Who is the strongest car enterprise intelligence center?

Since Tesla established its own Dojo a few years ago, car companies interested in intelligent driving and even the next generation development have turned their attention to this field.

At home, Xpeng Motors built a mental arithmetic center in Wulanchabu last August. At the beginning of June 5438+ 10 this year, Mimo Zhixing of Great Wall Motor announced that the "Snow Lake Oasis" intelligent computing center was put into use.

As for other competitors, they are also using various public cloud resources to enter this track.

Although all the facilities were declared as "the strongest" before release, after making a short video and sending it to Tik Tok for review, I really don't know which one will pass.

But one thing is certain, until the end of last month or even this time, the performance of Geely Star Intelligent Computing Center deployed by Geely Automobile Group in Changxing County, Huzhou was second to none in China.

Geely Xingrui, Huzhou's new "specialty"

At the end of February this year, the author was fortunate to be invited to Changxing County, Huzhou, to participate in the launching activities of the Star Smart Computing Center jointly organized by Geely and Ali.

Why choose Changxing County, Huzhou? This is a question I thought about before I left. After all, as far as personal cognition is concerned, there has always been a taste here in the last ten years, just like the corner of Jiangsu, Zhejiang and Shanghai.

But when I drove out of the South Taihu Interchange on the Shanghai-Chongqing Expressway, I immediately realized that there was no need to answer this question.

At this time, a strong southeast wind blew from behind. Its ferocity makes the car body shine. Heat dissipation has always been one of the important problems in computing centers. And such a strong wind is really an ideal place to build a computing center.

Geely Automobile is located in Changxing County.

It is almost common knowledge in Jiangsu, Zhejiang and Shanghai that it is windy around Taihu Lake. Whenever the temperature gets warmer in a year, a cold high pressure with Taihu Lake as the center is gradually generated, and the wind on the lake begins to blow around.

In addition, Dongque in Changxing County is close to Nanyunfeng and located in the center of the south bank plain of Taihu Lake. The highest peak at 52 1.5 meters above sea level on the south bank of Taihu Lake blocked a considerable part of the airflow and forced it to blow to both sides.

The intelligent computing center of Geely Star is backed by a hill called Fengshan on the west side of the windward side of Nanyunfeng. Coupled with the bypass effect brought by the isolated mountain, it is windy all the year round, and it is indeed an ideal place for Jiangsu and Zhejiang to deploy computing centers.

A few years ago, Ali Group built the Zhangbei Data Center of Ali Data Port in Xiaoertai Town, Zhangbei County, Hebei Province, fully considering the local windy and low temperature characteristics.

Although Jiangsu and Zhejiang can't expect a low average temperature in Zhangbei County due to latitude and climate problems, the strong wind blowing around Taihu Lake all the year round does have a significant effect on reducing the energy consumption of the cooling system of the computing center.

Besides, it is adjacent to the first-and second-tier central cities after all. Within 30 minutes' drive from downtown Huzhou, it takes 3 hours to the east to reach Shanghai and 2 hours to the southeast to reach Hangzhou.

Local governments facing the pressure of industrial upgrading, GDP assessment and other issues naturally welcome Geely Automobile Group's willingness to further lay out computing centers outside the production base.

Therefore, on the day of the delivery ceremony, many local city and county leaders in Huzhou also made a special trip to attend the meeting, which shows their importance.

After all, this is the first large-scale computing facility specially for automobile enterprises in Huzhou besides Unicom and Beidou.

The predecessor of Geely Star Intelligent Computing Data Center is an enterprise-level data center established by Geely Holding Group in Changxing County. Its main task is to provide cloud container services for the company's internal operations.

Such as website, OA office software, office cloud and so on. 202 1,1In February, the project of Geely Star Smart Computing Center was formally established, and the upgrade construction was officially started in July 2022.

Alibaba Cloud, a well-known cloud computing service provider, provides construction and follow-up operation and maintenance services for Star Smart Computing Center.

Geely Changxing Data Center, the total investment of the project is about 654.38+0 billion yuan, with a total area of 52. 12 mu and 5,000 planned cabinets. 2520 standard 5kW cabinets have been built in the first phase of Star Intelligent Computing Center.

The performance of Star Smart Computing Center is in a leading position among similar computing centers in China at present-the total cloud capacity of the whole system peaks at 865,438+billion times per second, and the maximum communication network transmission rate is 800GB per second.

In particular, its storage bandwidth reaches 4.5TB per second, which is 400 times that of Ethernet widely used in the industry.

At present, the Star Intelligent Computing Center also plans to have the second phase of the project. Once the second phase of the project is completed and landed, that is, all the planned 5,000 cabinets are deployed in place, and the system scale will eventually exceed EFLOPS, reaching an astonishing 654.38+0.2 billion times.

As a cloud support system for smart cars, the number of online vehicles supported by Star Intelligent Computing Center will reach 3.5 million, and the data storage scale will reach EB level.

Intelligent computing center era

What needs special emphasis here is the positioning of Geely Xingrui, the "intellectual center". It is different from traditional data centers and supercomputing centers.

Intelligent computing center is a rising concept in recent two years.

"Intelligent computing" means that the facility is not a supercomputer facility based on a general-purpose processor (CPU) in the traditional sense. A considerable part of its ultimate strength will be composed of A.I chips, which are designed to serve artificial intelligence projects.

Therefore, the key point to understand the importance of Starsharp and even the value of intelligent computing center is what is A.I chip.

As we all know, the operation related to artificial intelligence has its particularity. Whether it is the recognition of various voice signals, the discrimination of various graphics and images, or the popular deep learning training based on artificial neural network. When computers run such tasks, they usually face the problem that the tasks are trivial and complicated, but the number of parallel tasks is huge.

In this case, CPU with strong control (controller) and arithmetic logic unit (ALU) is usually very inefficient.

The reason why A.I project is poor is that it has its particularity, and there are often no complicated calculation tasks, but there are generally a lot of "small data" to be calculated at the same time. This means that a small number of high-performance controller +ALU combinations in CPU are limited in number, and it is impossible to efficiently parallel process the tasks composed of such subdivision projects.

The so-called A.I chip is opposite to CPU in structure-the area of a single controller and ALU inside the chip is small, but the total number is large. Even the first-level cache is divided into small blocks and allocated to each controller module.

These characteristics are obviously different from CPU, which makes it handy when dealing with A.I projects. Give a simple example: intelligent driving of cars.

The realization of automobile intelligent driving function usually depends on the perception of various external sensors on the surrounding environment and their own situation. Under the coordination of the autopilot domain controller, the optimal strategy of the current scene is realized.

In this process, when the visual sensor sends back the image signal, the system will disassemble the video signal into frames for interpretation, so as to understand the meaning of graphics in all areas of the whole image-either traffic signs, pedestrians, other motor vehicles and so on.

This means that the system will face the problem of processing a large number of small image blocks at the same time and identifying them separately.

Although each graph is not complicated, a picture is often composed of dozens of objects, and there are dozens of images every second (depending on the specific refresh rate of the vehicle vision sensor).

At this point, I believe many people have understood that the A.I chip is actually a specially optimized graphics processor (GPU).

After all, this kind of work can also be regarded as another type of massive graphics processing task to a great extent. Even if the tasks it performs have nothing to do with images, such as speech recognition, it faces a lot of complex multithreading problems, such as sound wave recognition and filtering.

And all kinds of machine deep learning training based on neural network also need to face complex multithreading problems.

Understanding what A.I chip is naturally solves the understanding of the positioning of intelligent computing center.

Judging from the current industrial development, artificial intelligence has three driving forces: data, algorithm and computing power, which are closely related to computing. For car companies, if they want to develop in intelligent driving, massive simulation and huge data collection are essential.

In this context, building a large intelligent computing center can obviously complete the iteration and evolution of its own model and algorithm faster.

In the field of intelligent driving research and development, if the traditional local model training mode is adopted, it will take about 3 months to complete the training of 1000 intelligent driving models. Using the rich A.I computing power of Star Intelligent Computing Center, the same work only takes 8 hours, and the research and development efficiency is improved by more than 200 times.

Of course, the Geely Star Intelligent Computing Center is not only used for the promotion of enterprise intelligent driving business and the research and development of related technologies. It can also perform the functions of supercomputing center, such as supporting automobile research and development. In addition, Geely's cloud support for users is also center-based.

As an integrated cloud computing platform of "cloud, number and intelligence", it can realize the compliance and enjoyment of all ecological data including aerospace, high-definition maps, smart energy, smart electricity, smart buildings, smart travel and so on under the Geely Holding Group system, and provide R&D and operational support for all links (monitoring, operation and OTA) of the automobile ecological chain.

In addition, Geely and Ali pioneered the intelligent cross-pool scheduling technology of computing power, which improved the utilization rate of computing resources by at least 20% compared with the existing technology of China Head Public Cloud Service Provider, and further accelerated the research and development of intelligent driving models and artificial intelligence services.

Of course, what we can't forget is that behind all this is Ali's figure. The three mental arithmetic centers mentioned above, except Xuehu Oasis, Xingrui and Fuyong, which are assisted by Volcano Engine, ByteDance, were all written by Alibaba Cloud.

In fact, most OEMs who have not built their own intelligent computing centers for the time being are doing similar work by using the public cloud provided by Alibaba Cloud.

As the third cloud service provider in the world and the first in Asia-Pacific, Alibaba Cloud has almost built a "production room" for the new generation of smart cars in China.

Among them, from the intelligent computing center built by car companies, the Star intelligent computing center is undoubtedly the most powerful at this stage.

Inside Xingruizhi Computing Center

However, on the intelligent and networked car track, the key to ultimate victory lies not in how powerful tools are mastered, but in how car companies use the tools in their hands to create products that lock in victory.

This article comes from the author's automobile commune, and the copyright belongs to the author. Please contact the author for any form of reprint. The content only represents the author's point of view and has nothing to do with the car reform.