At the 2024 China Computing Power Conference held recently, Zhao Zhiguo, chief engineer of the Ministry of Industry and Information Technology, said that the total scale of the computing center rack in use in mainland China exceeds 8.3 million standard racks, and the total scale of computing power reaches 246EFLOPS (exascale floating point operations per second), which is at the forefront of the world, and the industrial development continues to improve quality and efficiency.
At present, the pace of digital, high-end, and intelligent transformation of various industries is accelerating, and the development momentum and vitality are bursting out, and it is more urgent to need the "artery" of the new digital information infrastructure represented by computing facilities to smooth the flow of factors and economic circulation, and it is urgent for the computing power industry to leap from "quantity" to "quality" to build a strong foundation for high-quality economic and social development.
The total volume is growing rapidly
Behind the rapid development of artificial intelligence is the explosive growth of computing power demand. In recent years, the development of computing infrastructure in mainland China has achieved remarkable results, with the initial construction of a tier-optimized computing power supply system, the significant improvement of the comprehensive capacity of computing infrastructure, and the continuous innovation and development of the computing power industry, which continues to empower thousands of industries.
According to the data released by the Ministry of Industry and Information Technology, the total scale of computing power in mainland China is among the top in the world; With the in-depth promotion of interconnection, the 1ms (millisecond) delay guarantee capability between metro computing power nodes has gradually taken shape, the 5ms delay guarantee capability between regional clusters and surrounding major cities has been widely covered, and the 20ms delay guarantee capability between national hub nodes has been fully realized. The average power utilization efficiency (PUE) of computing centers across the country has dropped to 1.47, 246 national green data centers have been established, and more than 140 computing centers have reached the green and low-carbon level of 4A or above.
It is worth noting that the mainland has attracted foreign-funded enterprises and private enterprises to actively participate in the "computing power industry development phalanx", forming a new industrial development ecology of resource sharing, platform co-construction and value co-creation. There are more than 13,000 computing power application projects in various fields such as industry, education, healthcare, and energy, which strongly support the rapid development of artificial intelligence and lay a solid foundation for sharing the fruits of digital development on a larger scale and at a higher level.
According to the China Academy of Information and Communications Technology, by the end of 2023, the total scale of global computing infrastructure will reach 910EFLOPS, a year-on-year increase of 40%. United States and China rank among the top two in terms of computing infrastructure, accounting for 32% and 26% of the world's computing power, respectively.
At the same time, the structure of computing power in mainland China has been continuously adjusted, and the scale of intelligent computing has grown strongly. The scale of storage capacity continues to expand, and the proportion of advanced storage capacity continues to increase; The quality of transportation capacity has been significantly improved, and the network facilities have been continuously upgraded. As of June 2024, the scale of computing power in mainland China has reached 246EFLOPS, and the growth rate of intelligent computing has exceeded 65% year-on-year.
"As a new type of productivity integrating computing power, storage capacity and transportation capacity, comprehensive computing power is the core driving force to accelerate the development of the mainland's digital economy." Yu Xiaohui, president of the China Academy of Information and Communications Technology, said.
Zhou Yuefeng, Vice President of Huawei and President of the Data Storage Product Line, believes that the development of digital applications in the mainland is rapid, and the coordinated development of storage, computing, and networks will provide the mainland with a sustainable supply of computing infrastructure and provide better choices for the world. Huawei's integrated design of computing, network, and storage integrates computing power to build a computing power foundation for the AI era, enabling key industries such as government affairs, healthcare, finance, and manufacturing, and accelerating the intelligent transformation of society and industries.
The increment still needs to be improved
Li Huidi, deputy general manager of China Mobile, said that artificial intelligence is a major innovation at the level of the industrial revolution, and the breakthrough of AI technology and the diffusion of applications show significant scale effects, accelerating the transformation of the economy and society from "+AI" to "AI+". Among them, the scale effect of AI technology capabilities has promoted the explosive growth of training computing power; The scale effect of AI economic benefits will bring the demand for inference computing power far beyond training, and the computing power networking, computing-electricity synergy, and low-carbon development of 10,000 and 100,000 calorie intelligent computing clusters have become innovative directions.
The total scale of computing power in mainland China is growing rapidly, but from a structural point of view, the proportion of high-performance computing power needs to be improved. Zhao Zhiguo emphasized that the development of the computing power industry also needs to coordinate the "quantitative growth" and "qualitative improvement".
Guo Liang, chief engineer of the Institute of Cloud Computing and Big Data of the China Academy of Information and Communications Technology, said in an interview with the Economic Daily that the rise of new needs and new businesses such as artificial intelligence applications and large model training represented by generative artificial intelligence has promoted the rapid growth of the global intelligent computing scale. It is estimated that by the end of 2023, the global intelligent computing power scale will be 335EFLOPS, a year-on-year increase of 136%, and the growth rate will far exceed the growth rate of the overall scale of computing power.
The proportion of intelligent computing power in mainland China has also increased significantly, and intelligent computing centers are clustered and distributed. According to the calculation of the China Academy of Information and Communications Technology, by the end of 2023, the scale of intelligent computing power will account for nearly 30% of the increase in the overall computing scale, and the efficiency increase is obvious. At the national and local levels, the construction of intelligent computing centers has been actively promoted, and Beijing, Guangdong and other places have put forward the scale target of intelligent computing by 2025. From the perspective of regional distribution, the intelligent computing center shows a trend of cluster construction, and more than half of them are distributed in the eastern part of the mainland.
According to the "Report on the Evaluation System for the High-quality Development of Artificial Intelligence Computing Power", the development of computing power in mainland China is still facing insufficient supply of computing power and an imbalance between supply and demand; The level of computing power intelligence is low, and it is difficult to meet multiple application scenarios; Computing power is facing energy tests, and energy conservation and carbon reduction are urgent; The demand for diversified computing power has increased, and the level of universal inclusion is low; The completeness of the supply chain is insufficient, and the ecological construction needs to be improved; Problems such as simple performance evaluation and lack of measured performance of computing power urgently require the computing power industry to accelerate high-quality development from "quantity" to "quality".
In this regard, Zhang Dong, chief scientist of Inspur Yunhai, analyzed that high-quality computing power should have the characteristics of high computing efficiency, high intelligence efficiency, high carbon efficiency, sustainability, availability, and evaluation. Among them, high computing efficiency is to see how much computing power can be provided as a whole, and high intelligence efficiency is to use the smallest computing power to support a higher level of intelligence. According to public data, the average GPU utilization rate of the intelligent computing center in the traditional mode is less than 30%, and high computing efficiency is to solve the problems of low computing efficiency level and low utilization rate of computing resources. At present, the development of large models is in full bloom, and the intelligent emergence ability of large models has been greatly improved, but the modeling efficiency of existing large models is generally not high.
From big to strong
Zhao Zhiguo said that it is necessary to improve the construction layout mechanism, formulate development plans for major regional development strategies, and promote the efficient complementarity and collaborative linkage of computing power in the east and west. Improve the multi-allocation mechanism, formulate development plans for emerging industries such as artificial intelligence, build a number of intelligent computing centers, and reasonably allocate the proportion of general intelligent computing and supercomputing. Improve the interconnection mechanism, build computing power interconnection program rules, market rules, and quality inspection rules, and promote the formation of a development pattern of a unified market construction with a computing power market system serving the construction of a unified market.
In the view of Chen Zhenkuan, vice president of Lenovo Group, hybrid artificial intelligence will become a key leap from the accumulation of computing technology to the release of computing value, and a strong support for the realization of artificial intelligence inclusiveness. In the hybrid AI framework, personal and enterprise models, as well as individual agents and enterprise agents developed on their basis, will coexist and complement each other with public models to accelerate the inclusion of AI.
Yu Xiaohui said that it is necessary to further optimize the computing power supply structure and new infrastructure layout, improve the construction level of computing power centers, improve the relevant policy and standard system, strengthen technological innovation and R&D capabilities, rely on platforms such as the "intelligent computing ecosystem" to accelerate industrial empowerment and ecological improvement, and accelerate the high-quality development process of the comprehensive computing industry.
It is understood that the Ministry of Industry and Information Technology will promote greater breakthroughs in basic software and hardware technologies such as high-end chips, architecture design, large model algorithms, and high-speed transmission protocols, and establish and improve the standard system of the computing power industry; Cultivate a number of typical application scenarios and solutions empowered by computing power in key links such as R&D and design, manufacturing, warehousing and logistics, and marketing services in industrial enterprises; Focus on cultivating a number of computing power application solution providers for small and medium-sized enterprises, build a number of service platforms that integrate multiple resources and open multiple capabilities, and form a replicable and promotable development model in key areas such as industry, education, transportation, and energy.
Source: Economic Daily