Abstract: The power and utility of volume rendering is increased by global illumination. We present a hardware architecture, GI-Cube, designed to accelerate volume rendering, empower volumetric global illumination, and enable a host of ray-based volumetric processing. The algorithm reorders ray processing based on a partitioning of the volume. A cache enables efficient processing of coherent rays within a hardware pipeline. We study the flexibility and performance of this new architecture using both high and low level simulations.