CUDA-Based Real-Time Unbounded Ocean Rendering

Abstract

Height field and grid data are frequently updated during simulation of large scale ocean water, which reduces the frame rates of whole virtual scene. A CUDA-based framework for unbounded real-time ocean rendering is proposed in this paper. Firstly, height field data is calculated on GPU using FFT method. Then the whole view frustum dependent dynamic grid is created by an adaptive algorithm with continuous LOD, and positions of vertices are updated in CUDA kernels. Finally, normal map is generated using height field data on GPU and blended with perlin noise to shade ocean surface. Data transferred between GPU memory and primary memory is constrained during rendering loop. Experiments show the proposed framework achieves several times of speedup in grid updating efficiency than CPU implementation, and gets real-time and realistic ocean rendering results as well.