LOSTCIRCUITS

Textures and Triangles

"Guns don't kill detectives, Love does". Phil Marlowe once said and translated into AGP terminology it means: "Textures can be compressed, vertices can't", at least not where precision counts, that is, in CAD applications. The importance of vertices and tessellation in the overall bandwidth equation is probably best hi-lighted by ATi's TrueForm approach where only a limited number of vertices is needed to interpolate additional surface facets. Keep in mind that this is possible in gaming situations but definitely not when it comes to CAD applications where precision of each point in the three-dimensional space is of utmost importance. One example would be the reconstruction of any protein to the level of modeling the electron cloud of each atom along with intermolecular van der Waals forces for drug design purposes or else particle flow through a vector field. Onother example is Pro Engineer where three-dimensional assemblies, e.g. a complete car are designed to the last nut before even the first bolt is put to the real world test.

Quadro 4 980 XGL glamourshot, courtesy of nVidia, 8-layer PCB, 128 MB unified memory architecture-local frame buffer (note that the PCB of the real version is slightly different).

The opposite scenario would be true for most games that may sport some high level of detail in the foreground or first person but, aside from splendid textures, lack the number of geometry detail typical for CAD applications. Even if there is a high amount of triangles, vertex shaders can be used to stipulate behavioral templates like the swaying of grass in CodeCreatures. Because of the template nature of the shaders, the motions will still appear somewhat conformistic as opposed to a situation where every single blade and its motions were drawn separately which is what we would need in CAD applications.

If it was not possible to document the advantage of AGP8X over AGP4X with a standard desktop card, that still does not mean that there is none. An analogous situation would be to try and show the advantage better tires on a car that isn't capable to use the standard equipment to begin with. In other words, if the card cannot process the geometry and deliver internally enough texels to put the necessary layers of skin on the polygons, then no bandwidth increase in the world will result in any noticeable performance improvement.

In other words, if we are interested in seeing any differences between AGP 4X and 8x, we need very complex geometry with a huge number of polygons and a relatively low demand on T&L or any kind of eye candy that is enabled through internal processing of shaders. Keep in mind that the main function of shaders is the reduction of bandwidth because everything can be processed internally without even writing data back to the frame buffer. To see what is going on at the AGP interface, we further need the most powerful graphics processor we can get and in this case nVidia jumped right in and supplied us with the Quadro 4 980 XGL. For those who don't know what a Quadro 4 980 XGL is, in one sentence, it is probably the most powerful professional graphics card under US$100, oops, typo, US$1000 but here are the specs:

Next Page:    => Quadro 4 XGL At One Glance =>