum in early July, Xue Chunyu, CEO of Smart Mobility Group and chief operating officer of Furui Zhixing, believed that it was because of uncertainty about technology and the market. In the face of the above uncertainties, some insiders believe that the system planning ability of vehicle enterprises is insufficient, and only the general trend of automatic driving development is clear, but the specific landing is to take a step and look at one step. Ding Ding, vice president of Shanghai Xinwang Microelectronics, believes that there should be systematic planning to how to gradually exert the computing power, architecture, platform and efficiency within five years of the car, there is a clear logic that can be passed on to consumers a clear expectation, every six months and a year my driving experience has new changes, I believe that this buried price, consumers should be acceptable.
Vehicle companies, especially the new car-making forces, should continue to follow the "mass production generation, reserve generation, research and development generation" technology and market coordination system established by the automobile industry, and cannot peddle futures, let consumers bear the losses caused by uncertainty, and cannot take consumers as guinea pigs.
Forecast data from relevant agencies shows that in 2023, the penetration rate of L1/L2 will reach a peak of 49%, and by 2027, this data will be 39%, while the penetration rate of L3 will rise from 4.4% this year to 16%. Even if L4/L5 goes all the way up, it is still only 7.2%. At the same time, in terms of hash rate demand, L1 is less than 1TOPS hash rate, L2 needs 10+ TOPS hashrate, L3 needs 100+ TOPS hashrate, L4 requires 500+TOPS hashrate, and L5 needs 1000+TOPS hashrate.
At present, Yihang Intelligent, Light Boat Intelligent Navigation, Horizon and other enterprises have launched small computing power of automatic driving solutions, which can not only meet the intelligent improvement needs of automobile enterprises at this stage, but also realize data closed loop and help algorithms iteratively improve. At the public forum a few days ago, Wu Gansha, co-founder, chairman and CEO of Yishi Technology, introduced that Yishi Technology can achieve L2 capabilities with a computing power level of less than 10TOPS, and meet the data closed-loop and algorithm iteration. At the same time, this system uses the same set of all-scenario, unified architecture of the autonomous driving operating system U-OS with higher-level automatic driving capabilities.
Judging from the current degree of promotion of laws and regulations, moderate advance can solve the contradiction between the authority of the gradual demonstration area and market demand, but it is too advanced and enters the competitive cycle of labeling itself by relying on spelling power, which is not only unfavorable to itself, but also harmful to consumers.
From the perspective of promoting the development of autonomous driving technology, algorithm iteration is inseparable from high-quality data, and data needs to be collected by mass production vehicles deployed in large quantities. Therefore, the model of positive circulation should be that vehicles that meet the demand enter the private consumer market in large quantities at the right price, so as to achieve the bilateral effect of increasing the ability to drive themselves and selling on a larger scale.
T observation 06丨 large computing power is not equal to good intelligent driving