An optimal parallel algorithm for volume ray casting

Volume rendering by ray casting is a computationally expensive problem. For interactive volume visualization, rendering has to be done in real time (3D frames/sec). Since the typical 3D dataset size is at least 128³, the use of parallel processing is imperative. We present an O(log n) EREW algorithm for volume rendering using O(n³ ) processors which can be optimized to O(log³ n) time using O(n³/log³ n) processors. We have implemented our algorithm on MasPar MPl200. The implementation results show that a frame from 123³ data size is generated in about 3 seconds using 4096 processors