Currently, the wave of technological innovation led by AI large models provides a historic opportunity for the digital and intelligent upgrade of industries. As related research and development work gradually shifts from technology to application, AI large models are increasingly moving towards industrial applications, thereby triggering a revolutionary change in the global computing power structure. According to data from the China Academy of Information and Communications Technology (CAICT), the proportion of intelligent computing power has grown from about 10% to over 60%, with an annual growth rate exceeding 100%. Notably, China's development in this field is particularly rapid. In 2023, the global computing device computing power reached a scale of 1369EFlops, with China contributing 450EFlops, accounting for one-third of the global computing power, and growing at a rate of nearly 50%.

Green Computing Power Injected into "Super Projects"

To promote the domestic application of AI large models, the "national team" represented by China Mobile shoulders the main responsibility by actively deploying and building a new intelligent computing service system. Through measures such as construction of intelligent computing centers in locations such as Hohhot and Guiyang nationwide, and integrated computing power scheduling, they advance the implementation of national strategy of “Channeling Computing Resources from the East to the West”, directly addressing the pain points and difficulties encountered by AI large models during their application in enterprises.

As a representative of domestic green intelligent computing infrastructure providers, Shenzhou Kunpeng is deeply involved in the construction of China Mobile data centers. Based on the "KunTai pod2000 liquid-cooled cabinet" built with "Kunpeng+Ascend" processors, it creates a green, high-performance intelligent computing base, promoting the inclusiveness, industrialization, and large-scale development of AI large models. Taking China Mobile's Hohhot Data Center as an example, as the largest single intelligent computing center among global operators, it was selected as one of the "Top Ten Super Projects of State-Owned Enterprises in 2023." It introduces several technologies such as high-speed lossless network and cold-plate liquid cooling, supports the interconnected parallel training of ten thousand AI acceleration cards, and achieves an AI chip localization rate of over 85%, with an intelligent computing power scale reaching up to 6.7EFLOPS.

The green intelligent computing solution built based on the "KunTai pod2000 liquid-cooled cabinet" significantly improves the system’s overall cooling efficiency through technologies such as integrated cold plates and intelligent flow regulation, achieving a PUE of 1.15, effectively reducing energy consumption of various nodes by 30%. This equates to saving approximately 7.36 million kilowatt-hours of electricity per 100P intelligent computing unit annually. Meanwhile, the "KunTai pod2000 full liquid-cooled cabinet" boasts a maximum single cabinet power of over 60KW+, achieving an energy efficiency ratio 1.5 times that of the industry average, balancing "greenness" with "high performance." During actual delivery, the "KunTai pod2000 full liquid-cooled cabinet" adopts an integrated delivery method, effectively reducing deployment and operation complexity, greatly enhancing the construction and maintenance efficiency of intelligent computing centers, and lowering delivery and operation and maintenance costs.

Green Intelligent Computing for “Jiutian Large Model”

In terms of computing power application, the intelligent computing power provided by Shenzhou Kunpeng will serve as the computing base for the "Jiutian Natural Language Interaction Large Model," supporting super-large-scale, ultra-long stable, and ultra-high-efficiency large model training tasks.

As the first large model developed by a state-owned enterprise to pass both the national "Generative Artificial Intelligence Service Filing" and the "Domestic Deep Synthesis Service Algorithm Filing," the "Jiutian Natural Language Interaction Large Model" features enhanced industry capabilities, security and trustworthiness, and support for full-stack localization. It has released multiple versions with parameters of 9 billion, 13.9 billion, 57 billion, and 100 billion, which can be flexibly deployed across various cloud, edge, and end scenarios.

According to information from the State-owned Assets Supervision and Administration Commission (SASAC), based on the "Jiutian Natural Language Interaction Large Model," China Mobile has delivered a host of industry empowered achievements. First, it has created the "Jiutian·Zhongqing Base Large Model" open for co-construction by key industries, where several state-owned enterprises and top-tier hospitals are collaborating with China Mobile on large model development. Second, it has accelerated the construction of more than 20 industry-specific large models covering networks, customer service, home services, government affairs, travel, healthcare, industry, comprehensive governance, and enterprise communication. Among them, the Jiutian·Network Large Model has improved network tuning efficiency by over 30% in multiple provinces and increased the efficiency of handling performance issues by over 80%; the Jiutian·Customer Service Large Model has been applied on a large scale in China Mobile's 10086 online customer service, becoming the industry's first engineering case to use a large model in a massive customer service operations system; the Jiutian·Haisuan Government Affairs Large Model has been applied in the Heilongjiang digital government project, realizing provincial-level "one-stop service," "unified management," and "collaborative operations" to enhance intelligence and efficiency.

As the foundation of AI development, intelligent computing infrastructure is profoundly influencing the development and transformation of the digital economy. It is foreseeable that the industrial innovation brought by the deep integration of large models with industry scenarios will greatly energize and promote the digital economy. Committed to becoming a leading provider of innovative intelligent computing infrastructure, Shenzhou Kunpeng will continue to focus on customers, continuously launching better products, solutions, and services and injecting green intelligent computing into all industries.