The Tests 2
Even more perplexing than the OpenGL settings were the Direct3D settings. Direct3D settings had one thing going for them. You actually had access to a slider. If you look in the advanced section and proceed to the Direct3D FSAA area you will be greeted with a slider that has 9 distinct notches.
Then the mystery begins. What do these nine different settings do? We have heard tales of 2x being the middle and 4x being the end. With everything else being variations thereof. After a little conference call with NVIDIA we found out what these settings actually are.
2 Equivalent to 3dfx's 2x mode
3 Equivalent to 3dfx's 4x mode
4 4x with mipmaps set at the back buffer resolution
5 4x using gaussian blur
7 9x with mip maps set at the back buffer resolution
9 16x with mip maps set at the back buffer resolution
This explains why the game would start at a grinding halt when we had the slider at nine. Here we thought NVIDIA had really meant "beta" when they released the driver. One might say that 16x FSAA probably has some very detrimental effects on performance.
All these settings meant that we had a whole lot of combinations to compare. This is my job, but even I have to draw the line somewhere. I may not get to go home, but enough is enough, I will not give up my daily bucket of water for jaggies. If I had tested all the various combinations it would total to around… well I'm too lazy to bother figuring out all the combinations or permutations whatever they may be. Something had to be done about this. I figured that all the 9x and 16x settings were too slow to even bother with. I have to admit that I also wanted to avoid getting e-mail from that yokel that would complain about comparing 9x and 16x to the Voodoo's 2x and 4x, a somewhat valid point.
I had to decide which NVIDIA OpenGL FSAA settings to use in our tests. I started with value 0 on OpenGL and almost wretched. After seeing the text in Homeworld turn in to a fuzzy pile of mush I decided I would do NVIDIA a favor. I tossed out that test immediately. Setting 1 had a bit more promise but I decided to set it to the max and do the comparison from there. I narrowed OpenGL settings to the highest setting and gave myself something to compare.
I narrowed down the Direct3D choices to 4 settings after I threw out the 9x and 16x, but that still left too many combinations. I decided to take the 2x mode (setting 2) and the number 4 setting, which is 4x (MIPMaps at the back buffer resolution). These modes effectively take the best that NVIDIA has to offer in this set of drivers to pit against the 2x and 4x settings of the Voodoo5 5500. Apples to apples, yes?
We ran a new set of Quake3 benchmarks with the new drivers. We used the same test systems that we used in the previous test. Likewise, the comparison systems had not changed except for the drivers.