Development of low-power GPU-based next-generation weather prediction code

2014 Fiscal Year Final Research Report

• Project/Area Number: 25870223
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Computational science; High performance computing
• Research Institution: Tokyo Institute of Technology
• Principal Investigator: SHIMOKAWABE Takashi 東京工業大学, 学術国際情報センター, 助教 (40636049)
• Project Period (FY): 2013-04-01 – 2015-03-31
• Keywords: 気象計算 / 高性能計算 / GPU / 大規模計算 / 高生産フレームワーク / スーパーコンピュータ

Outline of Final Research Achievements

Numerical weather prediction is one of the major applications in high-performance computing and is accelerated on recent GPU supercomputers. Skillful programming techniques are required for obtaining good parallel efficiency on these supercomputers. To apply these complicated optimizations to various mesh-based applications easily, we have developed a high-productivity and high-portability framework for multi-GPU computation of mesh-based applications. By using this framework, the programmer can write user code just in a standard language and generate program code optimized for the GPU supercomputers. We have implemented the weather prediction code ASUCA on a multi-GPU platform by using this proposed framework; ASUCA is a high-resolution meso-scale atmospheric model being developed by the Japan Meteorological Agency. The framework-based ASUCA performed on TSUBAME2.5 supercomputer at Tokyo Institute of Technology has achieved high-resolution simulation results with high performance.

Free Research Field

格子法に基づいた大規模物理計算