Realtime ray tracing on current CPU architectures

Abstract

In computer graphics, ray tracing has become a powerful tool for generating realistically looking images. Even though ray tracing offers high flexibility, a logarithmic scalability in scene complexity, and is known to be efficiently parallelizable, its demand for compute power has in the past lead to its limitation to high-quality off-line rendering. This thesis focuses on the question of how realtime ray tracing can be realized on current processor architectures. To this end, it provides a detailed analysis of the weaknesses and strengths of current processor architectures, for the purpose of allowing for highly optimized implementation. The combination of processor-specific optimizations with algorithms that exploit the coherence of ray tracing, makes it possible to achieve realtime performance on a single CPU. Besides the optimization of the ray tracing algorithm itself, this thesis focuses on the efficient building of spatial index structures. By building these structures from scratch for every frame, interactive ray tracing of fully dynamic scenes becomes possible. Moreover, a parallelization framework for ray tracing is discussed that efficiently exploits the compute power of a cluster of commodity PCs. Finally, a global illumination algorithm is proposed that efficiently combines optimized ray tracing and the parallelization framework. The combination makes it possible to compute complete global illumination at interactive frame rates.