16x FSAA on NV30?

[K.I.L.E.R]

Has anyone done any testing? Dave, are you going to test it and post results?

I think it's whacky that they would include 16x FSAA.

 

[THe_KELRaTH]

I was looking at some comparative images on NVnews and I'm damned if I can see a difference.

 

[BoardBonobo]

The 43.40 dets were leaked by Dell and they are for the QuadroFX. That's why 16xFSAA is available. The QuadroFX supposedly supports AA sampling that high, but I guess it must be a mix of MSAA+SSAA and some post filtering.

 

[jb]

Ok whats the framebuffer requirements for x8/ x16 AA?

What would be the requirements for x4 MSAA + SSAA?

Would that even fit on the left over memory of the ultra?

Could the drivers be switching down to the highest AA it can handle (We know that drivers will switch down on you if you ask for an AA setting that the card can not handle)?

And they small perfromance hit going from x8 AA to x16 AA just does not add up....

 

[Thowllly]

I think it's whacky that they would include 16x FSAA.

Back in the V5 days, there were some Detonator drivers that let you use 16x AA even on the GF1.

 

[BoddoZerg]

16x on a GF1? Might as well use DX9 refrast...

 

[Bambers]

Yep I ran the 16x (straight 4x4 OGSS) on my 32mb gf256.

In fact then there were quite a few options for AA:

1x2,2x2, 2x2 + 9tap blur, 3x3 and 4x4 with 2x2,3x3 and 4x4 each having a non quality mode that kept the mip boundries at the places for the original res.


4x4 was limited to 512x348x16 on my 32mb card though . motocross madness 2 ran at smooth framerates with that however

 

[arjan de lumens]

Hmm - I remember being able to run 16X FSAA on a RIVA TNT (!) with pre-detonator drivers, even at high resolutions (1024x768) - looked great, but the performance generally didn't exceed ~1fps.

 

[Reverend]

I'm sure it can be done because if I'm not wrong I think I read that the NV30-Quadro line has this available.

 

[Dave Baumann]

Can anyone point me to these driver please?

 

[Nick[FM]]

Can anyone point me to these driver please?

Sure: http://www.winhelpline.info/download/dlm_download.php?id=60

*edit: http://www.faqxp.de/5e2635ef2983aa20cf73197a9e3b47d7/60/win2k_xp_43.30.zip for the lazy ones.. and of course http://download.guru3d.com/detonator/

 

[Dave Baumann]

Gawn!

 

[jb]

well...what do you think? what does the FSAA test show you...tell us pppplllleeeeaaasssseee

 

[BoardBonobo]

Strangely enough the drivers still only list the following modes:

4X Mixedsampling (skewed grid 8-tap filter)
4X Mixedsampling (skewed grid)
4X Multisampling (Gaussian)
4X Multisampling
4X Supersampling (skewed grid 8-tap filter)
4X Supersampling (skewed grid)
4X Supersampling (Gaussian)
4X Supersampling
4X Supersampling (LOD bias)
2X Supersampling (Quincunx)
2X Quincunx
2X Multisampling
2X Supersampling (vertical)
2X Supersampling (horizontal)

So 16x must be made up of some combination of these modes?

 

[tEd]

4x OGMS with 2x2 SS probably

 

[demalion]

So this isn't just line AA or something?

Anyways, I don't see how it could be just brute force AA just considering memory demands, so I wonder what it is.

 

[Dave Baumann]

OK, what you'll note about these drivers is that 8xS has been replaced by 8X and no longer says DX only - same with 16X. Here are the sample patterns:

8X:

16X:

What you'll notice is that the 8X and to a greater extent the 16X modes produce a slight amount of texture funkyness/blurring, heres a straight 4X shot for reference:

4X:

I would say that this is most likely the reason that 8X wasn't used in the end, and they shipped with 8xS instead - there obviously appears to be some instability in this mode with NV30.

Just to prove that it does operate under OpenGL, heres an output with Basic's earlier OGL FSAA tester:

16X OpenGL:

(Errr, you might need to use something on that shot to play around with brightness/gamma or something to see the sample points - they are there if you can't immediately see them)

Its quite interesting to see that do appear to effectively be able to make the SS+MS mixed modes work under OpenGL, when the MSAA extension does not suppoer supersampling - perhaps they are not using this method anymore, or have their own.

As I said, I think the texture corruption is why 8x/16x never shipped as they were initially said to in NV30. There is a specific reason why these mode are in these drivers, but I'm not at liberty to discuss.

 

[tEd]

4xMS with 1.5x2xSS

One can only hope they improve the AA engine on nv35

 

[OpenGL guy]

Just to prove that it does operate under OpenGL, heres an output with Basic's earlier OGL FSAA tester:

16X OpenGL:

The edge quality is scarcely better than 4x AA... If you look along the edges, you can see that there are up to 9 interpolations. However, some of the interpolations are over such a small area (one pixel) that they don't significantly help improve the image quality.

Samples taken on an ordered grid just don't make for good antialiasing: The end result is little better than 4 samples even thought nvidia claims 16.

Edit: I should say little better than 4x AA on a rotated grid.

 

[Colourless]

Dave, I'm not impressed. You should not have saved the screenshots as 256 colour GIFs. You have got dithering in the gradients, meaning they are pretty much useless to be able to tell the number of interpolations.

You really should use 24 Bit PNGs.