GF3 AA modes comparison - what's happening here ?

[pocketmoon_]

An original GF3, latest drivers, and aTuner to unlock all AA modes.

http://homepage.ntlworld.com/pocketmoon/comp.jpg

From left to right

No AA 4xAA 8xAA


Notice the microsoft logo? The 8xAA is much sharper than the other modes. Checking all modes available, the 8xAA, 6xS and 4xSupersampling all show much sharper detail.

Any idea for a AA novice?

 

[OpenGL guy]

Notice the microsoft logo? The 8xAA is much sharper than the other modes. Checking all modes available, the 8xAA, 6xS and 4xSupersampling all show much sharper detail.

Any idea for a AA novice?

Supersampling renders everything at a higher resolution, so higher resolution mipmaps will be used. This gives better texture filtering results.

 

[Bambers]

Unless you're using an 8500 in which case the mipmaps get a bit screwy in all except the nice non ordered 2x mode which doesn't show up very much in any 6xxx/9xxx drivers

 

[Lincoln Bauman]

This is something that has so far been completely ignored about Nvidia's higher level AA settings.
Supersample AA has several benefits associated with it that can have a noticeable impact on image quality.
Personally I'm not entirely disappointed with the AA modes offered by the GFFX, although I am annoyed
that the straight multisample modes appear to be unchanged from previous methods.
It's a shame that Nvidia did not expend more effort on the multisample modes and use a rotated grid for them.
They would have then had the total AA package, with fast good looking MS modes, and also SS modes for those finicky games that need it.

Lincoln

 

[antlers]

This is something that has so far been completely ignored about Nvidia's higher level AA settings.
Supersample AA has several benefits associated with it that can have a noticeable impact on image quality.
Personally I'm not entirely disappointed with the AA modes offered by the GFFX, although I am annoyed
that the straight multisample modes appear to be unchanged from previous methods.
It's a shame that Nvidia did not expend more effort on the multisample modes and use a rotated grid for them.
They would have then had the total AA package, with fast good looking MS modes, and also SS modes for those finicky games that need it.

Lincoln

I suspect there are fundamental incompatibles between their method for framebuffer or Z-compression and any sort of multisample pattern that is not OG.

 

[Crusher]

There you have visual proof, NVIDIA will decapitate you! And if you don't use FSAA, not only will you be decapitated, you'll become the amazing Flying Torso!

 

[Tahir2]

There you have visual proof, NVIDIA will decapitate you! And if you don't use FSAA, not only will you be decapitated, you'll become the amazing Flying Torso!

Only if you wear a Microsoft T-Shirt.. anyone who does deserves a healthy decapitation anyways - - - LOL

 

[pocketmoon_]

This is something that has so far been completely ignored about Nvidia's higher level AA settings.
Supersample AA has several benefits associated with it that can have a noticeable impact on image quality.Lincoln

Well on my GF3 4xSupersampling runs at 43% the no-aa fps and looks very nice in the apps I've tried in on. Of course I'm limited on the resolutions I can run at!

 

[Althornin]

Well on my GF3 4xSupersampling runs at 43% the no-aa fps and looks very nice in the apps I've tried in on. Of course I'm limited on the resolutions I can run at!

Is that 4xSSAA, or is that 4xS?
The 4xS is a skewed grid MS/SS AA combo mode.

 

[OpenGL guy]

This is something that has so far been completely ignored about Nvidia's higher level AA settings.
Supersample AA has several benefits associated with it that can have a noticeable impact on image quality.

The extra filtering of textures isn't always desired. Both SSAA and MSAA have drawbacks. SSAA tends to overfilter and has a large impact on performance. MSAA does nothing for texture aliasing but is much faster.

 

[KimB]

I suspect there are fundamental incompatibles between their method for framebuffer or Z-compression and any sort of multisample pattern that is not OG.

Z compression? Perhaps, but framebuffer data does not depend on position, so framebuffer compression would not be affected by MSAA sample possitions.

And z-compression would probably be planar in nature, such that it would still work with arbitrary sample positions, but would take a few more transistors.

The way I see it, nVidia saw it as a tradeoff, and didn't feel that the added work required in getting non-ordered-grid MSAA to work properly wasn't worth the money involved in producing it. Hopefully they'll change that stance now, as they really are behind in the FSAA department (note that I ascribe that by looking at its FSAA performance in more advanced titles...the FX does provide the best AA for older titles that are not close to fillrate-limited, such as CounterStrike, where the xS modes are usable...but that really doesn't mean anything to me, though it may mean something to some).

 

[KimB]

The extra filtering of textures isn't always desired. Both SSAA and MSAA have drawbacks. SSAA tends to overfilter and has a large impact on performance. MSAA does nothing for texture aliasing but is much faster.

Yes, definitely. Personally I'm hoping that before too long we'll see more advanced texture filtering to eliminate the need for full-screen supersampling entirely. That is, artifacts in magnified textures could be dealt with through bicubic filtering, and aliasing in alpha test textures (or other similar tests executed in the pixel shader that can cause aliasing that MSAA can't account for) could be dealt with either through smooth step functions, or texture-specific supersampling.