基于GPU加速的潮流模型及其在岛群二维水动力数值模拟中的应用

The application of GPU accelerated tidal model to two-dimensional hydrodynamic numerical simulation of archipelago

摘要: 本文基于GPU（Graphic processor unit）并行算法建立了非结构网格有限体积法海洋模型（GPU-OM），并实现了对长山群岛的潮流数值模拟。计算结果表明，该模型可以对岛群复杂岸线及地形条件下的潮流运动进行精细准确的计算。此外，通过对比GPU-OM、FVCOM和MIKE 21的并行加速能力，本文模型在Tesla K20X上的计算效率明显优于32线程的FVCOM模型和48线程的MIKE 21并行计算模式。

Abstract: In this work, an unstructured grid finite volume method accelerated with GPU parallel algorithm (GPU-OM) ocean model is developed.The model is implemented to simulate the tidal current situation in the Changshan archipelago water area.Simulate result shows that the GPU-OM can accurate simulate the water with complex shoreline and topography.In addition, the parallel acceleration of GPU-OM, FVCOM and MIKE21 are compared.The performance of GPU-OM executed on Tesla K20X is much better than that of FVCOM with 32 threads and MIKE 21 with 48 threads.