Exploiting programmable graphics hardware to accelerate ray tracing

Abstract

The production of photorealistic images in computer graphics by modelling light interaction has always been the topic of much research. Ray tracing has been slowly moving towards the realm of interactivity due to increasing processing speeds, research on speeding up ray tracing and research on reorganising ray tracing to better exploit every resource available. Programmable graphics hardware improvements are increasing at a fantastic rate. The extent of their processing speed increases shatters Moore's law. Graphics hardware today possess programmable vertex and fragment shaders which enable the hardware to be used in ways different from the standard rasterisation pipeline. One such example is ray tracing. In this dissertation we partition the ray tracing task between the CPU and the programmable graphics hardware to achieve a significant speedup over implementations using only the CPU. Our system exposes very efficient use of the programmable stages present in existing commodity graphics hardware by offloading ray-triangle intersection tests from the CPU to the GPU. Our method performs virtually all the ray-triangle intersection tests on the GPU and uses the available memory bandwidth efficiently. We demonstrate quasi-interactive frame rates on complex scenes made up of several thousand triangles.