LOSTCIRCUITS

Z-Occlusion Culling

In all honesty, the companies that deserve the credit for z-occlusion culling are Imagination Technologies (PowerVR) and ATi with their hyper-z architecture. As mentioned above, in traditional graphics engines, every single pixel within a given frame has to be rendered. In more primitive 3D games, this is not really too much of an issue, however, with increasing complexity of the game environment, more levels are added to any frame. Well-known examples are Quake2 and Quake3 Arena and the cheat mode enabled by some ASUS drivers. This cheat mode could be used to render the foreground semi-transparent and, hence, lifting the veil off the details further in the back which, with a normal graphics card would be occluded. Occluded, however, does not mean that the pixels are not rendered, on the contrary. This phenomenon of carrying around a huge workload of invisible pixels is called overdraw and can amount to 70-80% of the entire workload the graphics engine has to deliver. If only 20-30 % of all pixels are really visible, the overdraw is 300 - 400 % (to use a different terminology).

The flexibility as consequence of the programmable nature of the vertex shader is probably best illustrated by this little cartoon (courtesy of nVidia). The vertex shader is capable of using different algorithms for subsequent sets of vertices which is in contrast to the "fixed" functions of conventional designs. .

In principle, there isn't much functional difference between the nFiniteFX vertex shader and the conventional T&L functions of any older GPU except for its programmability and extended capabilities. The main issue here is the programmability which provides superior flexibility for game programmers.

Next Page:    => special effects and pixel shader =>