Computing power is in short supply, how to achieve low-carbon and efficient development?

"The growth rate of smart computing in 2023 will exceed 70% compared with 2022. The supply is still in short supply, and the market demand for computing power is hot."

With the various developments of large models, smart computing has become a hot topic. According to data released by the China Computing Power Platform, as of the end of 2023, China's total computing power is 23 billion EFLOPS(computing power indicator), of which the proportion of smart computing has reached 30%. On the other hand, tight computing power has become a "pain point" for many large model manufacturers. At the 2024 Global Digital Economy Conference Intelligent Computing Innovation and Development Forum held on July 4, many experts discussed this.

Computing services are constantly sinking

Currently, computing services are constantly sinking.

Li Jie, deputy director of the China Academy of Information and Technology, said that the demand for smart computing market is very strong, with a domestic growth rate of 30%. Overseas markets also have huge potential. According to tests by Google, Temasek and Bain, by the end of 2023, Southeast Asia's digital economy revenue is expected to be US$100 billion.

Wen Ku, chairman of the China Communications Standardization Association, pointed out that in order to stimulate the development potential of my country's smart computing power industry, it is necessary to strengthen the industrial ecological construction of smart computing power, promote efficient collaboration among operating enterprises, social manufacturing enterprises, and industry regulatory authorities, and promote industry-university-research Deep overlap of application, promote the coordinated development of cloud and edge computing power, strengthen the leading role of smart computing power, optimize the computing power structure, and improve the empowerment level of smart computing power applications.

Wang Xiaoli, deputy secretary of the Party Committee of the China Academy of Information and Communications, said that as computing power penetrates into hundreds of industries, innovative application points of computing power in traditional industries should be actively explored, such as in-depth layout in industrial Internet, smart medical care, urban management and other fields. "Encourage cross-industry and cross-field R & D cooperation, strengthen the communication and cooperation mechanism for multi-industry integration through communication platforms such as the 'Computing Power Industry Platform', accelerate industrial digitalization and intelligent transformation, and promote high-quality development of the real economy."

Large model energy consumption makes carbon emissions issues increasingly prominent

But on the other hand, as computing power demand grows, electricity is becoming one of the constraints. Many manufacturers have begun to "worry" about the energy consumption problem of large models.

Yang Zhichao, general manager of the strategic development department of Vision Link Power, said at the forum,"We have encountered some problems in practice. Low power supply is an important indicator for building a low-cost computing power center. The prerequisite for the widespread use of computing power may be extensive. Can the cost of electricity be reduced?"

At the same time, this has also made the issue of carbon emissions increasingly prominent.

The British "Financial Times" reported that Google pointed out in its annual environmental report on July 2 that the company's greenhouse gas emissions will reach 14.3 million tons of carbon equivalent in 2023, 48% higher than the 2019 benchmark and 13% higher than 2022. According to a report by the Guardian on July 2, Microsoft's president admitted in May that the company's emission reduction goals were affected by its artificial intelligence strategy. Microsoft's carbon emissions have increased by nearly one-third since 2020, largely due to the construction of data centers.

How can computing power be low-carbon and environmentally friendly?

"Currently, the demand for efficient and low-consumption intelligent computing resources is becoming increasingly urgent." Wang Xiaoli believes that we should focus on basic research and promote innovation in computing power technology. Increase investment in research and development in areas such as intelligent computing architecture, artificial intelligence chips, and artificial intelligence algorithms.

"Our country is still developing efficiently in this area." At the forum, Li Jie calculated,"As of the end of 2023, the national data center consumes 150 billion kilowatt-hours of electricity a year. In recent years, the growth rate of data center power consumption is about 15%, but the entire computing power market is developing rapidly, far exceeding 15%."

Mao Dongjun, member of the Party Leadership Group and Deputy Director of the Beijing City Bureau of Economy and Information Technology, said that Beijing continues to expand the scale of computing power cooperation between Beijing, Tianjin, Hebei and Mongolia, and plans to build ultra-large-scale smart computing clusters that support large-scale training needs of trillion-level parameters. At the same time, we also actively explore efficient allocation of computing power resources and green and sustainable development paths, striving to achieve energy conservation and emission reduction while ensuring the supply of computing power.

At the same time, many experts believe that early establishment of industry standards is conducive to the healthy development of computing power.

Wen Ku pointed out that it is necessary to improve the level of industrial standardization, increase technological innovation, and on the basis of practice, accelerate the formulation of international standards, national standards, industry standards, and group standards related to the intelligent computing power industry to form a complete standard system.

Wang Xiaoli also said that in order to deal with the current problems such as inconsistent interfaces and lack of standards in some fields in the development of computing power, it is necessary to conduct in-depth research on the characteristics and needs of computing power resources, formulate a set of scientific, reasonable and forward-looking standards system, and accelerate the formulation and implementation of a series of standards and specifications.