If this is your first visit, be sure to
check out the FAQ by clicking the
link above. You may have to register
before you can post: click the register link above to proceed. To start viewing messages,
select the forum that you want to visit from the selection below.
I'm sure someone here knows the real reason why, but from what I've read I'll make a guess.
It seems that the FAA is operating on the final rendered image, at a subpixel level (otherwise it would possibly be looking at EVERY triangle in the picture, including those that are not in view because they are covered by other triangles in the scene). Further, I don't think FAA works on the triangles anyway. That sounds more along the lines of edge FAA, while I think FAA is actually looking at pixels themselves, meaning it could theoretically AA pixels in the textures on a large triangle just as effectively as the edge of triangles in the scene. This means it works more along the basis of where visible edges in an image are, with no real regard to whether the edge is on or in the triangles of the polygons making up the image.
Anyhow, if that's correct, it seems that maybe the stencil buffer itself is actually being used as part of the FAA process, possibly for storage of the information needed when comparing the subpixel information?
Quick note: I'm no way shape or form any sort of hardware level techy (I don't, nor have I ever, design, or know anything about designing hardware.). So these are purely guesses based only on the slightly ignorant knowledge I do have combined with what I've seen so far.
"..so much for subtlety.."
System specs:
Gainward Ti4600
AMD Athlon XP2100+ (o.c. to 1845MHz)
Accually I am sure FAA is applied to the trangles. That is why they also support 3D AA lines, and that is why Tom made this comment.
"(FAA) does not recognize lines that are rendered in the form of textures."
Which I think is dumb. Since most textures won't need AA and applying AA to them will making them blurry. Blurry textures is the second biggest problem with FSAA.
I am not sure how the stencil buffer conflicts with FAA. I can't see how some pixel distortion would make me want to switch back to 4xFSAA. Unless it causes major problems.
The stencil buffer is used for shadows because it can make a cutout or outline of an object like person or a crate. It looks at the polygons and determines the edges of the object. this info is then used to project a shadow where it is suposed to be in respect to the light source.
FAA also looks for the edges of objects, because this is usualy where the jaggies come from. It then it determines how to blend the colors of the edge between the object and the background so that the edge is softened and the jaggies are reduced.
Right off the bat i can see why this could be a conflict. Maybe the stencil buffer is used to find the edges of the objects so they can be AA'd.
Also, this can be used with aniso i think, so textures will look good regardless. plus this AA method won't muddy textures like FSAA does.
That makes sense (the explanation of how it likely works, and why the stencil conflicts with FAA). But since at least a few of my current games use stencil shadows, it might cause some problems. Still, that's why I think a card with this sort of bandwidth available having 4xFSAA as a fallback is a good thing (and probably not taking too hard a hit for the slower method). The anisotropic filtering should still aleviate much of the effect that the FSAA will have on textures too (just like it does with GF4).
"..so much for subtlety.."
System specs:
Gainward Ti4600
AMD Athlon XP2100+ (o.c. to 1845MHz)
You're all wrong when you're thinking that FSAA generally blurries textures. First of all you have to say about which form of FSAA you're talking. Do you mean Multisampling with blurry filter (=Geforce3/4 Quincunx) or do you Multisampling without blurry filter (=Geforce3/4 2x/4x) or do you mean Supersampling (Ati, Kyro, old NVidia)?
Quincunx: relatively fast, blurry filter blurries edges (what is good), but also blurries textures (what is not good).
Multisampling 2x/4x: relatively fast, average edge smoothing, does nothing to textures.
Supersampling (any form): slow, average on edges, *sharpens* textures (like anisono does, but not that much as anisono)!!!
Matrox of course compared its new FAA mode to Quincunx to make its new FAA technique look as good as possible. But to be fair we should state that FAA is no better on textures than normal Multisampling, while Supersampling FSAA is even superiour texture-quality-wise to both FAA and Multisampling FSAA. However, supersampling is terribly slow. Also why sharpening textures by using slow Supersampling, when anisono does a much better job? So 16x FAA is definately a great thing, because it does what is most important, namely smoothing the edges without too much performance penalty.
This is a quite good article about Supersampling, Multisampling and Quincunx from a *very* reputable site. Use Babelfish if you don't understand german. Let me just translate one paragraph manually (page 11, last big paragraph):
"Let's get rid of one often appearing misunderstanding: <b>Multisampling does NOT blurry textures!</b> Not a bit. <b>Multisampling does just nothing with textures.</b> With Supersampling you can make textures sharper. But thanks to the anisotropic filter, which you can apply additionally to all AA-Methods, this additional sharping is not really necessary. <b>Texture blurrying comes only by Quincunx smoothing.</b> But especially if this smoothing is used combined with Multisampling, like the GeForce3 does, the blurryness gets very strong."
Comment