Will GeForce FX have RGSS FSAA?

[AdamK47]

I remember reading a while back about speculation on the various FSAA modes the GeForce FX (then NV30) would have. Two of them were Supersampled FSAA modes. Will nVidia have the good old 3dfx FSAA implemented into the GeForce FX? Thats been on my mind for nearly two years since the aquisition of 3dfx. I'm hoping that nVidia will use one of the most well known pieces of technology developed by 3dfx and make it available again.

 

[Bambers]

Theres currently 2 other threads along this line we don't need another

 

[AdamK47]

I did see those. One was close with RGMS FSAA, not RGSS FSAA. The other talks about the MS FSAA modes. I want my Super Sampling!

 

[Joe DeFuria]

Short answer: not according to what nVidia has told us so far.

Longer answer: there is the possibility that nVidia hasn't told us everything about their AA capability. If that's the case (perhaps whether it works or not depends on the success of their silicon re-spin), nVidia might surprise us with something.

However, I would doubt if they implemented a SuperSampling technique as their AA approach...not because of quality but speed. It's a lot easier to sell speed....

 

[Joe DeFuria]

Also, note that ATI was originally suppossed to implement a jittered-grid supersampling version of SmoothVision 2.0 to compliment the currently available multisampling one.

http://www.ati.com/vortal/r300/educational/main.html

(Click on the smoothvision section for info).

Whether or not ATI follows through on this remains to be seen. They may wait to get DX9 drivers shipped and up and running and initially debugged before adding the supersampling feature if at all.

Or...they may wait for NV30 to be shipped....so that none of the reviewers "accidentally" enables supersampling AA and compares performance to NV30.

 

[KimB]

It's pretty much certain that nVidia will never support a full supersampling technique, though the 6xs and 8x modes are a combination of multisampling and supersampling (In either case, I don't think there is more than 2x supersampling being done).

 

[Sage]

if ATi introduces a jittered-grid sampling method, they had better be prepared to eat shit and die because of the performance it would take do do enough jittered samples to be effective (i think its something like 64+ ? they might get by with 16 looking acceptable) but i'd buy one the moment they do it! and im not exactly a fan of the R300...

 

[Bigus Dickus]

Why on Earth would you need 16+ (or 64+) jittered samples to look good? Shouldn't that give better IQ for a given number of samples than a straight ordered grid?

 

[megadrive0088]

I sincerely hope that by the time XBox 2, PS3 and GameCube 2 arrive
(using GPUs that are several generations beyond R300-NV30) the
anti-aliasing issue will have been completely resolved. I'm so tired of the disappointments with every new GPU as far as a complete AA solution.
There better be a teriffic AA method selected by then, and enough samples of that method to make AA a non-issue once and for all

 

[EasyRaider]

if ATi introduces a jittered-grid sampling method, they had better be prepared to eat shit and die because of the performance it would take do do enough jittered samples to be effective (i think its something like 64+ ? they might get by with 16 looking acceptable) but i'd buy one the moment they do it! and im not exactly a fan of the R300...

Uh, they already have a jittered grid. That is, the grid is not ordered. I think you meant stochastic (sp?) sampling.

 

[Bambers]

Well the SV 1 on the 8500 certainly never turned out as planned. It's apparent in the first SV driver version that it does actually appear to adjust the grid for 2xQ and 4xQ on a pixel by pixel basis (I would put a screenshot here but it appears I've deleted from my webspace) but as everyone probably knows by now the AA was almost completely reduced to OGSS in the 9xxx/6xxx drivers. From the ati 9700 presentation it appears that the supersampling it has is the same as the one in the first SV. Maybe they got it working properly this time, or maybe not...

 

[Simon F]

Why on Earth would you need 16+ (or 64+) jittered samples to look good? Shouldn't that give better IQ for a given number of samples than a straight ordered grid?

In the past I found that, with only a few samples per pixel, an image rendered with a jittered grid tended to look quite noisy and more disturbing than the errors in the non-jittered grid.

 

[Basic]

Problem here is (as it often is) differing definitions.

ATI has a jittered grid in the sense that it looks rather random when you look at the grid for one pixel. But it is not realy random. It's carefully designed to have the sample points at optimal positions for horiz and vert lines, and rather good for the rest.

The number of such patterns are limited, but there's still a bunch of them. Rumours says that ATI can switch between different such sampling patterns on a per pixels basis in a pseudo random way. (Practical tests shows that they don't seem to switch pattern all that much though, if at all.)

So ATI does jittered samples in the sense that give good quality with reasonable number of samples.

I agree that pure random sampling (poison disc distribution) need an unreasonable amount of samples before it's effective. Simon has showed some realy good examples here.


Regarding GeFX, I doubt we'll see RGSS FSAA on it ever. Except if someone does it manually with an accumulation buffer.
(Or maybe if someone tries the 2x or 5xRGSS tricks I propsed here some years ago. )

 

[KimB]

I'd say it's more likely for somebody to do it manually using a fragment program (On the NV30...unlimited texture reads). This is the only reason you'd ever need to do it, and doing it this way would be vastly superior, since the only problems with MSAA are with regard to textures, some just need additional AA beyond texture filtering. It's far superior to SSAA as it will produce essentially the same image quality without the performance hit (supersampling only applied where it's needed, not all over).