Rendering realistic, computer-generated scenes can be a time-consuming exercise even on the most powerful computers. But computer scientists at the University of California, San Diego have developed a technique that accelerates certain graphics computations by two or three orders of magnitude, at very little cost. "It's really simple and requires only a few extra lines of code," says Craig Donner, a fourth-year computer science Ph.D. candidate in UCSD's Jacobs School of Engineering. "It was just an idea that nobody had thought of or implemented." That idea applies adaptive refinement to graphics chips, and was presented in August at SIGGRAPH, the leading academic conference for computer graphics. In a technical sketch (pictured) and companion article, Donner and Computer Science and Engineering professor Henrik Wann Jensen detailed the technique for rendering a scene by tiling the screen and disregarding those tiles that are inactive. The result: significant speedups, depending on tile size and the algorithm used.
"It was three or four times faster as an overall speedup," explains Jensen, who is affiliated with the California Institute for Telecommunications and Information Technology [Cal-(IT)2]. "This technique provides significant speedups by exploiting computational similarity, coherence and locality."
Adaptive refinement is not new, but the UCSD team is the first to apply it to graphics processing units (GPUs), and specifically to the fragment processor, one of two programmable areas on a GPU. The project grew out of Jensen and Donner's work with colleagues from Stanford University on photon mapping with graphics processors.
Photon mapping simulates the indirect illumination of a scene fairly efficiently, and typically has been done on powerful workstations. But driven by the demands of gamers, chip designers have dramatically improved the power and speed of GPUs. "They've gone from barely being able to run the simple 2D graphics of your desktop graphical interface, to being able to run these really complicated games that you see coming out now at full frame rates, 60 frames per second," says Donner. "It's pretty amazing."
Read more... Source : PsyOrg
