Parallel Considerations for Real-time Volume Rendering

Current technology dictates that parallel processing is required to perform real-time volume rendering for any signi cant data set sizes and image qualities. This paper compares parallel architectures and algorithms which perform volume rendering, as well as a few alternative architectures not previously applied to the volume rendering problem. Since the memory to processor bandwidth is the major obstacle to overcome when attempting to achieve real-time frame rates, the communication patterns will be the main feature discussed. Also of interest to the performance of the architecture/algorithm is the degree of parallelism. The goal of the current research in the Cube project is 30 Hz volume rendering, therefore the architectures/algorithms are ranked according to raw performance. To create an even eld for comparison, the performance is normalized to the number of processing elements.