Fang Lei of Nine Chapters Cloud received an interview from CCTV Financial Channel: The implementation of L3 has triggered a surge in computing power demand, and cloud training has become the core support.
With the approval of the first batch of Level 3 autonomous driving vehicle models by the Ministry of Industry and Information Technology, China's autonomous driving industry has entered a new stage of commercial application from the testing phase. The program "World of Business" on the financial channel of China Media Group conducted a special report on this. Fang Lei, the chairman of DataCanvas, was interviewed to deeply interpret the core value of computing power in the implementation of Level 3 autonomous driving and the industry trends.
The report pointed out that in addition to relying on in - vehicle chips, the development of Level 3 and higher - level autonomous driving has an extremely large demand for cloud computing power during the training phase. Currently, some automakers have carried out the training of autonomous driving models through the computing power centers of DataCanvas distributed across the country. The commercial implementation of Level 3 will further boost the leap - forward growth of the computing power market scale.
Fang Lei emphasized in the interview that the implementation of Level 3 is a landmark event for the computing power industry, which will trigger a concentrated explosion of computing power demand in the automotive and autonomous driving fields. "In a typical scenario, the training of an end - to - end autonomous driving model requires a computing power cluster of more than 4,000 (computing power) cards to run continuously for several months, and the investment scale is extremely large."
When talking about the cost structure of computing power, Fang Lei introduced that if the chip procurement cost is not considered, cloud - based training accounts for more than half of the total computing power cost. Although the unit training cost of models with the same intelligence level has decreased sharply, due to the continuous increase in the industry's demand for more intelligent models, the total training cost is still increasing.
Facing this surging market demand, as a provider of AI infrastructure and intelligent computing cloud, DataCanvas is fully prepared. The company's computing power scale has maintained rapid growth in the past two years, and it is expected to achieve a 2 - 3 times growth rate every year in the next 3 - 5 years, continuously improving the computing power supply capacity to escort the large - scale development of the autonomous driving industry.
To precisely meet the core needs of the intelligent driving field, DataCanvas' intelligent computing cloud has specifically developed a full - link intelligent computing solution suitable for autonomous driving scenarios. Focusing on the core needs of autonomous driving, such as "training with massive data, simulating complex scenarios, and lightweighting vehicle - end models", it optimizes the computing power scheduling framework, which can support the efficient training and reasoning of end - to - end autonomous driving models and intelligent cockpit interaction models, and is compatible with the differentiated chip combination solutions of different automakers. In addition, by leveraging transfer reinforcement learning technology, DataCanvas' intelligent computing cloud can significantly improve the data training efficiency in complex road conditions and edge scenarios, helping automakers quickly iterate on model performance to meet the strict requirements of Level 3 and higher - level autonomous driving for safety and stability, and providing solid computing power support for the large - scale implementation of intelligent driving technology.
The commercial implementation of Level 3 autonomous driving not only opens a new stage of intelligent transportation but also promotes computing power to become the core barrier of industrial competition. With its large - scale computing power supply and scenario - based solutions, DataCanvas is deeply involved in the construction of the autonomous driving industry ecosystem. In the future, it will continue to empower the intelligent upgrade of automakers through technological innovation, helping China gain an edge in the global intelligent driving race.