Sugon Chairman: Bridging the "Last Mile" of Computing Power Applications
The Chairman of Sugon, in an article published in People's Daily, proposed that China needs to strengthen domestic "large computing power" systems and refine intelligent computing services to bridge the "last mile" of computing power applications. The article noted that more than 500,000 GPU-card computing clusters will be deployed overall this year, among which 30,000-card clusters grew more than twofold year-on-year, and 100,000-card clusters achieved a "from 0 to 1" breakthrough; the fully domestic 100,000-card AI supercluster Sugon 8000 (Dengfeng) was completed in July this year. Intelligent computing services are showing trends such as software-hardware collaboration, supercomputing-intelligence integration, and a shift in delivery models from "machine-hours" and "card-hours" to "tokens."
Body · AI translation
Computing infrastructure is an important foundation for the leap in intelligent productivity, and intelligent computing services are a key lever for accelerating the transformation of production relations.
As artificial intelligence moves from the "digital world" to the "physical world," industries across the board are ushering in a new round of productivity transformation and leapfrogging, and the intelligent economy has become the core engine for activating the real economy. Against the backdrop of the transition between old and new growth drivers, the large-scale deployment and value creation of artificial intelligence require us not only to build strong new-type infrastructure capabilities, but also to innovate service systems for new quality productive forces scenarios, so as to bridge the "last mile" of computing application.
The State Council executive meeting held in June pointed out that efforts should be intensified to advance breakthroughs in artificial intelligence innovation, accelerate key technology research and the construction of ultra-large-scale intelligent computing clusters, strengthen the supply of high-quality data, reinforce guarantees for factors such as talent and funding, and support enterprises in carrying out basic research and frontier exploration. It is necessary to deeply implement the "AI+" initiative, leverage China's advantages such as a complete industrial system and rich application scenarios, and promote the accelerated large-scale commercial application of intelligent products and services.
In the era of artificial intelligence, the supporting role of computing infrastructure is further strengthened. Building strong domestic "large computing power" systems is an important foundation for accelerating breakthroughs in artificial intelligence technology and application exploration. On the other hand, the scenario-based deployment of artificial intelligence requires refining and optimizing the intelligent computing service system, unleashing the effectiveness of "large computing power," promoting a flourishing of computing applications with novel models and distinctive features, and transforming computing power into real productivity.
This year's Government Work Report proposed "implementing new infrastructure projects such as ultra-large-scale intelligent computing clusters and computing-power-electricity coordination." Against this backdrop, the pace of developing domestic "large computing power" will continue to accelerate—according to relevant plans, a total of more than 500,000 cards' worth of computing clusters will be deployed this year, among which 30,000-card clusters will grow more than twofold year on year, and 100,000-card clusters will achieve a breakthrough "from 0 to 1."
For example, China's first fully domestic 10,000-card artificial intelligence supercluster—Sugon 8000 (Dengfeng)—was completed in July this year. It was built on full-link, fully self-developed AI infrastructure capabilities spanning "chips, computing, storage, networking, cooling, applications, and services," achieving native integrated fusion of all types of computing and covering various scientific research and industrial scenarios such as scientific computing, large-model training and inference, and industrial simulation.
English text is being prepared.
These new trends require us to continue focusing on the full-stack ecosystem of the computing industry and build more complete computing infrastructure and service systems for the large-scale application of artificial intelligence.
Specifically, first, we will continue to advance the development of a series of chips and core components, forming an innovation trend from general-purpose to easy-to-use, from compatibility to innovation, and from following to leading, and continuously enriching the chip product portfolio; second, we will build a new generation of globally leading advanced computing systems, achieving "full-scenario coverage, integration of quantum, supercomputing, and intelligent computing, integrated space-air-ground, and tight coupling of computing and electricity"; third, we will continuously upgrade computing services, break through application barriers, and rely on platforms such as the national supercomputing internet to bridge the "last mile" for domestic large computing power to reach all industries.
Computing infrastructure is an important foundation for the leap in intelligent productivity, and intelligent computing services are a key lever for accelerating the transformation of production relations. As global intelligent computing services accelerate their evolution toward an ecosystem, China's artificial intelligence industry urgently needs to transform technological accumulation into service advantages and comprehensively improve the large-scale supply and inclusive level of computing services. I believe that in the future, domestic "large computing power" that is more widely applied and smarter will enter thousands of households and serve all industries, realizing greater value in advancing the construction of a digital China and improving the level of digital and intelligent development.
(The author is Chairman of Sugon and Director of the National High Performance Computer Engineering Technology Research Center. This article was compiled from an interview by our reporter Gu Yekai.)