PDA

View Full Version : NVIDIA antialiasing mode thread - version 2.0


Pages : [1] 2

StealthHawk
02-26-04, 03:51 PM
Welcome to version 2.0 of my NVIDIA anti-aliasing mode thread. The old post can be found here (http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=15821). That thread was segmented and had some outdated information, so a new thread was required for easier reading. All screenshots are high quality .png. Screenshots were taken with a GeForce3 and Detonator 44.03 with the exception of 2x1 and 1x2, which were taken with 53.03. Screenshots with a "GFFX" suffix succeeding the FSAA mode name were taken on a GeForceFX5900Ultra with Detonator 45.23. All screenshots were compared by me with what was actually displayed on the screen, so I can attest to their accuracy.

Depending on what card you own, and what driver you use, different FSAA modes will be available to you. Driver 42.68 or higher is required to use (most) of the following modes. GeForce3, GeForce4, or GeForceFX users all have the same FSAA modes available after using tweaker programs, the control panel may or may not offer all modes. I suggest using either RivaTuner (http://download.guru3d.com/rivatuner/) or aTuner (http://www.3dcenter.de/atuner/index_e.php).

Forceware/50 Series Driver
Forceware has brought several changes to FSAA modes offered in the control panel. "2xQ" in the control panel is Quincunx. "6x" in the control panel is really 12x(also known as 16x in D3D in some earlier drivers). "8x" in the control panel is 8xS. What this means to you is that if you only use the control panel, "6x" and "8x" will offer better IQ than they did in earlier drivers. Of course, this also means they will have a larger performance hit. All FSAA modes may be re-exposed and properly named by using RivaTuner and by making changes to the .cfg file. See the second post in this thread for details. Two new(and several other weird modes which seem to cause texture aliasing) modes were added in this driver, 1x2 and 2x1 Super Sampling.

The FSAA modes below use their 40 series driver(42.68+) naming conventions.

Edge quality(Listed from worst to best)
0x (http://www.mrclam.com/~stealth/0x.png)
1x2 Super Sampling (http://www.mrclam.com/~stealth/1x2.png)
2x1 Super Sampling (http://www.mrclam.com/~stealth/2x1.png)
2x (http://www.mrclam.com/~stealth/2x.png)
QCA (http://www.mrclam.com/~stealth/qca.png) | QCA GFFX (http://www.mrclam.com/~stealth/FXqca.png)
4x (http://www.mrclam.com/~stealth/4x.png)
4x Super Sampling (http://www.mrclam.com/~stealth/4xSS.png)
4xS (http://www.mrclam.com/~stealth/4xS.png)
6x (http://www.mrclam.com/~stealth/6x.png)
8x (http://www.mrclam.com/~stealth/8x.png)
4x+9tap (http://www.mrclam.com/~stealth/4x9.png) | 4x+9tap GFFX (http://www.mrclam.com/~stealth/FX4x9.png)
12x (http://www.mrclam.com/~stealth/12x.png)
8xS (http://www.mrclam.com/~stealth/8xS.png)

For comparison, you can find some shots taken on a GeForceFX5900Ultra by MikeC here (http://www.nvnews.net/reviews/sapphire_radeon_9800_pro/images/aa_comparison.shtml)(40 series driver. Note: "6xS" = 6x and "8xS" = 8x.). Or shots taken on a GeForceFX5950Ultra here (http://www.nvnews.net/files/graphics/screenshots/fsaatester/aa_comparison.shtml)(Forceware/50 series. Note: "6xS" = 12x and "2x2 SS" = 4x Super Sampling). For the most part, FSAA modes look exactly the same or are perfectly equivalent when comparing a GeForce3 and GeForceFX. QCA looks better on a GeForce3. 4x+9tap looks better on a GeForceFX than it does on a GeForce3. On the GeForce3 shot, you can see some weird anomalies where part of the image is distorted.

1x2 only does AA on horizontals and diagonals, and 2x1 only does AA on verticals and diagonals. Since neither of these modes do AA at all angles, they are gimmicks which also come at a high performance hit.

While 6x and 8x provide slightly better IQ at vertical and diagonal angles, 4xS is superior at horizontal angles. In my opinion, 4xS provides a better IQ/performance ratio than 6x or 8x.

You may notice that 6x and 8x look almost exactly the same. 8xS and 12x are also almost exactly the same, while 8xS is slightly better on some diagonals. 4x+9tap provides very good edge AA quality, roughly equal to 8xS/12x. Of course, straight edge quality does not tell the whole story. Let's see what each FSAA mode does to a 3d rendered scene to get the whole image quality picture and draw some accurate conclusions.

3d scene quality(Grouped for similarity and worst to best where applicable)
QCA (http://www.mrclam.com/~stealth/qca3d.png) | QCA GFFX (http://www.mrclam.com/~stealth/FXqca3d.png)
4x+9tap (http://www.mrclam.com/~stealth/4x93d.png) | 4x+9tap GFFX (http://www.mrclam.com/~stealth/FX4x93d.png)
0x (http://www.mrclam.com/~stealth/0x3d.png)
2x (http://www.mrclam.com/~stealth/2x3d.png)
4x (http://www.mrclam.com/~stealth/4x3d.png)
1x2 Super Sampling (http://www.mrclam.com/~stealth/1x23d.png)
2x1 Super Sampling (http://www.mrclam.com/~stealth/2x13d.png)
4xS (http://www.mrclam.com/~stealth/4xS3d.png)
6x (http://www.mrclam.com/~stealth/6x3d.png)
8x (http://www.mrclam.com/~stealth/8x3d.png)
12x (http://www.mrclam.com/~stealth/12x3d.png)
4x Super Sampling (http://www.mrclam.com/~stealth/4xSS3d.png)
8xS (http://www.mrclam.com/~stealth/8xS3d.png)

For a more comprehensive look at scene quality, see my "Effects of Super Sampling AA on Textures" post here (http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=23427).

Both QCA and 4x+9tap are blurring the whole scene.
4xS clears up vertical parts of the scene compared to standard 4x or 2x MSAA.
6x and 8x clear up horizontal parts of the scene compared to 2x or 4x. Notice that 8x actually clears up some angles better than 6x does.
12x clears up vertical angles better than 6x or 8x. 12x clears up horizontal angles at least as well as 6x, but not as well as 8x. 12x cleans up vertical angles at least as well as 4xS.
8xS clears up all angles better than or at least as well as 4xS, 6x, or 8x.
4x Sumper Sampling cleans up all angles to the same degree as 8xS.

Other Information
GeForce3, GeForce4Ti, GeForce4MX, and GeForceFX achieve FSAA in a variety of ways. Through a filter(QCA, 4x+9tap), Multisampling(2x, 4x), Super Sampling(1x2, 2x1, 2x2), or a combination/hybrid of MS and SS(all modes higher than 4x), the breakdown of which is as follows:4xS: 2xRGMS + 1x2OGSS
6x: 4xOGMS + 1.5x1OGSS
8x: 4xOGMS + 2x1OGSS
8xS: 2xRGMS + 4xOGSS
12x: 4xOGMS + 1.5x2OGSS

Performance numbers done by 3DCenter can be found here (http://www.3dcenter.de/artikel/geforcefx_aa_modi/index3_english.php), which match my own findings. Owners of high end GeForceFX cards will find that pure Multisample modes have a much lower performance hit than those shown at 3DC.

If you want to see the sample patterns, download the attachment in this post. You will see that 4xS, 6x, 8x, 8xS, and 12x all are hybrid modes with employ Super Sampling and Multisampling. Hence why these modes clear up textures.

StealthHawk
02-26-04, 03:52 PM
By default, I don't believe that either aTuner or RivaTuner expose all the FSAA modes. Personally, I recommend RivaTuner (http://download.guru3d.com/rivatuner/) because you can enable or disable different FSAA modes on a whim.

All you have to do is edit the RivaTuner.cfg file found in your RivaTuner installation directory. Open RivaTuner.cfg and scroll all the way to the bottom until you reach the section entitled [ForceWareFSAAModes]. Then replace what's there with this:FSAAMode01 = 2 x 1 (SS)
Required01 = 170h,18Fh,200h,
FSAAMode02 = 1 x 2 (SS)
Required02 = 170h,18Fh,200h,
FSAAMode05 = 2 x 2 (4x SS)
Required05 = 200h,
FSAAMode0E = 2x
FSAAMode0F = Quincunx
Required0F = 170h,18Fh,200h,
FSAAMode10 = 4x
FSAAMode12 = 4x 9-tap
Required12 = 170h,18Fh,200h,
FSAAMode13 = 4xS
Required13 = 170h,18Fh,200h,
FSAAMode17 = 6x(12x)
Required17 = 200h,
FSAAMode19 = 8x(8xS)
Required19 = 200h,
FSAAMode1A = 16x
Required1A = 308h,30Fh,338h,33Fh

FSAAMode16 = old 6x
FSAAMode18 = old 8x

FSAAMode0C = Fun Mode 01
FSAAMode14 = Fun Mode 02
FSAAMode0A = Fun Mode 03

Alternatively, you could download the zip file attached to this post where I edited the ForceWareFSAAModes section for you.

stncttr908
02-26-04, 08:18 PM
I think I said this in the previous thread but it bears repeating.

Can someone please get this guy a front page article!

SnakeEyes
02-27-04, 07:57 AM
Agreed. This is good investigating work, and a great presentation of it. It would be really nice to see it shown where all nVidia-based card owners that visit the site are most likely to notice it.

XanderF
03-03-04, 01:25 AM
Out of curiosity, how does this map out to the FSAA options in the 56.56 control panel?

The options currently are:
2x
2xQ
4x
6xS
8x

Most are pretty obvious, but I'm curious about 6xS. New mode? Or, 4xS relabelled? If it's a new mode, where did 4xS go?

It still shows up in RivaTuner, but that choice is less than totally desirable currently. First, no way of knowing if it's the 'same old' 4xS that has been in the driver control panel for a while. This is kinda backed up by the Rivatuner startup comment that it doesn't yet support this Det build. Secondly, what I'm looking to do is set this up as an application profile - something that RivaTuner's options don't take into account yet.

XanderF
03-03-04, 01:29 AM
Okay, bit of tinkering here with RivaTuner and the new control panel.

When setting both DX and OpenGL in RivaTuner to '4xS', and going into control panel under the 'Global Driver Settings', the slider shows 'Off' as the option. However, in the detail window it DOES say '4xS'.

Setting the slider to '6xS' and then taking a look at RivaTuner shows it now set to the old '6x' option.

Not sure what all that means. Apparently, 4xS isn't an option for application-detected FSAA....kinda a drag.

ChrisRay
03-04-04, 11:36 PM
http://www.3dcenter.de/artikel/ati_nvidia_aa_performance/index_e.php



Has anyone read this article? I was reading and simply couldnt believe what I read. Saying 4x Super Sampling has the same effect as 4x Multi Sampling.

Make claims with "AA" percentages, I am beginning to wonder if they even tested some of these modes, Because their conclusions on some of their results seem pretty impractical to me.




Did 4x Super Sampling EVER have the same pattern on the FX cards as it does multi sampling? Didnt for my Geforce 4 Ti

Bambers
03-05-04, 03:38 AM
Originally posted by XanderF
Most are pretty obvious, but I'm curious about 6xS. New mode? Or, 4xS relabelled? If it's a new mode, where did 4xS go?


6xS has been in and out of the drivers for ages. It's often confused for 12x because the app that shows the sample patterns doesn't handle non interger supersample factors very well and so using that 6xS looks almost identical to 12x.

In fact the screenshots linked to in the top post of this thread are actually 6xS and not 12xAA. You can see this in the 'dithering' in some of the horizontal mid gradients caused by the pattern being upside down everyother pixel.

ChrisRay
03-05-04, 03:49 AM
I think its interesting but this program is reporting me different results.


http://www.users.on.net/triforce/d3d_fsaaviewer/d3d_fsaaviewer-3.zip

There seems to be a disagreement that I am trying to pinpoint about how 4x Super Sampling is handled.


this tool is reporting it very differently.



Stealth have you updated it?


Anyway, Download the tool and see for yourself.

StealthHawk
03-06-04, 10:13 PM
Originally posted by XanderF
Out of curiosity, how does this map out to the FSAA options in the 56.56 control panel?

The options currently are:
2x
2xQ
4x
6xS
8x

Most are pretty obvious, but I'm curious about 6xS. New mode? Or, 4xS relabelled? If it's a new mode, where did 4xS go?

I'm assuming that the control panel options in 56.56 are the same as in every other 5x.xx driver, but I haven't tested it. 4xS was removed from the control panel in all 5x.xx drivers. No idea why :confused:

Originally posted by ChrisRay
Has anyone read this article? I was reading and simply couldnt believe what I read. Saying 4x Super Sampling has the same effect as 4x Multi Sampling.

Make claims with "AA" percentages, I am beginning to wonder if they even tested some of these modes, Because their conclusions on some of their results seem pretty impractical to me.




Did 4x Super Sampling EVER have the same pattern on the FX cards as it does multi sampling? Didnt for my Geforce 4 Ti

Yes, it has the same pattern- both use Ordered Grid. The reason why the sample patterns look different is that one is Multisampling and one is Super Sampling. The FSAA tester you link to shows geometry on one side, and textures on the other. MSAA of course only does work on geometry, and SSAA does work on both. Edge-wise(not counting alpha textures), they should be the same or at least equivalent.

Originally posted by Bambers
In fact the screenshots linked to in the top post of this thread are actually 6xS and not 12xAA. You can see this in the 'dithering' in some of the horizontal mid gradients caused by the pattern being upside down everyother pixel.

No, it is 12x. There are 12 samples being taken.

ChrisRay
03-15-04, 04:20 PM
Question Stealthawk. I havent seen this 6xS mode Ever, Whatever happened to it?


Noticed Atuner doesnt really offer that many AA modes, Or they seem to be disabled now on my driver set.. hmm..


http://www.3dcenter.de/artikel/ati_nvidia_aa_performance/index2_e.php

StealthHawk
03-16-04, 06:54 PM
Originally posted by ChrisRay
Question Stealthawk. I havent seen this 6xS mode Ever, Whatever happened to it?

My guess is that the "real" 6xS was killed because it produces artifacts? Anyway, it seems to have been removed a long time ago.

Noticed Atuner doesnt really offer that many AA modes, Or they seem to be disabled now on my driver set.. hmm..

What driver are you using?

ChrisRay
03-16-04, 07:40 PM
Originally posted by StealthHawk
My guess is that the "real" 6xS was killed because it produces artifacts? Anyway, it seems to have been removed a long time ago.



What driver are you using?


the 56.85 drivers ect,


Was noting that the old 8x/6x arent thetre and neither is 4xQuincunx.


Not that it matters, hardly any reason at all to use those modes. And They unlocked in Riva tuner.



in Regards to that 6xS, It had a pretty unique pattern, I wonder how efficient it was compared to the other modes.. Look rotated too... Wonder why it caused artifacts,

And I wonder why Nvidia doesnt use Rotated Grids for more of its Super Sampling Modes,

StealthHawk
03-17-04, 03:46 PM
Originally posted by ChrisRay
And I wonder why Nvidia doesnt use Rotated Grids for more of its Super Sampling Modes,

Yeah, the modes that use Rotated Grids offer the better IQ compared to ones that don't. Witness 4xS and 8xS compared to 6x, 8x, or 12x. The truth is that they wouldn't be able to make many more modes that have RG components, since the cards only support 2x RGMS. And I guess the highest mode of SS that NVIDIA is willing to use is 2x2 OGSS either due to physical limitations, or for performance reasons.

They could probably make some other modes, but they would look a lot like 4xS, which would kind of defeat the point of expending the extra effort. Especially when NVIDIA is officially de-supporting 4xS, removing it from the FSAA slider and all that. I'm glad NVIDIA made the move to replace the old 6x and old 8x with higher IQ versions in the control panel. Maybe someone from NVIDIA read my old post and agreed with me that those modes weren't very useful :p

ChrisRay
03-17-04, 05:39 PM
Well Stealth, I have heard Rotated Grid Super Sampling can cause artifacts in many cases.

For Example. ATis Quality 4x AA is Rotated Grid Super Sampling. But Apparently it reverts back to Ordered if any call for fog is requested (Dont understand why)

Maybe thats why Nvidia refuses to implement pure Rotated Grid Super Sampling modes? I thinka RGSS implement of 2x2 would be great. and their 2x SS modes could stand some rotation :)

Makes me wonder why Nvidia even bothered implementing 2x on the geforce 2 cards, It didnt do a thing for horinzontal edges...

SnakeEyes
03-17-04, 08:55 PM
If that is true about ATIs RGMS (they don't allow super sampling under Windows drivers, though maybe on the Mac..), it would be strange. I can't imagine a single reason or condition where a rotated grid vs. ordered grid would cause any strangeness beyond hitting different angles in the scene more efficiently (90vs 45, etc.). If you run across anything explaining this, please link it up, as I'd love to see what the deal is (<-- Snake is really curious about rendering stuff, even if he IS a n00b when it comes to understanding the really technical side of things -->).

ChrisRay
03-17-04, 10:04 PM
Originally posted by SnakeEyes
If that is true about ATIs RGMS (they don't allow super sampling under Windows drivers, though maybe on the Mac..), it would be strange. I can't imagine a single reason or condition where a rotated grid vs. ordered grid would cause any strangeness beyond hitting different angles in the scene more efficiently (90vs 45, etc.). If you run across anything explaining this, please link it up, as I'd love to see what the deal is (<-- Snake is really curious about rendering stuff, even if he IS a n00b when it comes to understanding the really technical side of things -->).


I honestly dont know. I know ATIS 4x is Rotated Grid and so its 2x in Quality Mode, Unless Fog is enabled AFAIK, Then it reverts back to ordered, Since Super Sampling affects textures too its not impossible to assume that it could do something that would cause rendering errors,

Nvidia has been pretty adamant about keeping most of its Super Sampling Methods ordered.

SnakeEyes
03-18-04, 08:41 AM
Maybe some of the Beyond3D gurus will read these, and if they know what's going on will feel free to enlighten us. :)

StealthHawk
03-18-04, 07:38 PM
Originally posted by ChrisRay
For Example. ATis Quality 4x AA is Rotated Grid Super Sampling. But Apparently it reverts back to Ordered if any call for fog is requested (Dont understand why)

Are you talking about R200's FSAA? Anyway, where did you here this?

Makes me wonder why Nvidia even bothered implementing 2x on the geforce 2 cards, It didnt do a thing for horinzontal edges...

The reason they implemented FSAA at all was to steal the Voodoo5's thunder. And of course, 2x has a much smaller performance hit than 4x, which was even more important at the time of a card like the gf2.

ChrisRay
03-18-04, 10:56 PM
Originally posted by StealthHawk
Are you talking about R200's FSAA? Anyway, where did you here this?



The reason they implemented FSAA at all was to steal the Voodoo5's thunder. And of course, 2x has a much smaller performance hit than 4x, which was even more important at the time of a card like the gf2.


Ya talking about about the r200s AA, Only reason I was privy to this was because way way back around Cat 3.0 release, (Maybe cat 3.2) The R200s AA quality was improved,

But something about z-fog forced the r200 back into Ordered Grid Mode because it caused an onscreen corruption. They were talking about the futuremark forums :P

StealthHawk
03-19-04, 01:59 AM
Have you tested this with a sampling pattern viewer?

ChrisRay
03-19-04, 02:36 AM
Originally posted by StealthHawk
Have you tested this with a sampling pattern viewer?


Well How can you enable Fog With a viewer?? ATIs r200 Gives Rotated Grid in Quality Mode,

and Ordered Grid in Performance Mode. It was my understanding it reverted back to Ordered(Performance mode) If it was enable. Hmm.


The R200 has like 10 Super Sampling Methods

2x Perf/Quality

3x Perf/Quality

4x Perf/Quality

5x Per/Quality

6x Per/Quality


Some of these modes completely useless, Like 3x/5x But they are interesting to see the sample Patterns for.

Like 3x is pretty bizarre, Should do a r200 AA comparison thread. Too bad ATI limits the resolutions on half of these AA modes, Rendering some of them alot less fun.

StealthHawk
03-19-04, 02:03 PM
What I meant was have you tested Quality 4x on a R200 to see that normally it uses RGSS now, when there is no fog?

BTW, I wanted to continue this conversation via PM, but your inbox is full :(

ChrisRay
04-01-04, 07:25 PM
Stealthhawk still havent been able to test or Compare,


Tho I'm later gonna try this method of Comparing AA shots,


One thing I have noticed, You Can still enable 16x in Drivers and the Drivers recognise it (Even Nvidias COntrol Panel)

But it does not seem to work in Direct3d (In D3d it reverts back to old 8x I think) (Pre8xS)

Sorry about jpeg but if you magnify on the edges it does appear something ie being done on this screen.


I chose this app because its easy to switch between d3d and openGL with wc3.


First Shot is 16xAA

Next is 8xS



Apologize about compression, I really wanted png for these screenshots, And I'd be willing to e-mail png shots to anyone who has a spot for them


Gonna do some performance tests between the various modes. (Tho 8x D3d Mode can be different) since in its d3d

So lets see...

ChrisRay
04-01-04, 07:25 PM
Heres 16x in D3d (which is definately 8x in D3d)


I know the screens arent exactly the same. But compare the non moving background rather than the elf herself or the grass ect