7900 GTX Anti-Aliasing Investigated

[BFG10K]

I hope you find it an interesting read.

Click.

 

[tuteja1986]

awesome article

Dude you should apply for editor job at some tech site. Its easy , just create a portfolio with all your works and send emails to the senior editors at many Tech Hardware site. Geezee try cnet , they sure as hell can use someone as smart as you , because they got bunch idiots doing horrible hardware review. Go check gamespot.com hardware editor's review , "shockingly terrible" ? these people get paid around 80K to 120K easy.

 

[videopho]

BFG10k
In the FSx game, I've found changing the aa & af really doesn't change the framerate as much as I've seen in any other games and your investigation in F/C verifies my finding too. May be as little as 5-6% in penalty in fps as you move up the aa/af slides in the control panel. Nonethless, my GTX card can only muster 9-10fps when on the runway and ~19-20fps in the air, using 2aa/4af @ 1920X1200 resolution with everything else set at mid-level setting eg sceneries, cloud etc.

That being said, I'm wondering the FSx game engine is designed or geared more toward dx-10 gen g/c. I'd love to see some benchmarks on the FSx preceeding the G80 hard launch.
Great article of yours, btw.

 

[BFG10K]

awesome article

Thanks - not many responses though. Must be a quiet weekend or something.

May be as little as 5-6% in penalty in fps as you move up the aa/af slides in the control panel.

If the 8xS and 16xS modes are not causing large performance hits for you it probably means the game in question is very CPU limited.

 

[yhelothar]

Nice article.
And just to think that old geforce2's only came with 2xSS and 4xSS.. they really weren't going anywhere using such powerful anti aliasing algorithms on such feeble cards.

 

[schneiderguy]

were the benchmarks run on a single 7900gtx or SLI'd 7900gtx's?

edit:nevermind i looked at the link in your sig

 

[xtknight]

Very well-written, all-encompassing, and I like it that the screenshots are all in exactly the same player position among the modes. Say a texture is 512x512, and, hypothetically, that the texture is being displayed on a wall and it's exactly 512x512 from the player's point of view. Now say the texture contains a line (y=x). If 2x2SSAA were to be done on that scene, would the line in the texture look finer? Is it able to improve the quality of jagged elements in a 512x512 texture by scaling it to 1024x1024 and then downscaling it again?

 

[BFG10K]

Now say the texture contains a line (y=x). If 2x2SSAA were to be done on that scene, would the line in the texture look finer?

Yes. Because full screen super-sampling takes texture samples it anti-aliases the entire texture including aliasing within itself.

So for example if your texture was solely composed of cross-crossing lines they would be anti-aliased in a similar fashion to how multi-sampling would anti-alias criss-crossing polygon edges.

This is exactly the effect you are seeing on the green grille to the right of the gun. You also see it on the (non transparent) texture lines just above the green lights on the back wall (look at the difference between 4x and 8xS in that area).

Is it able to improve the quality of jagged elements in a 512x512 texture by scaling it to 1024x1024 and then downscaling it again?

Full screen super-sampling renders the whole scene bigger and the textures are simply mapped to the large resolution. The actual textures themselves are still the same size however.

It's basically the same as raising the resolution except the resolution is raised internally and then scaled back down to the original resolution with the excess pixels being used to smooth out the remaining pixels.

 

[potato28]

Wow good comparison BFG10K! You should really consider writing for a tech site.

 

[josh6079]

Nice. Very nice.

:thumbsup:

 

[Gstanfor]

I reccomend people save BFG10K's images to their machines, then grab a copy of The Compressonator and use the image comparison function.

 

[Dethfrumbelo]

Some people in that thread were questioning the 'almost perfect' assessment. How clean it looks also depends on the pixel pitch of your monitor and whether it's an LCD or CRT. If they're looking at these images on a 19" LCD at 1280x1024 the aliasing will be much more noticeable than 1920x1440 on a 19" CRT, a world of difference.

 

[CP5670]

As others have said, excellent review. It might also be interesting to see how the top 4x4 SS mode looks. You would have to turn down the resolution a lot, but I'm wondering how it compares to 16X on those grate-like textures.

Some people in that thread were questioning the 'almost perfect' assessment. How clean it looks also depends on the pixel pitch of your monitor and whether it's an LCD or CRT. If they're looking at these images on a 19" LCD at 1280x1024 the aliasing will be much more noticeable than 1920x1440 on a 19" CRT, a world of difference.

Well, I wouldn't call it "literally perfect" either, but it's definitely a substantial improvement over any of the other modes.

These SS modes in general are very nice. I find them too slow in newer games, but they do a lot for almost any game older than two or three years.

 

[BFG10K]

Some people in that thread were questioning the 'almost perfect' assessment.

When I look at the images @ 1600x1200 on my CRT the railing bars are practically perfect with 16xS.

How clean it looks also depends on the pixel pitch of your monitor and whether it's an LCD or CRT.

True but this applies for any kind of screenshot comparison. Regardless of the monitor you can still make a comparison between the images though.

It might also be interesting to see how the top 4x4 SS mode looks

It would be tough to make any kind of comparison because 4x4 doesn't work at resolutions above 1024x768.

 

[BFG10K]

Bump for those that missed it.

 

[JAG87]

good article dude.

nothing I did not know, but still a good read.

so what do you think the best performance/quality ratio is? I think 4xAA with 8xAF or 16xAF is the sweet spot. Anything higher is not noticeable and kills your fps. whats your opinion?

 

[BFG10K]

I think from a performance/quality ratio 4xAA is the best because it significantly reduces jagged edges with a very reasonable performance cost. However 8xS does look a lot better in most cases, even at lower resolutions.

I usually run a minimum of 4xAA but if I can get 8xS without dropping below 1280x960 I'll use it instead.

As for 16xAF, I've been running that as a mandatory minimum for four years now.

 

[Noema]

Originally posted by: josh6079
Nice. Very nice.

:thumbsup:

:thumbsup: X2

 

[Noema]

Originally posted by: Gstanfor
I reccomend people save BFG10K's images to their machines, then grab a copy of The Compressonator and use the image comparison function.

This is a very cool tool; thanks for linking it. Makes the difference much easier to appreciate.

 

[BFG10K]

Another bump in case anyone's interested.

 

[Munky]

Good writeup. :thumbsup:

 

[brikis98]

awesome job... i think i need to spend time tweaking my game settings now

 

[Markbnj]

Seriously, one of the best and clearest explanations of how anti-aliasing works that I have read, and I have been reading explanations of anti-aliasing for about twenty years now . Well done.

 

[DAPUNISHER]

:sun: ^ Thank you for taking the time and effort of writing this up. for our edification and/or education.

 

[SonicIce]

2post