Future of MSAA?

I have owned GF3 & GF4. Quincunx was noticeably worse on GF3.

It didn't help that you were mostly restricted to 1024x768 or less with AA for GF3 max (if you wanted any sort of performance) and the blurring is far more noticeable on lower resolutions.

I've been trying out 4x9 Tap (didn't really bother trying it before now) and I think I've found a new favorite AA mode. I can't tell the difference performance-wise in farcry from straight 4x. Makes a real difference (for the better) in IQ in Mafia too.
 
radar1200gs said:
I have owned GF3 & GF4. Quincunx was noticeably worse on GF3.
I owned them too otherwise i wouldn't say what i've sad. I haven't noticed ANY difference in Quincunx quality between these two cards, both were blurry as hell. And always thought about all the talk of improved 2xQ in GF4Ti as of marketing BS from the quality point of view (it was faster than on GF3, that was true).

It didn't help that you were mostly restricted to 1024x768 or less with AA for GF3 max (if you wanted any sort of performance) and the blurring is far more noticeable on lower resolutions.
I was using 1024x768 with both of them. It was more a question of my monitor back then than videocard anyway.

I've been trying out 4x9 Tap (didn't really bother trying it before now) and I think I've found a new favorite AA mode. I can't tell the difference performance-wise in farcry from straight 4x. Makes a real difference (for the better) in IQ in Mafia too.
Yeah, everything becomes blurry :LOL: As for Mafia -- just finished it on 6800U with 8xS ;)

2xQ is effectively dead. 4x9 weren't ever released to the public. Anyway 4x on a 6800 is way better than 4x9 on previous cards.

[edit] mistakes
 
I owned a GF3 and GF4 and Quincunx blurred equally with both cards. In fact, I recall being amazed at all the NV fan sites claiming that Quincunx was now 'fixed' with the GF4 based on screen captures when my eyes were clearly telling me otherwise.
 
I'm not seeing any blurring with the 4x9 tap, if anything it sharpens the textures more which is probably why I like it.

With the quincunx issue I guess some can see it and some can't. I have had the same monitor now since I first purchased my GF2 Pro, so I know it isn't different monitors causing the difference.
 
[url=http://www.beyond3d.com/forum/viewtopic.php?p=402564#402564 said:
In relation to Gamma Corrected FSAA Radar[/url]]If you apply AA to samples that have been gamma corrected, you are not working with values the app originally sent anymore and the resulting final color isn't based on colors generated by the app.

Greg, are you actually aware of what Quincunx or 4x9 is doing? I find your championing of these modes and your comments about Gamma corrected FSAA to be absolutely at odds with each other. Are you actually serious about these things?
 
Yes I know what Quincux is doing. I was waiting for someone to ask this question, actually.

The difference is that you don't HAVE to enable Quincunx if you don't want to - you have a choice. A choice you don't get with ATi and gamma correct AA.

EDIT: I don't think I'm championing these modes, simply saying I think they are good and some critisisms levelled at them are not valid ones.

Personally I wouldn't agree with nVidia's assertion that Quincunx is equal to 4x AA - it isn't. However, it is better than 2x and cheaper than 4x. It's a "3x" AA mode if you like.
 
radar1200gs said:
Yes I know what Quincux is doing. I was waiting for someone to ask this question, actually.

The difference is that you don't HAVE to enable Quincunx if you don't want to - you have a choice. A choice you don't get with ATi and gamma correct AA.
Let's just beat that dead horse some more, shall we? When you can enable gamma correct AA on your card, maybe you'll have a point.
EDIT: I don't think I'm championing these modes, simply saying I think they are good and some critisisms levelled at them are not valid ones.
How can you be critical of other people's opinions when you don't understand yourself? Your comment in another thread about 4x9 improving texture sharpness gives you away.
Personally I wouldn't agree with nVidia's assertion that Quincunx is equal to 4x AA - it isn't.
Wow, that's reassuring. :rolleyes:
However, it is better than 2x and cheaper than 4x. It's a "3x" AA mode if you like.
3x AA would require 3 samples per pixel. You can't get 3x from 2x. 2x + a blur filter is still 2x + a blur filter. And not everyone prefers the extra blurring.
 
I owned them too otherwise i wouldn't say what i've sad. I haven't noticed ANY difference in Quincunx quality between these two cards, both were blurry as hell. And always thought about all the talk of improved 2xQ in GF4Ti as of marketing BS from the quality point of view (it was faster than on GF3, that was true).

Actually there is a difference between the way the Geforce 3 and Geforce 4 + line handle AA, Paticularly quincunx.


Anyway isnt quincunx a 2x AA with a 5tap filter? Or was it a 4tap filter ;p
 
I think you'll find the filter is 5 tap, and it's likely the difference between GF3 and what came after is in how the filter is actually implimented (No, I don't know exactly how the filter works).
 
I think Quincunx/4x9tap modes are fine. Whether people like them or not is gonna be subjective. They are compromise IQ improvement modes. Just like ATIS temporal AA. Each having their own side effect. Whether you like their compromises for performance is completely dependent upon user.
 
The thing that changed from GF3 to GF4 was the position of the texture sampling point. But the difference isn't that big, really.

One thing that can never happen with quincunx or 4x/9tap enabled is that a white pixel is neighboring a black one (or the same thing for each color channel). The blur filter might improve gouraud shaded scenes, but it's terrible for textured rendering IMO. Quincunx looks worse than 2x RGMS to me, and it's notably worse than methods like Flipquad.
 
Xmas said:
The thing that changed from GF3 to GF4 was the position of the texture sampling point. But the difference isn't that big, really.

One thing that can never happen with quincunx or 4x/9tap enabled is that a white pixel is neighboring a black one (or the same thing for each color channel). The blur filter might improve gouraud shaded scenes, but it's terrible for textured rendering IMO. Quincunx looks worse than 2x RGMS to me, and it's notably worse than methods like Flipquad.

Looks works than 2x RGMS at edge removal? or just overall quality? Quincunx actually did do a good job (for performance hit) at removing jagged edges for its time. I mean I have seen cases on horizontal edges it was actually superior to 4x OGMS for removing jagged edges.

But yes it does have a rather undesirable effect on textures.
 
The blur filter must be what affects textures (I believe that part is done in the ROP's).

Quincunx is MSAA based and therefore, like all other MSAA methods can't affect textures except for those samples that fall on object edges.
 
radar1200gs said:
The blur filter must be what affects textures (I believe that part is done in the ROP's).

Quincunx is MSAA based and therefore, like all other MSAA methods can't affect textures except for those samples that fall on object edges.

The 5tap filter is done on ramdac on Nv25 + hardware. 9tap is done internally.
 
Yes, I was meaning 2x Quincunx above.

I'm not entirely sure what 4x9 tap does (have to look it up), just know that I like its output.
 
Back
Top