What do we know about gf fx fsaa?

[Fotis]

I've been reading on beyond3d forums that gf fx will be using the same ordered grid fsaa as gf4 and that radeon 9700 rotated grid is better.Also people are saying that 6x rotated grid fsaa is better than 8x fssa.
What do we really know about gf fx fsaa?Is all the above true?

[Lezmaka]

Some people say the 9700's AA looks much better than the GF4's. Some are now saying that it's because of the sampling patterns, but I remember back when it came out, that the big thing with the 9700 was the gamma correction.

So which makes the most difference? The pattern, or does the gamma correction in AA make a hug difference?

[StealthHawk]

Originally posted by Lezmaka
Some people say the 9700's AA looks much better than the GF4's. Some are now saying that it's because of the sampling patterns, but I remember back when it came out, that the big thing with the 9700 was the gamma correction.

So which makes the most difference? The pattern, or does the gamma correction in AA make a hug difference?

the sampling pattern is what should eliminate the jaggies, the gamma correction only affects colors and should prevent blurring.

[Uttar]

There are TWO things which increase the R300 FSAA quality compared to the NV25:

1. The R300 uses RGSS
2. The NV25 is not able to do antialiasing on transparent/translucide triangles.

AFAIK, Gamma correction only affects overall quality but is more visible when using FSAA.


Well, we aren't even 99% sure the NV30 AA is OGSS. While multiple sites claim that, the extremetech article which got comments from David Kirk pretty much says:
"We aren't authorized to say if it's RGSS or OGSS, but it'll look better than NV25's OGSS"

Conclusion? They probably use RGSS, but their algorithm isn't as good as ATI's one, and they don't want people to know ATI got a better technology than them. They want to make people think EVERYTHING in the GeForce FX is superior to the R300.
It just isn't perfectly correct.
Or maybe it just uses OGSS - But considering David Kirk's past interviews, if that was the case, I would have expected "Well, RGSS isn't all that great... And the performance is a little lower... And bla bla bla" .
Another explanation is that RGSS would significantly reduce the use of their Color Compression technology, but they refuse to say it because they don't want ATI to discover it too quickly.


I guess time will tell :) Any opinion?


Uttar

[Bigus Dickus]

I know that was just an oversight Uttar, but you probably meant RGMS and OGMS, not RGSS and OGSS (though the GF4/NV30 does use SS in it's 4Xs/6Xs/8X modes).

;)

Aside from that, your point #2: I don't think the R300 is any more capable of antialiasing transparent polygon edges than the GF4/NV30 is. In fact, until ATi implements mixed MS/SS like the GF4 and NV30 do in the 4Xs/6Xs/8X modes, they are actually behind in this regard.

Your "advantage #2" should have been gamma correct AA.

Oh... and the B3D interview states that 8X uses an ordered grid, while other modes use a skewed grid. Which makes you have to ask "why even give 8X then, if the quality is only comparable to the competitions 4X?"

[Uttar]

Originally posted by Bigus Dickus
I know that was just an oversight Uttar, but you probably meant RGMS and OGMS, not RGSS and OGSS (though the GF4/NV30 does use SS in it's 4Xs/6Xs/8X modes).

;)

Aside from that, your point #2: I don't think the R300 is any more capable of antialiasing transparent polygon edges than the GF4/NV30 is. In fact, until ATi implements mixed MS/SS like the GF4 and NV30 do in the 4Xs/6Xs/8X modes, they are actually behind in this regard.

Your "advantage #2" should have been gamma correct AA.

Oh... and the B3D interview states that 8X uses an ordered grid, while other modes use a skewed grid. Which makes you have to ask "why even give 8X then, if the quality is only comparable to the competitions 4X?"

Oh, yeah, thanks for correcting me on the RGMS/OGMS part :)

Well, I read the transparent poly bit at Anandtech some time ago. But then again, according to Anand, 16x4 = 48... Here's the direct link and quote:
http://www.anandtech.com/showdoc.html?i=1656&p=11


While we did not get much time to evaluate the image quality of ATI’s multisampling algorithms, ATI insists that it will be higher quality than NVIDIA’s implementation in situations where transparent textures are used. The best example of this is in the DM-Antalus level in UT2003; the gorgeous grass in the level is simply a high resolution texture that is alpha blended, so you can see what’s behind it. According to ATI, NVIDIA’s multisampling algorithm will merely ignore aliasing within these polygon edges due to their transparency whereas the R300 will not.


Just looked at the Beyond 3D Radeon 9700 review, and they also state that.
Would be interesting to actually proof this one way or another...


Also, as I said in my original post, it is my understanding that Gamma Correction is active even if FSAA isn't active. But this certainly is the point I'm the least certain of. Radeon 9700 reviews doesn't seem to give many details on that.

However, nVidia's Intellisample tech paper ( http://www.nvidia.com/docs/lo/2415/SUPP/TB-00651-001_v01_Intellisample_110402.pdf ) seems to indicate this is active in ALL cases on the GeForce FX. My guess is that ATI just didn't give many info. I'd be surprised if the R300 doesn't support Gamma Correction in non-AA cases. But who knows...


Uttar

[Bigus Dickus]

I think there is a function in either OGL or DX (but not both IIRC) that can use pixel shaders to help in antialiasing alpha textures. However, ATi still hasn't implemented this because it would cause "odd" behavior in some legacy games. So, until they actuall enable the feature, their multisampling is limited in the same way that nV's is.

[StealthHawk]

Originally posted by Bigus Dickus
Oh... and the B3D interview states that 8X uses an ordered grid, while other modes use a skewed grid. Which makes you have to ask "why even give 8X then, if the quality is only comparable to the competitions 4X?"

wouldn't the simple answer be devious in nature? they can say their 4x FSAA has a lower performance hit than the competitions', even if it comes at the cost of quality.

that is, assuming NV30 still uses OGMS for 4x as compared to RGMS.

[Bigus Dickus]

The 6XS is a skewed grid, the 8X mode is an ordered grid.

From the GeForce FX launch interview at B3D. My guess is that 2X and 4X MSAA is rotated or "skewed" grid, and would be comparable in quality to ATi's 2X and 4X.

6Xs is probably a 1 x 3 ordered grid super sample of a 2X1 rotated grid Multisample. Could be something different though, so we'll have to wait and see. Quality is probably somewhere between ATi's 4X and 6X RGMS modes, but with the benefit of some texture antialasing.

Apparently 8X uses ordered grid for both the super and multisampling portions.

[StealthHawk]

Originally posted by Bigus Dickus
Apparently 8X uses ordered grid for both the super and multisampling portions.

8x = 8sX :confused:

if NV30 really uses RGMS for 4x, it would seem a waste of bandwidth for an OGMS 8x that would offer very little in IQ. so the 8sX thing makes some sense, but one would think the performance hit would be insane.

edit: ok, i read the interview and it says that 6x and 8x are a combination of MS and SS, so that answers that question.

[Bigus Dickus]

Yes, I would guess that the performance hit of 6X and 8X(s) will be substantial... especially 8X.

I think the only reason it is there is as a marketing tool: "we offer 8X AA, our competition only offers 6X."

Despite the fact that 6X RGMS will probably look better, and will likely have a smaller performance hit (but as for absolute performance, that's anyone's guess at the moment), consumers will buy it. That's marketing. ;)