Implementing Classical Ray Tracing on GPU – a Case Study of GPU Programming

General purpose GPU programming has become a topic of intensive research recently. In this context, GPUs are typically used as general-purpose stream processors. Although is usually suites most of existing algorithms, it is far from ideal for ray tracing as it generates extensive memory bandwidth. We take an alternative approach and instead implement the entire algorithm in a single pass, as in classical ray tracing. This allows reducing memory bandwidth and increasing the performance of the algorithm. The results are demonstrated on several test scenes.