*This post will be updated as new information is discovered*
This post is meant merely to summarize my findings and make for easier reading.
First of all let me talk about some recent developments. The results of this thread are only valid for cards up to the GeForce4Ti series. The GeForceFX series behave differently according to how the driver is configured. Quality and speed are both drastically different between the 3 settings on a GeForceFX card! The naming scheme Application/Balanced/Aggressive has changed to Application/Quality/Performance in driver 43.45. It was changed to Quality/Balanced/Performance in driver 43.51. It has further been changed to Quality/Performance/High Performance in driver 44.03.
As always, things can change in future drivers. There is no way of knowing whether or not things will continue to change, or in what ways they will change. You can find Detonator drivers at the following places
Important points of interest for pre-GeForceFX cards(mainly GeForce3/4):
1) The "Balanced" setting in Detonator 40(ie, the 4x.xx drivers) gives the same raw results as previous drivers.
2) The "Application" setting seems to give lower benchmark results in 3Dmark2001, although no quality is gained/lost from "Balanced." "Application" does not degrade performance in the OpenGL tests I ran such as the Quake series. UT2003 Demo ran slower with "Application" when using D3D or OpenGL. Performance in 3Dmark2001 with "Application" is now equal to "Balanced" with driver 44.03.
3) The "Aggressive" setting is marginally faster than "Balanced" on the magnitude of 0-1%.
4) Enabling "Texture Sharpening" generally increases the level of Anisotopic Filtering by one level, up to the maximum of 8x.
There are some exceptions though. Using 2x FSAA with no AF and TS gives better performance than using 2x FSAA and 2x AF, although the quality is not as good. Also using 4x FSAA with 2x AF and TS gives better performance than using 4x FSAA and 4x AF, and quality is the same.
5) The alledged FSAA booster as described in this thread does nothing when using the system described above, an AthlonXP 1800+, GeForce4Ti4200, with 512MB RAM.
6) Application/Balanced/Aggressive does not force trilinear in raw benchmark situations or with AF enabled. If the software is set to use bilinear, bilinear will be used. If the software is set to use trilinear, trilinear will be used.
7) Application uses 'quality optimization' Rivatuner AF, and Balanced/Aggressive uses 'performance optimization' Rivatuner AF in OpenGL. I cannot discern any quality differences between the two. Performance of anisotropic filtering with Application and Balanced/Aggressive is equal in driver 44.03, leading me to believe that Application may be using 'performance optimization' now(RivaTuner doesn't support 44.03 yet, so it is impossible to check).
8) I find no reason not to recommend using "Application" mode at all times now, since the performance quirks seem to have been worked out.
Facts about GeForceFX cards/drivers:
1) Application has the same quality you may be familiar with from previous nvidia cards. Balanced forces a mix between trilinear and bilinear. Aggressive forces bilinear. This applies only to drivers previous to 43.51.
In 44.03 Application uses trilinear, and both Balanced and Aggressive use bilinear. 43.51 should also follow the same behavior, since the changes to AF described below first occurred in 43.51.
2) Different patterns of AF depending on what degree of AF is selected and what the driver is set to(Balanced or Aggressive).
The AF sampling pattern for Balanced AF was improved in driver 43.51, so that it is very close to Application's quality, without true trilinear being used, of course. You can compare the old AF methods
to the new AF methods
, which were first implemented for GeForceFX cards in driver 43.51.
3) A word of warning, you might not always be getting trilinear filtering even if Application is selected in the driver control panel, depending on the game and the driver revision. As far as has been determined, this pertains only to UT2003.