1 INTRODUCTION

Interactive rendering systems provide a powerful way to convey information, especially for complex environments. Until recently the only interactive rendering algorithms were hardware-accelerated polygonal renderers. This approach has limitations due to both the algorithms used and the tight coupling to the hardware. Software-only implementations are more easily modified and extended which enables experimentation with various rendering and interaction options.

This paper describes our explorations of an interactive ray tracing system designed for current multiprocessor machines. This system was initially developed to examine ray tracing's performance on a modern architecture. We were surprised at just how responsive the resulting system turned out to be. Although the system takes careful advantage of system resources, it is essentially a brute force implementation (Figure 1). We intentionally take the simple path wherever feasible at each step believing that neither limiting assumptions nor complex algorithms are needed for performance.

The ray tracing system is interactive in part because it runs on a high-end machine (SGI Origin 2000) with fast frame buffer, CPU set, and interconnect. The key advantages of ray tracing are:

• ray tracing scales well on tens to hundreds of processors;
• ray tracing's frame rendering time is sub-linear in the number of primitives for static scenes;
• ray tracing allows a wide range of primitives and user programmable shading effects.

The first item allows our implementation to be interactive, the second allows this interactivity to extend to relatively large (e.g. gigabyte) scenes, and the third allows the familiar ray traced look with shadows and specular reflection (Figure 2).

  

Figure 1: The ray tracing system discussed in this paper explicitly traces all rays on a pool of processors for a viewpoint interactively selected by the viewer.

  

Figure 2: A portion of a 600 by 400 pixel image from our system running at approximately fifteen frames per second.

In the paper we stress the issues in ray tracing that change when we move from the static to the interactive case. These include achieving performance in synchronous or asynchronous (frameless) fashions (Section 2), and modifications to traditional Whitted-style lighting/shadowing model to improve appearance and performance (Section 3). We also discuss a few areas that might benefit from interactive ray tracing and show some of the environments we used in Section 4. We compare our work to the other work in parallel ray tracing in Section 5. We do not compare our work to the many object space methods available for simulating shadows and non-diffuse effects (e.g. Ofek and Rappoport [22]) which we believe comprise a different family of techniques. Our interactive implementation of ray tracing isosurfaces in trilinear volumes is described elsewhere [23].