In general, anti-aliasing greatly enhances the gaming experience by removing jagged edges, smoothing stair-steps on angled polygon edges, and most importantly in my opinion, eliminating the crawling that occurs when a moving polygon passes through the player’s field of view. One example of this crawling can be seen in the second animated logo when Deus Ex starts up.

Unlike the anti-aliasing solutions delivered by most other graphics card companies, 3dfx uses Rotated Grid Super Sampling in their AA. This can give better image quality than the alternative, Ordered Grid Super Sampling. OGSS uses a normal grid when sampling a scene, which means you get fewer color variations in the near horizontal and vertical areas. RGSS uses a rotated grid, and can therefore provide more color variations, decreasing the occurrence of jagged edges on those nearly horizontal or vertical polygons.

Another feature of the Voodoo5 that can be used today is its texture compression. Offering a slightly higher, fixed compression ratio, 3dfx’s FXT1 is the alternative to paying S3 and using their S3TC format. It can be used to compress textures using an 8:1 ratio under OpenGL, increasing load times somewhat. More information about FXT1 can be found in our article FXT1 Visited.

The Voodoo5 5500 PCI board has 64MB of ram and two of 3dfx’s standard issue VSA-100 chips, as seen on the AGP version as well as on the Voodoo4 and Voodoo5 6000 in the future. While the board actually has 64MB of ram, it can’t use that 64MB for textures as such. Each VSA-100 chip has access to 32MB, which it uses for storing textures. Textures are duplicated across the two memory sets, with only the frame buffer shared. In my opinion this is somewhat wasteful of memory, but I guess that’s the price you pay for the increased bandwidth. This increased bandwidth comes from the two chips working together in a Scan Band Interleave. As you’ll recall from the Voodoo2, when using two boards you could set them up so each one would render every other scan-line. Scan Band Interleave is the same idea, but the chips can now render a programmable set of scan-lines rather than just one. This is done to improve cache efficiency and performance over normal SLI.