The optimization and performance evaluation of ray tracer on GPU

Abstract

Ray tracing is an algorithm used to do image synthesis or rendering, which is well-known for its capability of producing a very high degree of visual realism. However, it demands a higher computational cost. It's internal characteristic made it suitable to be implemented in a parallel way. CUDA is a parallel computing platform and programming model invented by NVIDIA, which could be used to harness the parallel computing power of NVIDIA GPU. Therefore, we wish to combine the two, to improve ray tracing performance by using CUDA programming model. Moreover, we want to have a better understanding of CUDA applications' optimization. ☐ This thesis is focused on exploring how to use CUDA achieve good performance on ray tracing. First, we introduced the development of ray tracing algorithm. Then a metric which plays an important role in optimizing CUDA applications was discussed. Finally, we implemented three ray tracers, and by comparing their performance based on the metric we have discussed to identify the challenges of implementing a high- performance ray tracer with CUDA.