11. GPU Acceleration of a Non-Hydrostatic Ocean Model with Lagrangian Particle Tracking
Authors: Takateru Yamagishi (Research Organization for Information Science and Technology)Yoshimasa Matsumura (Hokkaido University)
Abstract: To achieve detailed simulations of several different types of particles in numerical ocean simulations, we have implemented and optimized a non-hydrostatic ocean model with Lagrangian particle tracking on an NVIDIA GPU. We have revised the algorithm for particle tracking and optimized the kernels for ocean dynamics calculation. The sorting of all particles at every time step was introduced to coalesce the access to the GPU global memory, and a texture cache was assigned to the ocean current velocity array to accelerate particle tracking. Thread-level parallelism was exploited with the help of additional calculations, and registers were effectively used for the ocean dynamics calculation. When comparing its execution on a Fujitsu SPARC64 IXfx to that on the NVIDIA K20C, the GPU-implemented model was three times faster. This model successfully reproduced the nonlinear distribution of the particles in the ocean.
Two-page extended abstract: pdf