You should also be able to have nVidia's QuickTweak icon in your tray area which allows very quick adjustments of custom D3D/OpenGL as well as forcing AA settings. Very few games will support Aniso and the many forms of AA out there, so manually overriding them can be very worthwhile. Also be sure to check out 3rd party tools like RivaTuner, Powerstrip etc. Here's some info on both AA and Aniso.
There are a lot of buzz words surrounding
Anisotropic Filtering but hopefully this will help to clear it all up a bit.
At a very basic level it enhances in game details and textures and helps to overcome the blurring effect of AA, although AA is not required for its implimentation it is rec to notice much diff for most games.
To get a bit deeper I need to get a little more technical. Texture mapping involves wrapping a texture's 2D sample around a 3D object to give the look of a brick wall, fur, wood etc, hence a texture map is used to apply textures to objects. Texture filtering dictates how texels (the elements of a texture) are used to produce the final pixels which make up the image. Through these processes textures are scaled up or down depending on distance and size. Often a series of pre-scaled textures are used to minimise distortion and maximise both speed and quality. Mipmap refers to the sequence of these scaled textures. 2 main types of texture filtering are BiLinear and TriLinear which are best for objects that are at right angles to the viewpoint. Aniso takes things a bit further by coping much better with the multitude of actual angles that a game may use. These 3 texture filtering methods obviously all come at a perf cost. BiL takes 4 texture reads per pixel, TriL 8 which effectively uses 2 lots of BiL which mostly helps the detail of objects further away. Aniso copes with many more variables and can map more accurately to angled textures, the end result is increased quality of textures esp on mid and long range objects.
The original
AA technique used by both nVidia & ATI was
Super-Sampling AA. This worked by rendering more pixels than were req for the final image on screen. So for a 800x600 image, 2xAA(SS) renders 1600x600 and 4xAA(SS) renders 1600x1200, therefore there is 2 horizontal pixels for every final 1 in 2xAA (4 for every final pixel in 4xAA) and this is where the extra info is found to base the final colour of the final pixel which then 'leaves' the card. That way there isn't quite the hit of actually running in a higher res but you gain much of the quality that would come from a higher res. There is a big hit from using SS-AA, for Rad8500 about 50% for 2xAA and 80% 4xAA.
The newer technique (well it's more of an enhanced 3dfx technique) currently used by nVidia GF3 & GF4 (perhaps the Xabre, and the Rad9700 I would imagine) is
Mutli-Sampling AA. This uses mathematics and complex algorithms to guess the correct AA for each pixel and largely blur the image (which is 80% of what AA does anyway) and that's why details and textures aren't as sharp with AA but are less jagged. AA-MS is much faster than AA-SS but does result in blurrier results, although this can be overcome with Aniso or GF4 4xS AA. The GF4 does use an enhanced version of AA-MS IIRC by taking the samples at slightly different locations which is intended to give better results. Also as stated GF4 have 4xS-AA as an option which uses some SS-AA techniques mixed with MS-AA which largely negates the need for Aniso and keeps textures and details sharp. 4xS is for DirX only but since OpenGL has Aniso it does't really matter. AA-Quincunx from GF3 & GF4 takes a 2xAA hit (little more for GF3) but achieves near 4xAA results, just more blurry (so is great with Aniso). It's to do with rendering 2xAA but shifting the pixels up a bit in order to give 4 pixel calculations. QxAA with Aniso is faster and better looking than 4xAA and is the best solution IMHO. There is little hit with MS-AA, GF4 take 20% hit for 2xAA/QxAA and 50% for 4xAA, GF3 is about the same but QxAA takes about 30% hit.
The Matrox Parhelia-512 uses
Fragment Super-Sampling AA which works out the key jagged lines and only performs AA-SS on these parts of the image, other AA modes blur (oh I mean AA) the whole image regardless of whether all pixels need this or not. This keeps the quality of AA-SS but removes most of the hits of unneccessarily blurring every single pixel. The downside is that some frames will req much more AA than others meaning the frame rate will be much more erratic, some jaggies are missed and it can't always be used. You will most likely need an average frame rate of around 100 FPS just to ensure the rate doesn't dip below 30 FPS, perhaps a slight exageration but you get the idea. If FxAA can't be used then the P512 has to resort to 4xAA (SS) which takes a big perf hit. All other hit results are from Q3A but using 3Dmark2001 as a guide the parhelia takes 25% hit for FxAA and a 70% hit for 4xAA. I would expect Q3A hits to be around 15% FxAA and 60% 4xAA though.
As with many things, AA & Aniso are mostly dependent upon a particular game's engine and implementation, in terms of both perf hit and quality. Beauty is certainly in the eye of the beholder and it is simply down to trying it and seeing for yourself. Obviously AA is at its most effective in lower resolutions, 800x600 for example. In 1024x768 15", 1280x960 17" and 1600x1200 19" the pixels are very small and the jagged effect of aliasing is also very small. If your monitor can't handle higher res or your CPU limits your gfx card's potential then it is a very good idea to enable AA and Aniso in any case. In terms of Aniso hit, GF4 takes 10% on 4tap (L2) and 25% on 16tap (L8) while Rad8500 takes 5% tap4 and 10% tap16, there are differences in final Aniso quality though. In short play around a bit in a few diff games and see what you prefer. HTH!
Facebook
Twitter