6 CONCLUSION

 Interactive ray tracing is a viable approach with high end parallel machines. As parallel architectures become more efficient and cheaper this approach could have much more widespread application. Ray tracing presents a new set of display options and tradeoffs for interactive display, such as soft shadows, frameless rendering, more sophisticated lighting, and different shading models. The software implementation allows us to easily explore these options and to evaluate their impact for an interactive display.

We believe the following possibilities are worth investigating:

• How should antialiasing be handled?
• How do we handle complex dynamic environments?
• How do we ensure predictable performance for simulation applications?
• What should the API be for an interactive ray tracer?
• How could an inexpensive architecture be built to do interactive ray tracing?

The first three items above highlight significant limitations of our current system: antialiasing is brute-force and thus too costly, and performance can be slow or unpredictable because there is a complex interaction between efficiency stucture build time, traversal time, and view-dependent performance. How much of these are due to the batch nature of traditional ray tracing methodology versus intrinsic limitations is not yet clear.

Additionally, we feel that an interactive ray tracer can help answer more general questions in interactive rendering, such as:

• How important are soft shadows and indirect illumination to scene comprehension and how accurate do they need to be?
• Are more physically accurate BRDF's more or less important in an interactive setting?
• Do accurate reflections give significant information about surface curvature/smoothness?

The ability to have more complete control over these features allows us to investigate their effects more completely.