Abstract: Latency contributes to image error and to motion sickness in head-tracked graphics displays. We attack the problem of latency by exploiting parallel scene generation, using multiple graphics engines to render images corresponding to a cloud of viewpoints. The real-time position of a 3D tracker is used to select among several just-generated images, rather than to generate a new image, for each frame. The system's scalable architecture is comprised of off-the-shelf components. When. future PC-class machines offer inexpensive high-speed 3D graphics, the design of the system will become economically attractive.