32-bit color

Why is 32-bit color such a big deal? There are two technical reasons:

A. Dithering
16 bits per pixel is only enough to represent about 65,000 colors. While that's a lot, it's not enough to show perfectly smooth color gradients. Imagine a sample image which goes from blue at the top to green at the bottom. With 16 bit color, you will see an obvious "stairstep" effect showing where each color stops and the next one begins. In order to lessen this effect, video cards use dithering to simulate more colors. They place small patterns of alternating color, typically in a checkerboard, to trick the eye into seeing an intermediate color.

Different cards have different approaches to dithering, and some are better at fooling the eye than others. One thing's certain, though: with 32 bits per pixel, you don't need to dither at all. You have so many colors-16.7 million, in fact-that the eye can no longer see the difference between one color and the next.

B. Multipass Rendering
Multipass rendering is becoming more and more important in 3D gaming. Put simply, this is blending more than one texture over the same area for added detail. For example, Quake uses two-pass rendering. The first pass gives you the standard wall texture, and the second pass adds a nice lighting texture for detailed shadows. Games like Unreal and Quake 3: Arena use even more passes to do cool things like volumetric fog and realistic reflections.

The main problem with 16-bit color and multipass rendering lies in the fact that the process is additive-each texture is blended with the previous one. There's a lack of precision in 16-bit color that causes artifacts when dithering . The problem is magnified as more colors are blended. Any inaccuracies in the color information are unnaturally amplified, leading to a result that is not quite equal to the sum of its parts.

Normally, the additional bandwidth required for 32-bit color is prohibitive. After all, 32 bits is double the color information for every pixel in every frame. For most 3D cards, including the TNT, you see a steep dropoff in framerate going from 16-bit to 32-bit color-on the order of 30 to 50 percent. However, The Rage Fury pulls off 32-bit color with a mere 20 percent performance penalty in Q2, DethKarz, and Turok 2. This is an excellent result.

However, there is a dark side to this otherwise good news. The Rage128 chipset has terrible 16-bit image quality. I noted this in my preview of the card, and the same exact problem is present in the shipping card with the latest drivers from the ATI web site. I have mixed feelings about such good 32-bit benchmark results knowing that there's no alternative: the 16-bit output is so poor that I consider it unusable.

ATI had a set of beta drivers that corrected the problem, but the resulting drop in performance was quite notable - so much that they included an option to "optimize for Direct3D," which simply disabled the fix and brought back the dithering problem.