Numerical weather simulation demands fast and high precision simulation over fine-grained grids to improve the accuracy of weather prediction. Because Graphics Processing Units (GPUs) offer extremely huge computational performance compared to conventional CPUs, GPUs have attracted a lot of attention in recent years as devices to accelerate many important classes of scientific applications, including numerical weather predictions.

In order to drastically shorten the runtime of a weather prediction code ASUCA developed by the JMA (Japan Meteorological Agency) for the purpose of the next-generation weather forecasting service, the entire parts of the huge code are re-written for GPU computing from scratch. By introducing many optimization techniques and several new algorithms, very high performance of 145 TFlops has been achieved with 3990 GPUs on the TSUBAME 2.0 supercomputer at Tokyo Tech. It is quite meaningful to show that the GPU supercomputing is really available for one of the major applications in the HPC field.

This figure demonstrates the real case of the ASUCA operation with both the real initial and the boundary data used for the current weather forecast at the JMA. This simulation was performed with a 4792 x 4696 x 48 mesh with horizontal mesh resolution of 500 meters using 437 GPUs of the TSUBAME 2.0 in single precision.