p359-parker.ppt

Download Report

Transcript p359-parker.ppt

Interactive Ray Tracing
Steven Parker
William Martin
Peter-Pike J. Sloan
Peter Shirley
Brian Smits
Charles Hansen
University of Utah
I3D 1999
presented by David Ventura
Leveraging Ray Tracing
• RT is inherently parallel
• Frame rendering time is sub-linear for static
scenes
• RT allows many interesting visual affects
The “Basic” Idea
• Synchronous mode
– Distribute rays to each processor
– Update frame when all sections are done
The “Basic” Idea
• Asynchronous mode
– Distribute pixel sets to each processor
– Constantly update screen
Tradeoffs and Features
• Lighting and material params must be set
• Model materials into different categories
– Diffuse, dielectric, metal, polished
• Indirect lighting sucks…
– Apply a tonal model!
Tradeoffs and Features
• Soft shadows are nice, but we need a low
computational substitution…
– Create penumbra falloffs by density occluders
Tradeoffs and Features
• Soft shadows and splines provide for good
representations
The Results
• 8 processors for interactivity,
but 64 for real-time
• Linear speedup to 64 processors
• Dynamic objects won’t happen
• Modeling and visualisation are
good apps
Conclusions
• Good for high-end machines and static
scenes
• Still a lot of questions…
–
–
–
–
Antialiasing?
Dynamic environments?
API?
Just how important is accuracy with interactive
simulations?