high performance computing on graphics processing units: hgpu.org

Computer graphics renderers for creating photo-realistic images use mainly unidirectional path tracing, having good results for scenes without caustics or hard cases. There are also few renderers with bi-directional path tracing implementation, however due to the complexity of the algorithm implementation, they almost exclusively target sequential CPUs. The thesis proposes a way of implementation of bi-directional path tracer on a parallel many-core architectures such as the GPU and provides a working implementation. Further this implementation is compared to a parallel implementation of the standard, unidirectional path tracer, in terms of quality and speed. Interactive frame rates have been achieved for parallel bi-directional path tracing, proving the possibility of using such algorithm on GPUs.