China uses 100,000 home-grown AI chips to build leading weather forecast system

China has developed a cutting-edge weather forecasting system capable of predicting weather up to 10 days in advance at an unprecedented resolution of 5km (3.1 miles), surpassing most major global systems. This achievement was made possible by a domestically built computing system known as Sugon 8,000, which utilizes 100,000 home-grown AI chips.

Most global forecasting systems currently have resolutions ranging from 5km to 10km. For example, the US National Weather Service's Global Forecast System has a maximum resolution of around 13km. The Chinese team's latest test using Sugon 8,000 achieved a 10-day global forecast in approximately one hour, meeting the requirements for routine weather forecasting.

The system's ability to handle smaller grid cells allows for more detailed weather system observations, though it comes with a significant computational cost. China aims to achieve 1km forecasts nationwide and resolution down to 100m (328 feet) in key areas by 2030. The Mesoscale Convective Vortex model, used by the Chinese system, was adapted to run on Sugon 8,000, a computing cluster that links 100,000 powerful computing units capable of simultaneous calculations.

This unique design allows Sugon 8,000 to handle different types of computing, combining the precision of scientific simulations with the speed needed for processing large data sets. The system has already been used for various applications, including protein research, molecular dynamics, and turbulence studies, demonstrating its versatility beyond weather forecasting.