aa methods, v5, gf3, gf4, 8500, 9700

[Brent]

please forgive my ignorance, a 3d programer i am not, but learning I am


Someone needs to make like a list or something of all the methods each card uses for AA at different levels, heh

I recieved an email from my current 9700 pro iq article regarding 'texture shimmering' You know, how textures will seem to shimmer or move when you pan across a screen. I know that the Voodoo5 eliminated this, it was pure RGSS correct?

Why then do not the GF3, GF4, and R9700 not eliminate this? Is it because of their Multi-Sampling nature of only AA'ing the edge of polys and not touching the textures themselves? (well except for the 8500 maybe?)

What type of AA does the GF3,4 use at 2X and 4X, RGMS right?

The 8500 is definitely OGSS or actually what ATI calls PJMS?

And the 9700 is RGMS but actually PJMS according to ATI for that progrmable jitter table.

I want to make sure i get all of this down and understand it clearly.

Thanks a bunch guys, I seem to lean on B3D knowledge quite a bit as you all definitely know what you are talking about

 

[Hyp-X]

What type of AA does the GF3,4 use at 2X and 4X, RGMS right?

GF3,4 has:

2x RGMS
2x RGMS with a 5 tap filter (Quincunx)
4x OGMS
4x OGMS with a 9 tap filter (OpenGL only for some reason)
2x RGMS + 2x OGSS (aka. 4xS) (Direct3D only for some reason)

The 8500 is definitely OGSS or actually what ATI calls PJMS?

Yes it's OGSS. They used non whole grid sizes at some modes.
Eg. 1,5 x 1,5

And the 9700 is RGMS but actually PJMS according to ATI for that programable jitter table.

I don't know if it's really programmable, but it shouldn't really be, and likely it isn't.
Actually 2x and 4x are matching RGMS, but 6x is not any more.
Read the B3D review of the card

 

[Dave Baumann]

Yes it's OGSS. They used non whole grid sizes at some modes.

No - it was at least RGSS (if not prgrammable) in some modes; there were issues in some cases (fog) which resulted in it dropping back to OGSS.

I don't know if it's really programmable, but it shouldn't really be, and likely it isn't.
Actually 2x and 4x are matching RGMS, but 6x is not any more.
Read the B3D review of the card

I've been told that 9700 is doing what 8500 should have done in all cases (albeit in a multisampling sense). There are several different sample positions that can be predifined as a kinda sample position map; dependant on ciriteria for a particular pixel different sample maps for different pixels can be chosen and used. Because there is only a limited number of sample positions these will be repreated throughout a frame.

If you look at the B3D review the 4X is actually slightly off the RG map 3dfx used because the top two sample points are slightly squished in together (IIRC). There ws only one sample position shown from the application becuase there is probably insufficient variance in the rendered scene to trigger different sample patterns.