NVIDIA SSAA/Downsampling

[KingOfsorroW]

Some numbers from built in crysis gpu benchmark:
dx10 32bit 1920x1200

SSAA (no aa/traa) 3890x1400 (1920*2x1200*2)
min: 7.16
average: 19.38
Max:20.64

8Qx, 2xSStraa 1920x1200
min:23,59
average: 35,36
max:38,31

4x,2xSStraa 1920x1200
min:28.02
average:41.70
max: 45.37

[ElSel10]

not that crap again

What part of it is crap? People post those tunnel shots of perfect AF, while shots of real world gameplay tell a different story.

[BoredByLife]

What part of it is crap? People post those tunnel shots of perfect AF, while shots of real world gameplay tell a different story.

Bug/cheat (which is game specific) vs. the used technology for AF.

[KingOfsorroW]

This is not correct based on the images in that link. When __SSAA (TSSSA) is applied it makes the jaggies blurry. To imply that it's "crystal clear" is just an inaccurate stab at my post to create an argument . One thing you avoided when replying to my post is that TSSAA doesn't remove the jaggies. They are still there making the use of it moot.

In that review:
http://www.pcgameshardware.com/aid,7...ames/Practice/

They compared the different modes of tssaa which gives exactly teh blurring you are reffering to. They say that the transparency anti-aliiasing is now rendered all over the screen, but it's not that the image itself is rendered multiple times... The thing is that TSSAA is not the SSAA. It is basically an addition to nvidia' usual FSAA which is MSAA or combination of the MSAA and CSAA... Note that I used teh minimal amount of the TSSAA in the tests - reason: because I dont see the real need in going above that value as the settings of 4x or 8x of the TSSAA makes the picture extremly blurry. But this, is not the SSAA. The full screen SSAA doesn't use msaa or csaa nor does it need the TSSAA.

Do you have oblivion installed? If so you don't have to own a dx10 compatible gpu to see the difference between msaa+tsaa and the real ssaa. All you need is download and install the nhancer and select the supersampling mode in the fsaa modes list and set it to 2x2. Disable the transparency AA (TRAA) and see the difference in picrure quality by yourself. Well you can actually do this in any other game it's just that in oblivion you may see the difference exceptionnally well.

EDIT: I made a correction in the first post so that people don't get confused by the SGSSAA in the first link.

[Eastcoasthandle]

In that review:
http://www.pcgameshardware.com/aid,7...ames/Practice/

They compared the different modes of tssaa which gives exactly teh blurring you are reffering to. They say that the transparency anti-aliiasing is now rendered all over the screen, but it's not that the image itself is rendered multiple times... The thing is that TSSAA is not the SSAA. It is basically an addition to nvidia' usual FSAA which is MSAA or combination of the MSAA and CSAA... Note that I used teh minimal amount of the TSSAA in the tests - reason: because I dont see the real need in going above that value as the settings of 4x or 8x of the TSSAA makes the picture extremly blurry. But this, is not the SSAA. The full screen SSAA doesn't use msaa or csaa nor does it need the TSSAA.

Do you have oblivion installed? If so you don't have to own a dx10 compatible gpu to see the difference between msaa+tsaa and the real ssaa. All you need is download and install the nhancer and select the supersampling mode in the fsaa modes list and set it to 2x2. Disable the transparency AA (TRAA) and see the difference in picrure quality by yourself. Well you can actually do this in any other game it's just that in oblivion you may see the difference exceptionnally well.

EDIT: I made a correction in the first post so that people don't get confused by the SGSSAA in the first link.

The article in and of itself is discussing __SSAA (in this case SGSSAA). In which they used TSSAA to example the effects thereof. What was provided (not implied) is the discovery of SGSSAA and used TSSAA in the photo examples. This is what I responded to in my 1st post in this thread. Albeit I didn't go into any extreme detail about.

From my understanding supersampling blurs textures because it is applying anti-aliasing to shaders, alpha textures, etc. While it has the potential to sharpen edges. SSAA is a brute force method of AA. In the article it's implied that a negative LOD was used (as is usually the case when SSAA is used). Yet the image still looks blurry. In the end IMO, the image still looks blurry at a costly performance hit. I've seen it and honestly don't see any benefit from it.

[KingOfsorroW]

The article in and of itself is discussing __SSAA (in this case SGSSAA). In which they used TSSAA to example the effects thereof. What was provided (not implied) is the discovery of SGSSAA and used TSSAA in the photo examples. This is what I responded to in my 1st post in this thread. Albeit I didn't go into any extreme detail about.

From my understanding supersampling blurs textures because it is applying anti-aliasing to shaders, alpha textures, etc. While it has the potential to sharpen edges. SSAA is a brute force method of AA. In the article it's implied that a negative LOD was used (as is usually the case when SSAA is used). Yet the image still looks blurry. In the end IMO, the image still looks blurry at a costly performance hit. I've seen it and honestly don't see any benefit from it.

You still don't understand. The article discusses the SGSSAA. Not the SSAA. The SSAA probably blurs something but it doesn't come even close to the blur of any other aa method simply because it doesn't apply anti-aliasing to anything in particular and/or selectively, it simply renders the whole image several times - that's why it is called a "brute force AA". And that is SSAA. Which makes the image look crystal clear as much as it is possible at all. Really, try it for yourself instead of typing.

[Eastcoasthandle]

You still don't understand. The article discusses the SGSSAA. Not the SSAA. The SSAA probably blurs something but it doesn't come even close to the blur of any other aa method simply because it doesn't apply anti-aliasing to anything in particular and/or selectively, it simply renders the whole image several times - that's why it is called a "brute force AA". And that is SSAA. Which makes the image look crystal clear as much as it is possible at all. Really, try it for yourself instead of typing.

I see, despite what I've posted you still want to argue. It's clear I understand but ultimately the images in the article are still blurry and using that technique (with SLI configuration at that) comes at a performance penalty. Which is what I originally said in post 10

[IDANPE]

What blur ?

TSSAA X4


MSAA X4


look at the floor to see the SSAA effect.

[Eastcoasthandle]

What blur ?

look at the floor to see the SSAA effect.

The blur referenced in the OP article.
Oh and look at the jaggies on the grated flooring towards the steps. Looks exactly the same to me.

[IDANPE]

not to me..
but yes, u need at least X8 TSSAA to see it without any jaggies.

[Eastcoasthandle]

not to me..
but yes, u need at least X8 TSSAA to see it without any jaggies.

Which is what the IQ pics show in the OP's link. The very thing I commented on when I said that the jaggies looked blurry. So we are back to square one, thanks!

[RVWinkle]

NV has better overall AF quality ever since G80. In fact, there have been no visible changes of the AF algorithm for the past four years -- GF100 outputs virtually the same texture quality as the old G80. The problem with ATi's case is the severe under-sampling, visible in most high-frequency/detailed textures all over the place, like subtle shimmering and sharp MIP stage transitions. The angular pattern -- being invariant or not -- should not be used as exhaustive quality measurement.

The weird thing about your statement is that back when the G80 came out everyone was accusing nVidia of cheating with af. The reality is that ATI has always always always had better image quality than any other competitor. Other aspects like drivers are a different story of course.

[ElSel10]

The weird thing about your statement is that back when the G80 came out everyone was accusing nVidia of cheating with af. The reality is that ATI has always always always had better image quality than any other competitor. Other aspects like drivers are a different story of course.

This statement couldn't be farther from the truth. Nvidia's angle-dependent AF algorithm used in G80, G92, GT200 and Fermi runs circles around ATI's algorithm used in the R600, RV670 and RV770.

Now supposedly ATI has bested the algorithm used in Fermi with RV870, being angle-independent; however, in real world usage it is impossible to tell the difference, and in fact there is even picture evidence suggestion it still isn't as good as Nvidia's algorithm used in anything outside that tunnel program. This would suggest that in order to make up for the performance impact of a fully angle independent algorithm, textures are being undersampled compared to Nvidia.

[Cybercat]

I have no idea what SSAA looks like on NVIDIA's hardware, but having messed with ATI's SSAA, I concluded it's a pile of blurry mess.

I suppose if I wanted to clear jaggies that way I could play games with my glasses off.

[Aerwidh]

I suppose if I wanted to clear jaggies that way I could play games with my glasses off.

Built in SSAA at no performance penalty (though possibly with a mild headache, I guess)

[Manicdan]

im lost in all the acrynyms being thrown around.

i do like how they can apply AA to any object with one of the methods (TSAA?) but im sure its very costly. might as well be like rendering 600fps, but do 60fps with 10 different pixel centers to average out.

AA being object specific does suck when theres alot of object which will look very jagged. but im sure the raw bandwidth needed to clean up those things is just too much for games like crysis. but an older game (like HL2 maybe?) where you can get 200fps at the highest settings, it might be worth it to clean up those extra items. but isnt that kind what catalyst AI is suppose to do? even though it sucks at it. all i know is that little walk through with the metal fence in CCC, always looks like crap with any setting.

[Chumbucket843]

I have no idea what SSAA looks like on NVIDIA's hardware, but having messed with ATI's SSAA, I concluded it's a pile of blurry mess.

I suppose if I wanted to clear jaggies that way I could play games with my glasses off.

dont jump to conclusions! that issue was fixed in ATi cards in november or so. before if you didnt manually adjust the texture LOD it would be blurry because of lower res textures. SSAA provides the best image quality possible.
http://www.pcgameshardware.com/aid,6...eviews/?page=6

[Macadamia]

actually it's kinda ironic.

DX10/11 doesn't let you set LOD as far as I know so SSAA will introduce blurring (nV's SSAA only works on DX10/11 as of now)

DX9's SSAA on Radeon 5ks have LOD offset so they're not that blurry anymore (but different compared to refras still); nVidia cards can get the same effect through nHancer I suppose.

[Cybercat]

dont jump to conclusions! that issue was fixed in ATi cards in november or so. before if you didnt manually adjust the texture LOD it would be blurry because of lower res textures. SSAA provides the best image quality possible.
http://www.pcgameshardware.com/aid,6...eviews/?page=6

I tried it VERY recently, in World of Warcraft.