The difference in image quality between images rendered at 60Hz for games and images rendered off-line for movies has been rapidly shrinking over the past few years, to the point where claims of the arrival of interactive film-quality graphics are now not unreasonable. While these two types of rendering address problems with a number of different characteristics (many of them unique challenges for interactive rendering), it is not unusual for offline rendering for film to require a million times more processing time to generate a single image than an image rendered for a game. How could this be possible?
In this talk I will discuss how the development of programmable graphics hardware over the past few years, currently delivering hundreds of gigaflops and tens of gigabytes of memory bandwidth per second, has in turn sparked the development of completely new algorithms for real-time rendering that are fundamentally well-suited to the characteristics of this hardware. Thanks to the enormous performance benefits of developing algorithms that map well to graphics hardware, researchers and developers have invented new approaches that borrow few ideas from offline rendering while still delivering excellent image quality.
I'll survey these new approaches, contrasting how they address rendering tasks with corresponding approaches in offline rendering. I'll present some ideas about how these two types of rendering can draw from each other, speculate how future developments in hardware architectures may impact interactive rendering in the future, and argue that off-line rendering will soon be a historical artifact, made irrelevant by the rapidly increasing quality of interactive graphics.
About the speaker:
Matt Pharr recently cofounded Neoptica, a company devoted to developing software for advanced graphics on next-generation architectures.
Previously, he was a member of the technical staff in the Software Architecture group at NVIDIA, where he also served as the editor of the book "GPU Gems 2: Programming Techniques for High-Performance Graphics and General Purpose Computation", a collection covering the latest ideas in GPU programming written by experts in the industry. He was one of the founders of Exluna, a company that developed rendering software and tools; Exluna was acquired by NVIDIA in 2002. He previously worked in the Rendering R+D group at Pixar, working on the RenderMan rendering system.
Matt and Greg Humphreys are the authors of the textbook "Physically Based Rendering: From Theory To Implementation", which has been used in graduate-level computer graphics courses at over ten universities (including Stanford). He holds a B.S. from Yale University and a Ph.D. from Stanford University, where he researched theoretical and systems issues related to rendering.
664 Douglass St
San Francisco, CA 94114