From today's perspective, the future of computation lies in parallelization. This is a central design focus of modern GPUs, whose enormous calculation capabilities are now available for general purpose programming. This thesis examines the portability of spatial queries to NVIDIA's popular GP-GPU CUDA platform. The two algorithms analyzed are the query on the R-tree spatial indexing structure and an intersection test between triangle meshes. Both of them offer potential for the high-level parallelization necessary for efficient GPU programs. The CUDA implementation showed no significant advantage over the single-threaded CPU one for the R-tree query, both for single as well as multiple parallel queries. The mesh intersection test on the other hand performed better using CUDA, at least when the tested objects were preselected by their bounding boxes using an R-tree. The entire CPU code of the examined program was written in C#. A pure C implementation could yield different results. The analysis could also be repeated fora multi-threaded CPU program and on newer hardware.     «

From today's perspective, the future of computation lies in parallelization. This is a central design focus of modern GPUs, whose enormous calculation capabilities are now available for general purpose programming. This thesis examines the portability of spatial queries to NVIDIA's popular GP-GPU CUDA platform. The two algorithms analyzed are the query on the R-tree spatial indexing structure and an intersection test between triangle meshes. Both of them offer potential for the high-level parall...     »