Abstract
This paper describes an architecture that enables the use of commodity off the shelf graphics hardware along with high speed network devices for distributed volume rendering. Several PCs drive a number of graphic accelerator boards using an OpenGL interface. The frame buffers of the cards are read back and blended together for final presentation on a single PC working as front end. We explain how the attainable frame rates are limited by the transfer speeds over the network as well as the overhead implied by having to blend several images together limits . An initial implementation using four graphic cards achieves frame rates similar to those of high performance visualization systems.