China's First Fully Domestic 100,000-Card AI Supercluster "Sugon 8000 (Dengfeng)" Debuts at WAIC
Sugon's "Sugon 8000 (Dengfeng)" appeared at WAIC 2026: fully domestic, HPC-AI fused, FP64–INT8 full precision, powered by Hygon and other domestic chips, with immersion phase-change liquid cooling and full-stack self-development.
A "museum-piece" domestic supercluster
Per IThome (July 17), at WAIC 2026 in Shanghai, Sugon unveiled China's first fully domestic 100,000-card AI supercluster, "Sugon 8000 (Dengfeng)," in the flesh — named a conference "treasure exhibit."
Technical route: HPC-AI fusion, FP64 to INT8
Sugon 8000 (Dengfeng) takes an "HPC-AI fusion" route for combined high-precision scientific computing and low-precision AI, supporting FP64 through INT8 full precision across scientific computing, large-model training, AI inference and industrial simulation — spanning quantum, astronomy, automotive, robotics and large models.
Key specs: domestic base + liquid cooling
- Compute base: powered by Hygon and other domestic chips;
- Network: a scaleFabric IB-like native RDMA high-speed fabric interconnecting the 100,000-card cluster;
- Storage: ParaStor distributed storage for the massive I/O of training and scientific computing;
- Cooling: immersion phase-change liquid cooling, supporting MW-class per-rack density, with domestic coolant and year-round free cooling to boost efficiency.
Significance: full-stack self-development, national supercomputing grid
The cluster claims full self-development across "chip, compute, storage, network, cooling, application, service," supports the Gridview management platform, and can plug into China's national supercomputing internet. Against constrained advanced-process and high-end memory supply, a 100,000-card, chip-to-cooling all-domestic cluster marks a substantive step for China's AI infrastructure toward "scale plus self-control."