Based on the latest artificial intelligence theory and advanced computing architecture, intelligent computing power is providing strong support for the development of artificial intelligence. The "Proposal of the Central Committee of the Communist Party of China on Formulating the 15th Five Year Plan for National Economic and Social Development" passed at the Fourth Plenary Session of the 20th Central Committee of the Communist Party of China proposes to "accelerate innovation in digital intelligence technologies such as artificial intelligence, break through basic theories and core technologies, and strengthen efficient supply of computing power, algorithms, data, etc." According to data from the Ministry of Industry and Information Technology, as of the end of June this year, the total scale of computing power center racks in use in China has reached 10.85 million standard racks, the scale of intelligent computing power has reached 78.8 trillion times per second, the storage capacity has exceeded 1680 exabytes, and 1509 large models have been released, ranking among the top in the world. At the same time, the application of intelligent computing power is deepening, and its coupling with various industries is constantly increasing, becoming a new foundation for the digital economy. Currently, what are the main aspects of intelligent computing power applications? How can new computing power products and service models provide strong support for high-quality development? The reporter conducted an interview. Driving the paradigm shift in scientific research and accelerating the generation of original innovative achievements, researcher Shen Enzhi is conducting research on the mechanisms and biological functions of non coding RNA (ribonucleic acid) in the Non Coding Nucleic Acid Biology Laboratory at West Lake University. Non coding RNA accounts for approximately 98% of the human transcriptome, and it not only participates in various basic life processes of living organisms, but is also closely related to the occurrence of many major diseases. ”Introduction to Shen Enzhi. Non coding RNA has a large quantity and is highly dynamic in living organisms, allowing it to interact and regulate with many other biomolecules. If traditional experimental methods are used for research, it requires a lot of time and effort, and it is difficult to analyze the complex regulatory relationships and find patterns. Nowadays, the Shen Enzhi team utilizes intelligent computing technology to efficiently analyze sequencing data, solving problems that could not be solved solely through experiments and greatly accelerating the research process. In recent years, West Lake University has built a leading computing power platform with the assistance of enterprises such as Inspur Information, supporting scientists to conduct interdisciplinary research. Shen Enzhi believes that interdisciplinary research between artificial intelligence and life sciences, especially computational biology, has significantly improved the speed and accuracy of scientific research, making many new discoveries impossible. Relying on intelligent computing power to achieve efficient model training and inference greatly accelerates the application innovation of artificial intelligence in the field of scientific research. Domestic universities and computing power enterprises are also accelerating the pace of industry university research cooperation, promoting the landing of intelligent computing power in the field of scientific research. The Intelligent Computing System Research Laboratory of Nankai University has improved the inference speed of a single fundus image by 2.4 times through the use of an efficient artificial intelligence computing open-source framework in the study of retinal image vascular segmentation, thereby promoting the clinical application of fundus image analysis technology; The Shanghai Artificial Intelligence Laboratory collaborated closely with Lingang Laboratory, Shanghai Jiao Tong University, Fudan University, and others to discover and validate two new cancer targets in just two months; Pengcheng Laboratory integrates computing power nodes from more than 20 cities through the China Computing Power Network, and relies on "Pengcheng Cloud Brain II" to serve tens of thousands of researchers... In the view of Sun Maosong, Executive Vice President of Tsinghua University's Artificial Intelligence Research Institute, artificial intelligence is expected to become an important tool for driving scientific paradigm innovation. In the paradigm shift driven by artificial intelligence in scientific research, it is important to be adept at utilizing intelligent computing power to discover "root node" problems, such as predicting protein structures with large models. "Solving these problems may bring fundamental changes to the field. The implementation of achievements is accelerating and adapting to diverse application scenarios. The State Council has issued the "Opinions on Deepening the Implementation of the 'Artificial Intelligence+' Action, 'which proposes to strengthen the overall planning of intelligent computing power and promote the supply of intelligent computing power that is universally accessible, economically efficient, green, and safe. The Yili Modern Smart Health Valley located in Hohhot, Inner Mongolia, has exclusive "health records" for the cows on the pasture. Based on the artificial intelligence computing power provided by Alibaba Cloud and the Feitian operating system, Yili has built an intelligent computing infrastructure of "one cloud, multiple cores, one cloud, multiple calculations", providing a stable and flexible deployment environment for artificial intelligence applications, and integrating decision algorithms and generative models into the production and supply chain fields. Relying on manual observation of the health status of cows is inefficient and subjective. Now, by using computer vision to recognize the eye characteristics of cows, we can monitor their health in real time and manage their feeding accurately, effectively improving the quality of milk production. ”Cheng Guoqiang, Director of Digital Technology at Yili Group's Digital Technology Center, introduced that in Yili's smart factory, intelligent computing power empowers automated production lines, continuously improving production and management efficiency. Currently, Yili has developed over 800 intelligent agents. The large model covers 70% of the supply chain scenarios, including order fulfillment, inventory turnover, and logistics timeliness, significantly reducing the risks of raw material lead times, backlogs, and stockouts. ”Jia Chaohui, Senior Director of Alibaba Cloud Solutions, said that in the future, they will also create an "intelligent computing cloud" and combine professional data and industry knowledge such as grass planting and cow health to jointly develop a large-scale model for the agricultural and animal husbandry industry. It is reported that since the first open source model was released in 2023, Alibaba's Tongyi Big Model has been downloaded over 600 million times worldwide, and the number of derivative models has exceeded 170000. With the continuous improvement of speed levels of high-speed trains, aerodynamic design has become a key factor in research and development. CRRC built a standard database of aerodynamic loads for high-speed EMUs based on existing simulation and test data. Relying on the Baidu PaddlePaddle in-depth learning platform and the PaddlePaddle scientific calculation kit, it built an intelligent aerodynamic simulation model. Chen Jian, Deputy Director of the Digital Innovation Department of the Science and Technology Quality and Information Technology Department of China Railway Rolling Stock Corporation (CRRC) and Head of the Artificial Intelligence Office Promotion Group, introduced that compared with traditional simulation methods, the simulation cycle of large models has been shortened from days of supercomputing resources to 10 seconds of a single graphics card, and the overall simulation efficiency has been improved by more than 30 times. The result error is less than 5%, which promotes the transformation of China's high-speed train exterior design from experience driven to data-driven, and promotes intelligent selection and rapid optimization of high-speed train exterior design. Baidu's AI basic technology base based on Wenxin Model and PaddlePaddle deep learning platform is helping thousands of industries to improve efficiency and reduce costs. Wang Haifeng, chief technology officer of Baidu, introduced that PaddlePaddle supports large-scale adaptive distributed deep learning training, and the effective training time of large models on the Wanka cluster accounts for 98%. Through the full chain collaborative optimization of large model compression, reasoning, and service deployment, the reasoning of large models with low latency and high throughput is realized. In addition, PaddlePaddle continued to actively carry out unified hardware adaptation, providing a strong guarantee for diversified computing power. Collaboration between industry, academia, and research, new products and models continue to emerge. Recently, Inspur Information released an artificial intelligence computing system optimized for intelligent agents. Among them, based on the meta brain SD200 super node artificial intelligence server, the DeepSeek R1 large model word element (the basic unit of model input and output) generation speed only takes 8.9 milliseconds; The latest release of the metabrain HC1000 super extended artificial intelligence server significantly reduces the cost of generating lexical elements. A supernode server is a compact computing unit that creates a logical unity within a limited physical space through extremely low latency bus level interconnect technology and minimalist interconnect architecture. The entire system can operate like a super sized computer. The key to developing supernodes is the innovation of computing architecture, which follows the development direction of applications and algorithms, is application-oriented, system centered, and effectively solves the bottleneck problem of intelligent computing power development. ”In the view of Liu Jun, Chief Artificial Intelligence Strategy Officer of Inspur Information, the widespread application of artificial intelligence continues to drive rapid growth in computing power. Developing intelligent computing power requires overcoming challenges such as system scale approaching engineering limits and high energy consumption, effectively balancing input and output. In the opinion of industry experts, only by continuously promoting innovation and breakthroughs in artificial intelligence computing architecture through software and hardware collaborative design and deep optimization can we further promote the deep integration of artificial intelligence technologies such as large models and intelligent agents with the real economy, and make artificial intelligence an innovative force that empowers thousands of industries. In 2024, under the organization of the Zhongguancun Management Committee of the Beijing Municipal Science and Technology Commission, several enterprises and institutions in the fields of chip, system, large model development, and application innovation, including Inspur Information, established the "Hypernode Computing Power Cluster Innovation Alliance". Relevant units closely cooperated and made significant progress in hypernode interconnection protocols, system development, standard formulation, application deployment, and other aspects. As one of the participants, Taichu (Wuxi) Electronic Technology Co., Ltd. actively collaborates with research institutions, operators, and computing power service providers to create intelligent computing solutions. Its high-density liquid cooled intelligent computing cluster has served more than 200 universities and enterprises, supporting over 120 quadrillion floating-point computing power demands per second. The innovation breakthrough of intelligent computing power cannot be achieved without the collaboration of industry, academia, and research, as well as talent support, "said Hong Yuan, the Chief Product Officer of the company. They have established a joint research and development team with Hunan University, Nanjing University of Posts and Telecommunications, Suzhou University, and others to jointly carry out scientific research and development. In addition, they have also built a teaching and training platform based on self-developed artificial intelligence acceleration cards and software hardware collaboration, cultivating a large number of young, dynamic, and practical talents. (New Society)