At the recent 2024 China Computing Power Conference, Zhao Zhiguo, Chief Engineer of the Ministry of Industry and Information Technology, stated that China's total scale of computing power center racks exceeds 8.3 million standard racks, with a total computing power of 246 EFLOPS (hundred billion billion floating-point operations per second), ranking among the top in the world, and the industry development is continuously improving in quality and efficiency, moving towards innovation.
The current pace of digitalization, high-endization, and intelligent transformation in various industries is accelerating, with development momentum and vitality emerging in competition, creating an urgent need for new types of digital information infrastructure represented by computing power facilities to facilitate the flow of elements and economic cycles, the "arteries". There is an imminent need for the computing power industry to leap from "quantity" to "quality", building a strong foundation for high-quality economic and social development.
Advertisement
The total volume is growing rapidly. Behind the rapid development of artificial intelligence is the explosive growth in demand for computing power. In recent years, China's computing power infrastructure has achieved significant development, with an initial construction of a hierarchically optimized computing power supply system, a noticeable enhancement in the comprehensive capabilities of computing power infrastructure, continuous innovation and development in the computing power industry, and sustained empowerment across various industries.
Data released by the Ministry of Industry and Information Technology shows that China's total computing power ranks among the top in the world; interconnectivity is being deeply advanced, with the capability to ensure 1ms (millisecond) latency between urban computing power nodes gradually taking shape, 5ms latency coverage widely spread between regional clusters and surrounding major cities, and 20ms latency capability fully realized between national hub nodes; the average power usage efficiency (PUE) of national computing power centers has been reduced to 1.47, with the creation of 246 national green data centers, and over 140 computing power centers achieving a green and low-carbon level of 4A or above.
It is worth noting that China has attracted foreign-funded enterprises, private enterprises, and others to actively participate in the "computing power industry development array", forming a new industrial development ecosystem of resource sharing, platform co-construction, and value co-creation. Computing power application projects in multiple fields such as industry, education, healthcare, and energy exceed 13,000, strongly supporting the rapid development of artificial intelligence and laying a solid foundation for sharing digital development achievements on a larger scale and at a higher level.
According to estimates by the China Academy of Information and Communications Technology, by the end of 2023, the total scale of global computing power infrastructure reached 910 EFLOPS, a year-on-year increase of 40%. The United States and China's computing power infrastructure scales are ranked first and second, with global computing power shares of 32% and 26%, respectively.
At the same time, China's computing power structure is continuously adjusting, with the scale of intelligent computing growing strongly; the scale of storage continues to expand, with the proportion of advanced storage continuously increasing; the quality of transportation has significantly improved, and network facilities are continuously upgraded. As of June 2024, China's computing power scale reached 246 EFLOPS, with the growth rate of intelligent computing exceeding 65% year-on-year.
"Comprehensive computing power, as a new type of productivity that integrates computing power, storage power, and transportation power, is the core driving force for accelerating China's digital economy development," said Yu Xiaohui, Dean of the China Academy of Information and Communications Technology.
Zhou Yuefeng, Vice President of Huawei and President of the Data Storage Product Line, believes that China's digital application process is developing very quickly, and storage, computing, and networking need to develop in coordination to provide sustainable computing power infrastructure for China and better choices for the world. Huawei, through the integrated design of computing, networking, and storage convergence, has created the computing power base for the AI era, empowering key industries such as government affairs, healthcare, finance, and manufacturing, and accelerating the intelligent transformation of society and various industries.Incremental Improvements Must Also Enhance Quality
China Mobile's Deputy General Manager, Li Huidi, stated that artificial intelligence (AI) is a significant innovation on the scale of an industrial revolution. Breakthroughs in AI technology and the diffusion of its applications exhibit a pronounced scale effect, accelerating the socio-economic transition from "AI+" to "AI+". Among these, the scale effect in AI technology capabilities has driven an explosive growth in training computational power. The scale effect in AI economic benefits will lead to an inference computational power demand far exceeding training, with innovations focusing on the power networking of ten-thousand-card and hundred-thousand-card intelligent computing clusters, the synergy of computing and electricity, and low-carbon development.
The total scale of China's computing power has grown rapidly, but structurally, the proportion of high-performance computing power needs to be increased. Zhao Zhiguo emphasized that the development of the computing power industry must also balance "quantitative growth" with "qualitative improvement".
Guo Liang, the Chief Engineer of the Cloud Computing and Big Data Research Institute at the China Academy of Information and Communications Technology, introduced to the Economic Daily reporter that the rise of new demands and businesses represented by generative artificial intelligence applications and large model training has propelled the global intelligent computing scale to show a high-speed growth trend. According to estimates, by the end of 2023, the global intelligent computing scale reached 335 EFLOPS, with a year-on-year growth of 136%, far exceeding the overall growth rate of computing power.
China's proportion of intelligent computing power has also increased significantly, with intelligent computing centers being concentratedly distributed. According to estimates by the China Academy of Information and Communications Technology, by the end of 2023, the scale of intelligent computing power accounted for nearly 30% of the overall computing power scale, showing a clear increase in efficiency. National and local levels are actively promoting the construction of intelligent computing centers, with Beijing, Guangdong, and other places setting intelligent computing scale targets for 2025. In terms of regional distribution, intelligent computing centers are showing a trend of cluster construction, with more than half distributed in the eastern regions of China.
The "Artificial Intelligence Computing Power High-Quality Development Evaluation System Report" believes that China's computing power development still faces issues such as insufficient computing power supply, unbalanced supply and demand matching; low intelligence level of computing power, which is difficult to meet diverse application scenarios; computing power facing energy challenges, where energy conservation and carbon reduction are urgent; increased demand for diversified computing power, with a low level of universal benefits; insufficient completeness of the supply chain, and the ecosystem construction is yet to be improved; simple performance evaluation, and a lack of actual measured performance of computing power, among other issues. There is an urgent need for the computing power industry to accelerate the high-quality development from "quantity" to "quality".
In response, Zhang Dong, the Chief Scientist at Inspur Cloud Sea, analyzed that high-quality computing power should have characteristics such as high computational efficiency, high intelligence efficiency, high carbon efficiency, sustainability, availability, and assessability. Among these, high computational efficiency is about how much computing power can be provided as a whole, while high intelligence efficiency is about supporting a higher level of intelligence with the least amount of computing power. Public data shows that the average GPU utilization rate of intelligent computing centers in the traditional model is below 30%. High computational efficiency is about solving the problems of low computational efficiency and low utilization rate of computing power resources. Currently, the development of large models is flourishing, and the intelligent emergence ability of large models has been greatly improved. However, the computational efficiency of existing large models is generally not high. In the current situation where the contradiction between computing power supply and demand is prominent, improving computational efficiency and achieving higher intelligence with lower computing power is the core goal of improving intelligence efficiency.
Moving from Big to Strong
Zhao Zhiguo stated that it is necessary to improve the construction layout mechanism, formulate development plans for major regional development strategies, and promote efficient complementarity and collaborative linkage between eastern and western computing power. Improve the diversified configuration mechanism, formulate development plans for emerging industries such as artificial intelligence, build a batch of intelligent computing centers, and reasonably configure the proportion of general computing, intelligent computing, and supercomputing. Improve the interconnection mechanism, build rules for computing power interconnection procedures, market rules, and quality inspection rules, and promote the development pattern of serving the unified large market construction with the large market system of computing power.Yu Xiaohui stated that it is necessary to further optimize the structure of computing power supply and the layout of new infrastructure, enhance the construction level of computing power centers, improve the relevant policy and standard system, strengthen technological innovation and R&D capabilities, and rely on platforms such as the "Smart Computing Ecosystem" to accelerate industry empowerment and ecological improvement, thereby accelerating the high-quality development process of the comprehensive computing power 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, architectural design, large model algorithms, and high-speed transmission protocols, and establish a sound computing power industry standard system. In key links such as research and development design, production manufacturing, warehousing logistics, and marketing services of industrial enterprises, a number of typical application scenarios and solutions empowered by computing power will be cultivated. A number of computing power application solution providers will be cultivated for small and medium-sized enterprises, and a number of service platforms integrating multiple resources and opening multiple capabilities will be built. Replicable and promotable development models will be formed in key areas such as industry, education, transportation, and energy.