What does AF exactly do?

It filters out those jagged edged lines you see when you play without AF, and AA blurrs (for lake of a better word) images to make them look better.

At least this is how I rudimentarily understand them.

~Adam
 
Best analogy I can give you is, it pre calculates the filtering needed on the games textures, rendering it ahead of time. This in effect, gets rid of the cumbersome rolling carpet effect of Trilinear where you see the textures being filted as you move. With Anisotropic filtering, you move over the textures in an fps such as far cry, and the ground textures will already be filtered ahead of time. The higher the method of Anisotropic, the further down you'll be able to view in the game filtered correctly. To do some tests between Trilinear and AF, find a good fps, zoom with a sniper rifle, and watch the ground. AF/AA work in tandem btw. The far lengths AF filter still won't be smooth looking until AA is applied also.
 
Short answer AF is just looking at more texture data to more accurately choose what texture data should be displayed.

Slightly more complex answer, its basically for when texture data is underrepresented at sharp angles to the screen more texture lookups are done.

The lot more complex answer, each pixel on the screen which hits a textured polygon maps to an area on the texture. Once the polygon is so far away each pixel covers more area than 1 texel (a pixel on the texture) when this happens filtering needs to be done of some sort. Ideally the color would be based on the average color that falls under the pixel, unfortunately thats expensive to calculate if the area the pixel covers on the texture is large. So to make it more close to the pixel coming one to one with the texels on the texture they come up with MipMapping which is smaller textures made from the larger texture that have been prefiltered down so each texel on them is the average of the 4 texels the higher level there. Then you pick the mipmap that corresponds the closest to a one texel to one pixel ratio. This is called Bilinear filtering.

Now an issue with bilinear filtering is that if a polygon is tilted and part of the polygon is drawn with one mipmap and another part with another mipmap level you get a seam where the two meet (one part looks blurrier than the other basically). Trilinear filtering takes this a step further and does 2 texture lookups one on the closest mipmap level to being to the one texel to one pixel ration and then the second lookup is on the 2nd closest mipmap level. These two lookups are then interpolated together based on how close the 2nd closest mipmap is to being the closest.

Trilinear filtering works great and perfectly almost if the polygon facing the user is perpendicular to the user. If the polygon is at a sharp angle to the user(say the walls or floor in a game) then the pixels on the screen on those polygons will be covering a much larger area of texels than a polygon at the same distance away from the user but parallel to the screen*. Therefore the result is walls and floors look blurry or jittery. To fix this more samplings are done at the best mipmap level to get a higher quality looking textured polygon. This is called Anisotropic filtering. All that is happening is additional texture lookups are being done to add in just the color data of what the pixel is actually covering and not unnecessary texture data which is blurring the result.

Another way to get similar effects is similar to the idea of mipmaps except instead of reducing both dimensions at the same time only reduce one. This is quite expensive unfortunately in memory since you need all the combinations of mipmaping this way.

Now what do those numbers mean with #x AF well they just saying how many texture lookups are being done

1x AF - Bilinear Filtering (does one lookup on the mipmap level closest to 1 pixel/1 texel ratio)
2x AF - Trilinear Filtering (texture lookups to the two closest mipmap levels)
4x AF - Finally what most people actually call AF with 2 lookups on each of the corresponding mipmap levels (well they all are to some extent)
8x AF - 4 on each
16x AF - Hope you can figure this out ;)

*(btw the pixel will always map to a quadrilateral on the texture in the polygon parallel to the screen it will be a nice square aligning well with covering only a single texel of area on the closest mipmap on the texture, on the otherhand for a polygon at a sharp angle the quadrilateral covering the texture of what should determine the color will be really wide if say its a wall or really tall if its the floor unfortunately the mipmap will be undersampling as a result if the mipmap is choosed based on the quadrilaterals shortest dimension causing jittery transitions or a blurry texture because its choosing the longest length of the quadrilateral(which is more common and normally more pleasing to the eye)).

And if you want to know just a little more check out these slides http://www.cs.jhu.edu/%7Ecohen/RenderingTechniques/Lectures/ImageTextureFundamentals.color.pdf
 
They run around the same fps in Painkiller but, my i didn't see difference in graphics.
 
Originally posted by Shawn
They run around the same fps in Painkiller but, my i didn't see difference in graphics.
Click to expand...

Are you forcing it on the ati control panel.

or Maybe you already had it on 8x before.
 
Good way to figure out if its on or not. Find a game with really high quality textures. Find a long wall in the game. Then get the angle of the wall so you are basically just looking down it and its not very wide on your screen. Take a picture. Change AF settings. Come back in and take another picture. Then compare.

AF is definately a difference I would say though. Was trying to find some good pictures from various journal articles just the only issue is apparently these papers never had websites due to they were written long time ago.
 
You must log in or register to reply here.
Back
Top