A few weeks ago we mentioned that we were going to have a Pepsi challenge of sorts. We persuaded IT to lend us two of our standard client machines. We yanked out the video cards and replaced one with a Voodoo 5 5500 and the other with a GeForce 2. Except for the video cards, we had two identical systems. We then used chocolate and shiny trinkets to lure various employees into our FSAA challenge. Our results were useless within 8 hours due to new driver releases. A third test took place. This consisted of me.
Our criteria for choosing games was greatly changed in this new round of testing. We found it only fair to test both cards on games where we could generate the same images on both computers. This effectively limited us to the games that could either generate replays or had built in CG sequences that showed off the desired graphical effects. We came up with Need For Speed 5, Descent 3, and Homeworld. Not only are these great examples of where FSAA does wonders, but they also contain demos and replays. Thus maintaining the same angles and sequences on both cards at roughly the same time. We found it beneficial to stagger the replays so that we could see the same angles and scenes on both monitors. If we were to have them running at the same time we would never be able see the same angles represented on both monitors. Unless we could use ours eyes independently of each other and look at both monitors at once, in which case we would really be freaks.
Our monitors were both calibrated to produce nearly the same image. We did this while in the normal desktop area. While the colors produced by the cards were different, the only things being judged here were those issues that dealt with FSAA. Ignoring color and other image properties was not very hard. All factors, such as crawling, pixel popping, and jaggies, were readily visible. The demos chosen had very easy to spot regions where these artifacts could be noticed.
We started at the lower resolutions and worked our way up, eventually cycling around to do all the resolutions in 32-bit, where applicable. The resolutions used for this comparison were: 640x480, 1024x768, and 1600x1200.
For performance tests we ran Quake 3 in its' various modes. Descent 3 was among the programs that were going to be used. However, results obtained from this program were rather inconsistent. Consequently this test was dropped.