Ray Tracing using 3D Grid Simulations

Bounding Volume Hierarchies (BVHs) and k-d trees have been used to create interactive ray tracing. Ray tracing dynamic scenes using nVidia’s R © OptiX TM has already provided thirty to sixty frames per second or better. In this paper, we implement space partitioning methods, such as grid method and the proximity clouds (PCs), on multiple GPUs. Our motivation is to investigate the use of such methods for medical applications, because there is direct one to one correspondence between 3D voxels used in space partitioning methods and 3D voxels in the volume data. In the past, proximity Clouds have worked well on static scenes, but object movement forces recalculation of the scene and some preprocessing cost. This paper investigates parallelizing these techniques on the GPU to determine the feasibility of dynamic scenes using them. Our scenes are made of spheres instead of volume data because at this time we do not know of any technique that can generate dynamically changing volume data. Interestingly, Proximity Clouds (PCs), which typically has large gains in rendering times compared to the 3D Grid method, emerges only slightly better than the 3D Grid method for dynamic scenes because both the processing and rendering costs are now added for dynamic scenes with spheres. The 3D grid method, due to its simpler preprocessing, may be the best choice for ray tracing the dynamic volume data on multiple GPU.