Your opinion about the not correct working FAA

[Radagast]

Hello.

Yesterday I asked the Matrox tech support in their forums about the FAA. I recognized - like some technical reviews before - that the FAA of the Parhelia does not AA'd the intersection of polygons. To see it in pictures (of UT2003) see the following threads:

http://forum.matrox.com/mgaforum/Forum10/HTML/004686.html

http://forum.matrox.com/mgaforum/Forum8/HTML/001491.html

There I asked where the reason for these not AA'd edges - and these are edges for the spectator - can be found and they (Richie and Haig) said that this caused by the way they detect the edges and that this can only be solved by changing the mechanism but if or when it will happen was not said.

So what do you think about this point? In my opinion a FAA without AA'ing every edge correctly is not better then the other AA'ing modes. So should they change or improve their mechanism of detecting?

[Greebe]

That is being looked into

[DGhost]

I would rather see Matrox AA edges inside polys caused by textures w/ "holes" in them.

i see what you are talking about, and i would like for them to be AA'd, but there is no easy way of doing it without killing performance.

[P5ycho]

providing the option to enable it at the cost of performance woould be the best choice, although i would prefer aliased textures with transparant parts more.

[McElvis]

Is the detection algorithm hardwired in the core , or is it software controlled?

[Radagast]

Good question.

I do not know. I hope it is part of the driver so that it could be changed. But I cannot tell you. Perhaps somebody else here can answer that question.

[Greebe]

HW, but some things can be patched via drivers

[mallowman]

Originally posted by DGhost
I would rather see Matrox AA edges inside polys caused by textures w/ "holes" in them.

Shouldn't that be the job of texture filtering? The alpha channel ought to be filtered just like color channels are, thereby smoothing out the edge. I'm just wondering, because I've seen that complaint a lot.

[Ali]

I like the big hints comming from Greebe. Nice big smilly faces on posts from a beta tester have to be a good sign.

The FAA would be the only reason for me to get a Parhelia at this stage, unless they drop the price considerably on the 64MB version.

Ali

[Pace]

I'm getting more undecided Looking at the Radeon9700, it offers easily more performance for the same price, and will last me till Parhelia II etc. However, the current Parhelia might last until then as well, and I get to enjoy the great DualHead and SurroundGaming on occasion too.

AA is a crucial factor, and with Parhelia down to £270 here ($380), I'd like to snap one up soon.

P.

[DGhost]

Originally posted by mallowman


Shouldn't that be the job of texture filtering? The alpha channel ought to be filtered just like color channels are, thereby smoothing out the edge. I'm just wondering, because I've seen that complaint a lot.

ah, last i checked if you have a hole in the texture (caused by alpha masking a portion of the texture to be totally transparent), and you are using texture filtering, it will not AA the edge with the rest of the scene, just with the rest of the texture.

this sort of thing would be easier to add in than poly intersections. all you would have to do is treat the area immediately adjacent to the alpha masked spot and treat it as poly edge.

[3dcgi]

This is one of the reasons Matrox doesn't claim FAA is FSAA. Others implementations antialias intersecting objects, but currently only supersampling AAs textures. Of course Parhelia can do this, but it is very slow.

If the ut2003 aliasing is really intersecting objects then it could have been avoided if Epic had tessellated their polygons to have a shared edge for the wall and opening. They made the tradeoff for lower polygon count.