Driver-applied MLAA on D3D9 games like GTA 4?

[DieH@rd]

Did you restart after hotfix?

 

[BRiT]

I don't believe the MLAA step to be 'free'. It was said to be as costly as the edge-detect shader anti-aliasing. Now it may seem to have no performance impact, but it hasn't been tested extensively on games that already use a lot of shading processing. The small handful of games I've seen benchmarked with it so far can't be said to push shaders.

 

[no-X]

According to users it currently works only with DX9 titles. Many DX9 titles aren't ALU-limited, so the MLAA is almost free. More significant impact on performance was reported for Darksiders, but I haven't exact numbers.

 

[Malo]

Does a really nice job with Risen, I have the Gothic 4 demo, didn't do anything for that fence at the beginning.

That was one of the worst parts of the demo, disgusting aliasing on that fence.

 

[Tamlin]

You can force AA in SC2, I'm interested as hell in GTA4 though. I've never found a good way to make it work in that.

You might have tried it already, but here are GTA4:









Taken from here

They are jpeg's, so they might have some compression artifacts.

 

[homerdog]

GTA IV all modded out like that is one of the best looking games I've ever seen.

 

[Babel-17]

Are you using the dx3d9.dll fix for Fallout New Vegas? Cause that causes that in that game.

OT but there's now one possibly better suited for ATI cards.

http://www.newvegasnexus.com/downloads/file.php?id=34970

 

[DavidGraham]

What is the difference between DLAA and MLAA ?

I remember reading somewhere about DLAA being much better , have less artifacts and is faster to implement too .

 

[Malo]

Dave on MLAA from the other thread:

With regards to MLAA, the DC comment is a little red herring; it started life as in HLSL but obviously what is present in the driver is not HLSL; it is platform indepentant so it should be in XP aready) and also API independant - we've had some reports of the driver not picking up the correct key's in DX10/DX11 but it should work, additionally OpenGL is still being worked on.

With regards to the hardware support, the LDS really is paying off here because we need to store chunks of the frame for the analysis and blending stages. Without the LDS you're probably going to be doing roundtips to memory constantly, which is going to have a big impact in performance.

 

[digitalwanderer]

OT but there's now one possibly better suited for ATI cards.

http://www.newvegasnexus.com/downloads/file.php?id=34970

Much better ATi d3d9.dll, better trees and transparent AA.

 

[brain_stew]

I don't believe the MLAA step to be 'free'. It was said to be as costly as the edge-detect shader anti-aliasing. Now it may seem to have no performance impact, but it hasn't been tested extensively on games that already use a lot of shading processing. The small handful of games I've seen benchmarked with it so far can't be said to push shaders.

Gothic 4 and GTA 4 aren't shader heavy games?

 

[pcchen]

I don't believe the MLAA step to be 'free'. It was said to be as costly as the edge-detect shader anti-aliasing. Now it may seem to have no performance impact, but it hasn't been tested extensively on games that already use a lot of shading processing. The small handful of games I've seen benchmarked with it so far can't be said to push shaders.

It's not "free" in the absolute sense but it can be very cheap so it's practically free. Since MLAA is a screen space algorithm, the ALU resources required depends only on the resolution and the frame rate. Basically, you can see it as Nx[number of pixels]x[frame rate] as the "FLOPS" required, where N is the operation required for a pixel on average (assuming this is not bandwidth limited).

For 1920x1080 60fps, it's Nx0.125 GFLOPS. So for a > 2 TFLOPS part it can essentially afford N to be something like 100 (i.e. 100 operations per pixel on average) and still only eats less than 1% of the theoretical computation power.

 

[BRiT]

Gothic 4 and GTA 4 aren't shader heavy games?

GTA 4 was released in 2008. No. It's not shader heavy.

Gothic 4 Arcania still only uses DX9 as it's playable on WinXP. I haven't looked into it nor had I seen performance numbers of MLAA with Gothic 4.

 

[AlNom]

hm... gallery update: http://www.iryokufx.com/mlaa/gallery/index.html

 

[homerdog]

Based on that (especially the Unigene stuff), MSAA is definitely still going to be my preferred form of AA. But in cases where MSAA is not available or is too expensive, MLAA looks much better than nothing.

I would like to see some examples of MSAA and MLAA being used together.

 

[Albuquerque]

Looking at those shots, I find very few cases where MSAA gets something "right" where MLAA wasn't able to do it just as well. Still, alpha textures are getting hammered and neither of those algorithms help really noisy shaders or fine details in the distance (ie chain link fences or foliage where you get 1-pixel coverage issues.)

I'm gonna piece together a set of 10.10a hotfix MLAA-hacked mobility drivers today and get this running on my laptop. My i5-520m and 5650 already do pretty well with 2xSSAA under Oblivion and Fallout 3. I wouldn't mind seeing what 8xMLAA does as a comparison

 

[homerdog]

Looking at those shots, I find very few cases where MSAA gets something "right" where MLAA wasn't able to do it just as well.

Look at the ship in the second, third, and fourth unigene shot. MLAA is not doing much for it, and it is likely even worse in motion (when it is most distracting). And it should not be uncommon to see situations like this in many games.

Now if they could make an algorithm that could detect when MLAA is not effective and request more samples, that would be very impressive. :smile: But I suspect there are deep and complex reasons why this would not be efficient..

 

[Albuquerque]

Since I do not "play" Unigene, I did not put much weight in the issues it exhibits. The only parts that I put any real thought towards were the aliasing problems with the various ropes in the scene -- outer edges seemed OK, but near-parallel ropes had aliasing problems with almost-1-pixel-but-not-quite coverage issues causing aliased gaps between the ropes.

MSAA didn't solve much there either, only SSAA truly fixed it. In which case, meh, I'm not really that concerned. It seems that, so far, all the true gaming shots have excellent results.

 

[homerdog]

Since I do not "play" Unigene, I did not put much weight in the issues it exhibits. The only parts that I put any real thought towards were the aliasing problems with the various ropes in the scene -- outer edges seemed OK, but near-parallel ropes had aliasing problems with almost-1-pixel-but-not-quite coverage issues causing aliased gaps between the ropes.

Well I've gone and circled a few trouble spots here. :smile:
Yes it's Unigene but I see no reason why it should have issues other games don't.

MLAA:


Compared to 8xMSAA:


Honestly I could've drawn a lot more red circles but I think you know what I'm talking about now. :smile:
MSAA just produces a much cleaner and more accurate image overall. And obviously this should be more noticeable in motion.

 

[Xenus]

Pretty much everything you circled where subpixel objects so it was kind of expected. As stated before a game with a bunch of subpixel objects everywhere doesn't benefit nearly as much from MLAA