RGMS and FSAA modes

[RumbuDRak]

Hello everybody,
can someone confirm my "research" ?

ATI CFAA

4x CFAA = 2x RGSM + Narrow Tent
6x CFAA = 2x RGSM + Wide Tent or 4x RGSM + Narrow tent
8x CFAA = 4x RGSM + Wide Tent
12x CFAA = 8x RGSM + Narrow Tent and 4xRGSM + Edge Detect
16x CFAA = 4x RGSM + Wide Tent
24x CFAA = 8x RGSM + Edge Detect

Is it right that ATI uses RGSM as the "base" for the CFAA modes (or is it another)? I heard about 6x Sparse FSAA on ATI RV6xx, but I can't find any mention about it in local article about R600 on B3D ezine. What is it anyway?

NVIDIA

8x CSAA = 4x RGMS + 4xCSAA
16x CSAA = 4x RGMS + 12xCSAA
16xQ CSAA = 8x SSMS (sparse sampled) + 8xCSAA

What the hell is Sparse Sampled ? Is it MultiSampling mode as RGMS? In your local article the 8xQ CSAA is not described, is it 8x SSMS?

Is there any difference between RGSM and RGMS or is it just bad spelling on the net ?

Sry for confusing/bad questions and especially sorry for my level of english .

 

[ChrisRay]

8xQ is just 8 sample multisampling with a sparse pattern.

 

[RumbuDRak]

Great, thanks. So.....

1, There's no such 6x FSAA with Sparse Pattern on contemporary ATI GPUs.
2, RGSM/RGMS is just bad spelling and it should be called RGMS (Rotated Grid MultiSampling).

Am I right?

 

[Kaotik]

Great, thanks. So.....

1, There's no such 6x FSAA with Sparse Pattern on contemporary ATI GPUs.
2, RGSM/RGMS is just bad spelling and it should be called RGMS (Rotated Grid MultiSampling).

Am I right?

1. R300 (R9500/9700) to R580 (X1900) support 6x Sparse Pattern MSAA, in R6xx (HD-series) and newer, this has been removed in favor of 8x Sparse Pattern MSAA (someone correct me if I'm wrong, might be Rotated Grid too?)
2. Yes.

 

[Blazkowicz]

NVIDIA

8x CSAA = 4x RGMS + 4xCSAA
16x CSAA = 4x RGMS + 12xCSAA
16xQ CSAA = 8x SSMS (sparse sampled) + 8xCSAA

no, you're clearly wrong, it goes like this :

8x : 4x RGMS * 2x CSAA
8xQ : 8x RGMS
16x : 4x RGMS * 4x CSAA
16xQ : 8x RGMS * 2x CSAA

then you have the 8xS, 16xS, 32xS modes that can be useful for legacy games, they are a mix of RGMS and ordered grid supersampling.

 

[AlNom]

http://developer.nvidia.com/object/coverage-sampled-aa.html

8x = 4xMSAA + 8CS
8xQ = 8xMSAA + 8CS
16x = 4xMSAA + 16CS
16xQ = 8xMSAA + 16CS

However, going by B3D's own review of the G80, it seems that nVidia is counting each MSAA sample as a coverage sample in the coverage sample number.

So it's really,

8x = 4xMSAA + 4 "other" coverage sample points
8xQ = 8xMSAA = 8CS
16x = 4xMSAA + 12 "other" CS points
16xQ = 8xMSAA + 8 "other" CS points

edit: 8xQ and 16xQ are sparse sampled

 

[Colourless]

No cards do Rotated Grid Multisampling at levels other than 2x and 4x. At higher levels they do sparse sampling as it makes more sense. The only card that ever did Rotated Grid at a higher level was the Voodoo 5 6000.

 

[no-X]

DeltaChrome/GammaChrome used RG sampling 5x and 9x

 

[Blazkowicz]

well I was wrong afterall, didn't really think too much about it. with SSAA and MSAA combination, the number of X's does multiply.
(and, that v5 6000 can do any kind of sparse sampling if it wants to )

 

[Colourless]

Yes, it could do any sparse sampling it wanted to, but by default it was set to do rotated grid at 8x