The Bund Conference: Breaking through the computing power bottleneck of general artificial intelligence and green computing becoming a consensus

Since this year, the large model industry has experienced a blowout development. At present, more than 100 large models have been publicly released in China, which has accelerated the scarcity of intelligent computing power and led to the situation of "hard to find a GPU card" and "queuing up for reservation".

How will the industry respond to the explosive demand for intelligent computing power and the resulting surge in energy consumption? At the 2023 inclusion·The Bund Conference "Green Computing Forum in the Cloud Native Era" held on September 7, guests from the fields of government, industry, academia, research, and application shared the development direction under new trends.

At present, China attaches great importance to the green development of the computing power industry from a policy perspective. He Feng, deputy director of the Shanghai Communications Administration Bureau, proposed that Shanghai will carry out green computing power-related work around the "Computing Power Pujiang Action Plan", and proposed to accelerate the research and development of key green computing technologies, improve data center resource utilization; speed up the standardized measurement of data centers and definition; three key suggestions to deepen industry collaboration and jointly build a new green computing ecology.

(Picture: He Feng, Deputy Director of Shanghai Communications Administration delivering a speech)

Xie Ji, secretary-general of the China Association of Small and Medium Enterprises and former inspector of the Climate Department of the National Development and Reform Commission, said that the China Association of Small and Medium-sized Enterprises attaches great importance to the development of small and medium-sized enterprises in the green and low-carbon economy. The Association and relevant units in Shanghai jointly released the "SME ESG Information Disclosure Guidelines" and "SME ESG Star Rating" group standards to promote the green and low-carbon transformation of the majority of small and medium-sized enterprises, practice "green computing", and achieve "carbon neutrality" in the computing field. and".

Ant Group proposed to pay attention to the "end-to-end green computing power system", especially the green computing in the computing power application stage, to drive the breakthrough of computing efficiency bottlenecks and achieve green coordinated development. This proposition was also highly praised by the guests attending the meeting. recognized.

End-to-end green computing refers to green computing in the entire industry from power production and computing power production (including intelligent computing center builders, hardware manufacturers, and cloud manufacturers) to computing power application. In the "White Paper on Green Computing for Computing Power Application Links" released by the China Academy of Information and Communications Technology at the 2023 Computing Power Conference, it was mentioned that green computing in the computing power application link has the largest room for improvement in intelligent computing efficiency and the fastest results, and it has not yet been sufficiently recognized. areas that still need to be strengthened.

At the level of computing hardware infrastructure, the industry has achieved remarkable results through measures such as architecture upgrades and reduction of data center energy consumption. Duan Xiaodong, deputy director of China Mobile Research Institute, said that China Mobile has built a new intelligent computing center from the five core aspects of "new Internet, new computing efficiency, new storage, new platform, and new energy saving" and built a unified standard, leading technology, compatible and open Intelligent computing base, "To face the explosion of intelligent computing power, it is necessary to build a computing power network through an open, collaborative, inclusive, and win-win approach to promote the construction of a better green and energy-saving computing power foundation in the future intelligent computing era."

Using AI technology to achieve energy-saving optimization of data centers and related industries is also an important practical direction. Liu Yunxin, Professor Guoqiang of Tsinghua University and chief researcher of the Intelligent Industry Research Institute, and Zhang Yimin, chief scientist of Intel China Research Institute and director of the Academic Cooperation Center, shared the latest results, including core enabling technologies in two directions: energy-efficient intelligent computing systems and utilization AI improves quality and efficiency, and then creates an intelligent computing platform for "dual carbon", using data-driven intelligent decision-making to empower the greening of the industry.

Statistical data estimates that after the data is generated, the proportion that can be effectively retained, used and brought into play is only 0.2%. Large amounts of data cannot effectively generate business value, and the bottleneck lies in current data computing and processing capabilities. In this regard, Jin Hai, a professor at Huazhong University of Science and Technology, vice chairman of the China Computer Federation, and IEEE Fellow, proposed three models of near-data processing from the perspective of new data technologies: storage moving up, computing power sinking, and distribution expansion, in order to crack the "storage wall." "Effective solutions are proposed for performance bottlenecks. Looking into the future, Jin Hai believes that "data-centric computing architecture will become the trend of innovating big data processing models, which will also greatly optimize the computing power efficiency in data processing."

(Picture: Jin Hai, Professor of Huazhong University of Science and Technology, Vice Chairman of China Computer Federation, and IEEE Fellow, delivered a keynote speech)

In the era of big models, technology companies are at the forefront of the computing power shortage. He Zhengyu, chief technology officer of Ant Group and president of the platform technology business group, said that since the era of CPU general computing, Ant Group has begun to build a green and efficient intelligent computing system, forming low-carbon computing power selection, global resource scheduling optimization, workload optimization, green Monitor and evaluate the four core capabilities, which will provide "weights and measures" for green computing in the era of intelligent computing. He Zhengyu introduced that in 2022, this system will support the intelligent scheduling of computing power of 3 billion PV recommended on the Alipay homepage every day, saving 41.4% of machine resource consumption throughout the year. "In the era of large AI models, technology is sinking, new optimization space is opened, and targeted optimization of the underlying software stack should be paid attention to."

(Picture: He Zhengyu, Chief Technology Officer of Ant Group and President of Platform Technology Group, delivered a keynote speech)

End-to-end green computing requires collaboration between upstream and downstream industries to jointly build a green computing ecosystem. Ren Xiang, director of the Integrated Circuit Evaluation Center of China Electronics Technology Standardization Institute and leader of the standard testing group of the Green Computing Industry Alliance, said that gathering upstream and downstream resources in the industry chain through the Green Computing Industry Alliance can effectively promote standard innovation, talent training, international cooperation and technology Product development and application implementation.

Paying attention to the issue of carbon perception and constructing a green measurement and evaluation method is also the first step to systematize industrial development. Wang Chaoyang, General Manager of Alibaba Cloud IDC R&D Division, Zhao Shuai, General Manager of Inspur Information Server Product Line, and Zhu Youhong, Director of Data Center Operations Department of PICC Technology, each shared different opinions from the perspectives of green data centers, hardware and data processing, and constructively discussed It proposes to construct correlation indicators between computing power and carbon emissions, providing an industry reference for effective measurement of green computing practices.

At the Bund Conference, the value of computing power as the foundation for digitalization and deep intelligence has been recognized. Under the general trend of computing power resources being in short supply, it has become an industry consensus to drive intelligent computing power to break through bottlenecks through software and hardware collaboration.

The above is the detailed content of The Bund Conference: Breaking through the computing power bottleneck of general artificial intelligence and green computing becoming a consensus. For more information, please follow other related articles on the PHP Chinese website!