03,20,2026 18views
【Smart CityOnline Brand Column】On April 14, the 60000 card AI for Science computing cluster system provided by Zhongke Dawning was put into use at the core node of the national supercomputing Internet in Zhengzhou. As the largest AI4S computing cluster in China, it combines super intelligent full stack technology capabilities to meet users' multidimensional needs from cluster performance, software model optimization, scientific research application efficiency to scientific intelligent agent development, providing guarantees for the large-scale implementation of "artificial intelligence+scientific technology" in China.
The landing of this cluster is not only a showcase of technological achievements, but also a milestone in the deep integration of artificial intelligence technology and scientific research innovation in China. We look forward to the participation of experts and industry professionals from various fields to work together and seize the development opportunities of 'artificial intelligence+' with this computing cluster landing as a new starting pointXu Hongxing, academician of the CAS Member and president of Henan Academy of Sciencesrepresent
Six major advantages to create the greatestAI4S cluster
With the development of artificial intelligence technology, AI4S is becoming a technological powerhouse and the core of future industry competition, as well as a global competitive focus. China's 15th Five Year Plan clearly proposes to use artificial intelligence to lead the paradigm shift in scientific research and seize the commanding heights of artificial intelligence industry applications. To meet the needs of users for large-scale parallel computing, trillion parameter large model training, high-throughput reasoning, etc., the Zhongke Shuguang AI4S computing cluster has six major features:
Powerful computing power: Achieve 60000 card cluster deployment and provide world-class super intelligent fusion computing power.
Comprehensive accuracy: relying on independently controllable core chips, it can support full precision calculations with 8/16/32/64 bit widths, efficiently handling high-dimensional functions and complex scientific problems.
High speed interconnection: Through the first domestic InfiniBand lossless high-speed network scaleFabric series product, it can fully meet the ultimate demand of AI4S computing clusters for high bandwidth and low latency networks.
Storage computing collaboration: Through the design of "super tunnels" and AI data acceleration, the three-layer transmission collaboration from chips, systems to applications is achieved, avoiding storage IO bottlenecks.
Flexible scheduling: The intelligent scheduling mechanism enables the system to flexibly match and schedule the computing, storage, and network resources of the cluster according to task requirements, with a concurrent job scheduling efficiency of over 10000 times per second.
Stable and reliable: relying on intelligent operation and maintenance, digital twin system, and submerged phase change liquid cooling technology, the system availability reaches 99.99%, ensuring long-term stable operation of the cluster.
thousandDouble Increasemulti-domainAI4S efficiency
Li Bin, Senior Vice President of Zhongke Shuguang, stated that the Shuguang AI4S computing cluster not only has outstanding hardware performance, but also has compatibility with mainstream computing ecosystems and extensive research application iteration support capabilities through software and hardware collaborative design, fully breaking down the barriers between traditional supercomputing and intelligent computing.
In terms of application, through extensive and deep cooperation with universities, research institutions, and enterprises, the cluster has completed large-scale parallel computing testing in multiple fields such as biology and materials, and has achieved multiple outstanding results: simulating protein folding at a scale of 30000 calories is 1000 times faster than traditional algorithms; Realize trillion atom liquid water molecule dynamics simulation with a scale of 45000 calories, breaking the world simulation scale record while increasing efficiency by more than 3 orders of magnitude; Assist in expanding the scale of turbulence direct simulation to millions of billion grids, significantly improving research efficiency.
In addition, with the implementation and operation of the cluster, the national supercomputing Internet platform (scnet. cn) has also built the largest AI4S computing infrastructure in China, linking more than 3 million CPU cores and more than 200000 GPU cards in total, and accessing the national integrated computing network scheduling system to provide inclusive AI4S computing services for universities, research institutes and enterprises across the country.
Completed in a few hoursAI4SModel development to reduce the development of scientific intelligent agentsandApplication threshold
In order to promote the engineering of AI4S, Dawning AI4S computing cluster gathers massive data sets, industry knowledge base, large models, scientific and engineering computing software tools, etc., and becomes the innovation base of AI4S ecological co construction and sharing, combining the integrated concept of "mathematical modeling" of the national supercomputing Internet and carrying the first one-stop development platform OneScience of scientific large models in China.
OneScience integrates dozens of AI4S hotspot models and datasets, allowing users to complete scientific model development in multiple disciplines such as earth science, bioinformatics, fluid simulation, and material chemistry within hours, solving pain points such as complex development environments and difficult data acquisition.
On the supercomputing Internet platform, Dawning AI4S computing cluster users do not need to face tedious software configuration and IT processes. They just need to put forward requirements through natural language. Through the "super scientific computing agent", they can automatically disassemble tasks, call models, schedule computing power, complete end-to-end delivery, and reduce the completion time of scientific research tasks from days to hours.
From the release of scaleX 10000 card super cluster in December 2025, to the launch of 30000 card national super computing Internet core node in February 2026, and to the release of 60000 card AI4S computing cluster system today, "Dawn Speed" has not only refreshed the record of computing cluster engineering deployment capability, but also achieved a technological leap from AI to AI4S, so that "AI+" can take root in all disciplines, helping China seize the commanding heights of AI industry applications.
