A Comparison of Parallelization and Performance Optimizations for Two Ray-tracing Applications

Ray-tracing is a compute-intensive technique to render three-dimensional graphics and is embarrassingly parallel. There are many examples of ray-tracing applications from both academic and public domain projects. We ask if the parallelization and optimization issues are similar. This paper uses two sequential raytracing applications (POV-Ray (Persistence of Vision Raytracer) and PBRT (Physically-based Ray Tracer)) to compare the issues. We describe every step of the process, from parallelization, debugging/ verification to optimization of the parallelized codes. We use a performance model of the applications to guide us on the speedups’ predicted upper bounds. We also profile the execution characteristics of both applications in their final optimized version. The general experience is that POV-Ray needed more efforts to parallelize and debug than PBRT, but less efforts to optimize.