Sugon 8000, China’s First Domestically Produced 100,000-Card AI Supercomputing Cluster, Goes into Operation

Recently, China's first domestically produced 100,000-card AI supercomputing cluster, Sugon 8000, was officially commissioned and connected to the Zhengzhou core node of the National Supercomputing Internet. In its first week, it operated at full capacity, processing over 150,000 jobs daily and peaking at more than 500,000 jobs in a single day. It has completed adaptation for numerous super-fusion applications covering large model training, high-throughput inference, and scientific computing. At the recent 2026 World Artificial Intelligence Conference and High-level Meeting on Global AI Governance, Sugon 8000 was listed among the top ten "treasures of the venue".

The Sugon 8000 adopts a "super-fusion" technical route, integrating high-precision scientific computing and low-precision intelligent computing capabilities into the same system. The system supports precision from double-precision 64-bit down to 32-bit, 16-bit, 8-bit, and even lower, providing flexible mixed-precision calling capabilities to meet computing needs for scientific computing, large model training, AI inference, and industrial simulation. The system consists of 3 billion electronic components, with interconnect cabling totaling over 1,600 km and a total weight of 1,500 tons. Its structural assembly precision is less than 0.05 mm. From core component development to system deployment, all is domestically produced, providing full-chain self-developed AI infrastructure capabilities covering chips, computing, storage, networking, cooling, applications, and services.

The distributed storage technology used in Sugon 8000 achieved first place in both the production full-node and 10-node categories of the global storage system performance benchmark IO500 list released in June this year. The storage system can support massive data reads and writes during large model training and scientific computing, ensuring efficient operation of the large-scale cluster. Sugon 8000 submerges all computing equipment in fluorinated liquid for cooling. The fluorinated liquid has a boiling point of around 50 degrees Celsius, while the equipment operates at 80 to 90 degrees Celsius. When the equipment temperature rises, the fluorinated liquid boils and undergoes phase change, entering a condenser via transfer pipes, and the cooled gas is converted back into fluorinated liquid for closed-loop recycling. Technologies such as high-speed networking, digital twin operation and maintenance, and multi-fusion scheduling collectively enhance the operational efficiency and application adaptability of the ultra-large-scale cluster.

Upon commissioning, Sugon 8000 was connected to the National Supercomputing Internet and, through it, to the national integrated computing network, providing computing services to research institutes and industrial users. It supports scenarios such as weather forecasting, materials design, energy exploration, biomedicine, quantum computing, fluid simulation, industrial simulation, and large model training and inference. In April 2024, the National Supercomputing Internet, initiated by national ministries, was officially launched. It has built the country's largest integrated computing network and application service platform, aggregating over 3.5 million CPU cores and 250,000 GPU cards. After the Zhengzhou core node of the National Supercomputing Internet went online, it launched a universal resource package for new users, including 200 card-hours of computing power, a 10 million token call quota, and 500 GB of storage resources, along with access to mainstream large models. The National Supercomputing Internet has accumulated over 1.4 million users, with the average user age continuously decreasing, allowing everyone from middle school students to professional researchers to access high-end computing resources.

Sugon 8000 has completed deep adaptation of over 300 key applications, covering more than 20 industry sectors. In typical scenarios, the system has supported the development and deployment of a new generation of weather large models, driven the migration and optimization of core imaging algorithms in the energy sector on domestic platforms, and accelerated ultra-large-scale tasks such as protein folding simulation in biology, materials, and fluid dynamics. In recent years, Sugon has built an industrial ecosystem consortium around a domestic general-purpose computing platform, now gathering over 6,000 industry partners.