Lund University
ACM SIGGRAPH 2006 Conference Abstracts and Applications
Abstract
To reduce bandwidth usage to textures in a graphics hardware architecture, it is common to use texture compression. Traditionally this is only done with static textures, and they are compressed offline. However there are applications where it would be desirable to be able to compress dynamically generated textures on the fly (cubemaps, hardware accelerated GUIs). We present a method for compressing dynamically generated textures using the programmability of modern GPUs. We show that, by choosing a compression scheme suited for streaming architectures and allowing a slight decrease in image quality, compression can be done up to 100 times faster using the GPU. We also show that compressing dynamically generated textures can significantly reduce texture bandwidth usage.
Paper
Master's thesis
by Oskar Alexandersson and Christoffer Gurell on the same topic [thesis.pdf]