Abstract:

[Objective] This study is to evaluate the performance of the WRF model on the Kunpeng processor for the widespread application of domestic chips in the meteorological field. [Background] In recent years, the amount of meteorological observation data has been increasing, and supercomputers have become the main support for the development of meteorology. The research on the technologies of supporting meteorological information systems by the domestic chips is helpful in promoting the wide application of domestic chips in this field. [Method] The Kunpeng 920 processor based on the ARM architecture is tested by benchmarks in three aspects: STREAM, HPL, and HPCG. Additionally, WRF, a typical application in weather forecasting, is taken as an example to demonstrate the good performance of the Kunpeng processor in model migration, computing efficiency, and scalability. [Results] According to the test results, the Kunpeng processor achieves excellent memory-access bandwidth, floating-point computing performance, and high scalability. Also the porting process on the Kunpeng platform is simple, the application performance meets users' requirements on timeliness and scalability, and the forecast result is the same as that of the existing platform. [Conclusions] Therefore, the Kunpeng processor is capable of being widely used in the meteorological applications.

Key words: high performance computing, ARM architecture, Kunpeng 920, WRF, performance evaluation