Conclusion

Well based on the first series of test images, we can conclude that it is indeed possible to reduce the visibility of the dithering patterns, by filtering the dithered frame-buffer data before sending it to the RAMDAC. This technique results in an increased color resolution, but it is impossible to say by how much this resolution is increased. 3dfx claims that with the filter they use, the color depth is increased from 16-bits to 22-bits. Right now, I believe that this is probably a bit optimistic. 22-bit color is very, very close to 24-bit color and AFAIK tell the difference between the 24-bit test images and the so-called 22-bit test images is visible. I'm not going to risk making a guess how much the color depth really is, however. I personally believe that it is increased, but I don't think that that is the main goal. The main advantage of the technique is the reduced visibility of the ugly dithering patterns.

Everything you do has a cost, and in this case, the price you pay is a reduction of the spatial resolution. Small details are being smeared out by the filter and this can result in horizontal artifacts. In general, everything ends up being a bit more fuzzy and blurred.

So what can we say:

Better than normal 16-bit dithered? Yes!
Equivalent to 22-bit? Probably not.
Reduces visibility of dithering patterns? Yes.
Has no impact on small details? Definitely not!
An acceptable replacement for true 24-bit color? Nope... sorry 3dfx.

Hmm... now if 3dfx would send me a Voodoo3 board then I would be able to check out how good it really is... no pressure =^)
Nevertheless, feel free to send all your comments, reactions and hardware offers (stop pushing -ed) to the email below ...