Ray Tracing on a GPU with CUDA – Comparative Study of Three Algorithms

Abstract

We present a comparative study of ray tracing algorithms implemented on a GPU for three published papers using different spatial data structures evaluated for performance on nine static scenes in walk-through animation. We compare the performance for uniform grids, bounding volume hierarchies (BVHs), and kd-trees evaluated on a GPU for ray casting and Whitted-style ray tracing. We show that performance of ray tracing with BVHs exceeds the performance of ray tracing with kd-trees for coherent rays. Contrary, the ray tracing with kd-trees is faster than that with BVHs for incoherent rays. The performance of ray tracing with uniform grids is slower than both ray tracing with BVHs and kd-trees except for uniformly populated scenes. We show that the performance is highly sensitive to details of implementation on kd-trees.