Bandwidth-efficient Hardware-Based Volume Rendering for Large Unstructured Meshes

Abstract

Recent advances in graphics processor architecture and capabilities have made the development of fast and efficient unstructured volume rendering methods possible. These techniques can be classified into two roughly delimited categories: cell projection based methods and GPU raycasting algorithms. However, both approaches are subject to limitations, respectively due to the main memory-to-GPU bandwidth for the former and due to the GPU per-fragment computation speed and memory size for the latter. These potential bottlenecks can be particularly limiting for large-size datasets, such as the ones produced by large-scale numerical simulation. In this work, we describe an enhancement to the cell-projection rendering method, allowing us to specify each tetrahedron with only 4 vertices and their associated data. By using a point sprite primitive, instead of a set of 4 triangles, we significantly reduce the amount of data transferred from the main memory through the graphics port for each frame rendered. We evaluate the impact of the different rendering stages of our method on the overall frame rate.