China's AI supercluster completes 10-day global forecast in under an hour

At that resolution, the model divides the global atmosphere into grid cells about five kilometers apart. That allows forecasters to show the spiral structure of typhoons and the movement of weather fronts much more clearly.

As grids become denser, computing units must exchange far more data, and simulation results must be written to storage promptly. Communication, storage and task allocation play an increasingly important role in overall efficiency, so simply adding more processors may not speed up the entire forecast proportionally.

Major modeling efforts are turning to accelerated computing platforms as they push toward global kilometer-scale forecasting. The "Sugon 8000" system combines general-purpose and accelerated processors and verifies that results from different computing units are strictly consistent, meeting requirements for both computing efficiency and result reliability,.

Read More:Founder of Chinese startup Spirit AI says robot brains set for 2027 breakthrough. The achievement is an important collaborative innovation between China's self-developed numerical model and domestically produced computing power, the announcement said. It lays a foundation for continued efforts toward kilometer-scale and hundred-meter-scale forecasting in this field.

Sugon Chairman Li Jun said the company aims to make high-performance computing an important infrastructure for understanding the Earth and foreseeing the future. "The value of advanced computing should be shown in solving major scientific problems and supporting industrial innovation," Li said.

The applications of the "Sugon 8000" extend beyond weather forecasting. More refined and timely forecasts affect flood-season preparedness, agricultural production, flight scheduling and power supply, so meeting computing time targets provides an important condition for these fields,.

Large-scale numerical simulation is also necessary for processes that are difficult to observe directly or reproduce through repeated experiments, such as structural changes in proteins and fluid movements in turbulent flows. The "Sugon 8000" has carried out large-scale computing projects in areas including protein research, molecular dynamics and turbulence simulation.

It supports nearly 200 industry applications and more than 500 mainstream large models, covering more than 30 scientific and industrial fields. Li said Sugon will deepen cooperation with research institutions, closely integrate computing technology with research needs, and provide tools for scientists to explore more complex problems.