China’s homegrown 100,000-card AI supercluster is advancing the integration of artificial intelligence and supercomputing, providing a powerful platform for scientific research and large-scale AI applications.

Completed and put into operation in July, the Sugon Dawning 8000, also known as Dengfeng, combines the two computing approaches on a unified infrastructure. Weighing 1,500 tonnes and containing 1,600 kilometres of internal cables, the system is housed in a 2,000-square-metre space.

Its high-speed network connects 100,000 AI computing cards, enabling them to work together on complex tasks, from high-precision scientific simulations to AI model training and inference.

The system is already handling more than 6,000 computing tasks simultaneously. Engineers are now working to improve its efficiency and ease pressure on heavily loaded computing resources.

Their efforts are also delivering results in weather forecasting. By optimising algorithms and making better use of computing power, engineers have cut a model’s response time from 10 milliseconds to three, roughly tripling its speed.