Parallel Implementation for Phase-Field Simulation of Flow Effect on Dendritic Growth with GPU Acceleration

Abstract

A Sola-phase field model combined Sola algorithm with phase-field model is established. It is difficult to implement real-time simulation as the computational grids increase. Taking pure SCN for example, the solidification microstructure evolution process in the presence of flow has been accelerated on a GPU with CUDA programming. The GPU implementation of the Sola-phase field model is introduced in this paper. The acceleration results of the dendritic growth simulation under flow by using a single NVIDIA GeForece GTX780 GPU with different memories are also evaluated. The results show that the GPU computation with the shared memory achieves the best acceleration effect, which is 56.16 times faster than that on a single CPU core for 2048 × 2048 grid size. In addition, the simulation results on GPU tally well with that on CPU, which indicates the reliability of GPU-accelerated phase-field simulation.