The typical way to render volumes using graphics hardware is 3D texture slicing. This is easy to do and well supported by the current systems, but because it has to write every pixel many times, typically hundreds, it is hard to get to realtime performance due to the enormous bandwidth requirements. Modern graphics hardware opens up a way to speed things up.
Modern graphics hardware is extremely efficient nad highly programmable. The lastest graphics chips have processing power that rivals supercomputers (the ATI Radeon X 1900 XTX has a peak performance of >400 GFLOPS). Their memory bandwidth is comparatively lower, which makes it a promising apporach to due as much work as possible with as little memory consumption as possible. Because they have become programmable recently, there are a lot of possibilities. One of them is porting a classic software-based volume rendering algorithm directly to the GPU, volume ray casting. This has been demonstrated by the visualization group at VRVis in the context of Virtual Endoscopy. Our goal is to generalize the techniques presented and integrate them into the OpenSG volume rendering framework.