TXAA Anti-Aliasing Makes Its Debut In Latest Update For The Secret World

[TimothyLottes]

I'm a bit confused by those calculations. I very often play with 4xSGSSAA and it usually carries a performance hit of about 50% vs. no AA. Meaning, if I have 100fps without any AA, I have 50fps with 4xSG.

That could easily be because the game is CPU bound with no AA (or GPU/CPU sync limited even if the CPU isn't loaded). Here is the math: 4xRGSSAA = 4x the pixel shading as no-AA. Double the pixel area = 2x the pixel shading (OGSSAA). 4xMSAA is base line 1x the area with no edges, then perhaps up to 0.25x to 0.5x more samples. So if you take 4xMSAA at 1.5x the amount of shaded samples compared to no-AA, then double the area you are at 3x vs 4x with 4xRGSSAA. One other factor on top of that is that increasing area via OGSSAA is more efficient on the GPU compared to RGSSAA due to how the super-sampling works in the pixel shader with MSAA.

Do you think it would be possible to get a more effective down-sampling filter for this hack? Also, there is a bug that prevents 3840x2160 on Kepler cards (red screen bug). Maybe there could be something done about that? Sorry to be a little OT, but it was too tempting after your very appreciated comment

Better down-sampling filter in on my list, but some might not like it as it would increase the amount of filtering (vs just point sampling as currently done). I've found that Kepler isn't allowing me to do custom resolutions over 2560x. So I've been intending to file a bug on that, BTW when I get the time. However I've never seen the red-screen from down-sampling.

 

[BallaTheFeared]

OT: Wheres my GK110 at Timmy?

 

[TimothyLottes]

They're not just failing to anti-alias fake geometry like trees, but somehow they're only anti-aliasing a limited selection of polygons too, which is just weird.

There are other factors with MSAA which may or may not be effecting BF3. For example the technique on slide 33 on this presentation could be an issue. It is easy to over optimize MSAA by being too liberal on the mask used to reduce shading cost for "complex" pixels. Another challenge for a game which is doing physically correct lighting and post processing with MSAA is that post processing needs HDR inputs. However if you tone-map after standard MSAA (which is needed for correct post), then the standard MSAA box filter stops "working" on mixed HDR/LDR edges.

 

[boxleitnerb]

Better down-sampling filter in on my list, but some might not like it as it would increase the amount of filtering (vs just point sampling as currently done). I've found that Kepler isn't allowing me to do custom resolutions over 2560x. So I've been intending to file a bug on that, BTW when I get the time. However I've never seen the red-screen from down-sampling.

Yeah exactly. When trying to create a custom resolution above ca. 2880x1620 on Kepler, you get a red screen or an error message. Interestingly, even on a puny GT520, 3840x2160@60Hz is possible without a hiccup.
The question about the better down-sampling filter was in regard to the driver hack, not TXAA. You're closer to the source, so maybe...

Enough OT from me, thx for listening!

 

[TimothyLottes]

Yeah exactly. When trying to create a custom resolution above ca. 2880x1620 on Kepler, you get a red screen or an error message. Interestingly, even on a puny GT520, 3840x2160@60Hz is possible without a hiccup.
The question about the better down-sampling filter was in regard to the driver hack, not TXAA. You're closer to the source, so maybe...

Enough OT from me, thx for listening!

BTW, I'm able to reproduce the red screen on my GTX680 in TSW, and I'm able to reproduce the resolution cap. So I've filed these bugs at NVIDIA so that they can get in the pipeline to be fixed.

Thanks again for the note on the problem!

 

[thilanliyan]

It's good to have technical people from companies (I don't mean sales people) on these boards listening to what enthusiasts are yearning for.

Thanks to you Timothy Lottes for getting involved in this conversation. It is great to see. :thumbsup:

 

[boxleitnerb]

I agree, thank you!

 

[ViRGE]

Better down-sampling filter in on my list, but some might not like it as it would increase the amount of filtering (vs just point sampling as currently done). I've found that Kepler isn't allowing me to do custom resolutions over 2560x. So I've been intending to file a bug on that, BTW when I get the time. However I've never seen the red-screen from down-sampling.

I don't know what kind of information you have access to internally, but publicly at least you're actually not the first person to run into that high resolution downscaling problem. Apple ran into it in developing the Retina MacBook Pro, and apparently it was bad enough that there wasn't a workaround available from NVIDIA, requiring them to implement their own scaler as a shader program.

http://www.anandtech.com/show/6023/the-nextgen-macbook-pro-with-retina-display-review/7

 

[RussianSensation]

Yikes! Most bizarre thing about it is that there is more blur with TXAA than FXAA.

It's a double-negative. Not only is there more blur, but it changes the contrast so the bright colours get more washed out and blacks get less deep. The whole point of a high-resolution IPS monitor is to get more detailed textures and crisper colours/deeper blacks than you can get on a TN. With TXAA you are literally just saying you wasted $ on an IPS 2560x1440/1600 monitor. It looks straight up like a console game.

 

[SirPauly]

Based on static shots -- not much of an argument.

 

[cmdrdredd]

So you're saying that in motion the colors suddenly get more vibrant and the fog magically disappears? And all the details pop back up like nothing happened? No it doesn't

 

[SirPauly]

Imho,

No, there is a trade-off of a noticeable blur or softening of the image for an impressive anti-aliasing while moving. Secret World suffers from a lot of aliasing that ya can't see on a static image and is curbed, very much so with TXAA. The key is in motion and the strength of the feature; to only discuss the cons is not objective.

These are the cons to me: Proprietary, can't be forced on due to the need for game developer support, noticeable blur that may be distracting to gamers that demand sharp, how about adding a multi-sampling feature as well instead of just offering a TXAA setting?

Positives: Innovative, different choice, game needed another quality setting due to the sheer amount of aliasing, wonderful job of cleaning up the aliasing, specifically temporal noise, efficient based on the anti-aliasing offered to me.

The feature to me is not a replacement over-all, but an enhancement for someone to consider, may enjoy to improve immersion.

 

[cmdrdredd]

The other negative is the current performance hit received. Needs more work.

Maybe it can be cleaned up like FXAA injectors clean up the blur that FXAA leaves with post sharpening.

 

[SirPauly]

I would like to see a sharper setting for the gamers that demand more sharpness but also to leave it alone for that cinematic look while moving as well. Hopefully as this feature matures there may be flexibility for sharp and smooth.

 

[f1sherman]

Not possible if you want to get rid of temporal aliasing.

You need bigger than pixel-wide filter mask to get rid of it.

http://timothylottes.blogspot.co.uk/2011/10/sharpness-vs-temporal-aliasing.html

 

[SirPauly]

Timothy offered this:

Note at some point the LOD bias could also be applied to TXAA just like RGSSAA. This is something I'm looking to improve upon on later TXAA releases once I get the jitter to the point where I like it (then you'd have 2 shaded samples/pixel on still images too, so LOD bias starts to make more sense).

 

[f1sherman]

The question is how much of LOD bias added details will survive after TXAA gets applied.

Frankly I doubt he'll end up doing that. It's more of a hack, a brute force one, anyway.

 

[RussianSensation]

Imho,

No, there is a trade-off of a noticeable blur or softening of the image for an impressive anti-aliasing while moving. Secret World suffers from a lot of aliasing that ya can't see on a static image and is curbed, very much so with TXAA. The key is in motion and the strength of the feature; to only discuss the cons is not objective..

The cons are so massive that any benefit of anti-aliasing is immediately irrelevant imo. The game looks like a console game, from all your shots, and other posted screens, and the video in motion from NV's website. It's blur fest everywhere. If I wanted to spend $500 on a GPU and make games look like PS3, I'd get a PS3. If aliasing loses texture sharpness and details, and puts opaque like coating on colours, it's degrading image quality on the entire screen while only fixing aliasing. Worse yet, FXAA still aliases in TSW and doesn't blur textures as much. If TXAA improves in future games, great. I hate the washed out look of consoles games. It just sounds like you hate aliasing so much that you'd sacrifice deep blacks, vibrant colours and texture detail for that 1 feature. To each his own.

 

[SirPauly]

With all due respect, have the common decency to investigate in motion before one blankets the feature with irrelevant or use console gaming.

 

[MrK6]

With all due respect, have the common decency to investigate in motion before one blankets the feature with irrelevant or use console gaming.

The thing is we already have AA methods that do well both in motion AND in static elements (UI or 3D). TXAA is only bringing negatives to the table in this situation.

 

[SirPauly]

I don't disagree with that but offering a traditional super-sampled may of translated into totally unplayable at x2 or x4 in this specific title.

This is why I enjoyed this quote:

I am not saying this is traditional or conventional super-sampling.

Yeah TXAA is not conventional super-sampling. Certainly I'd rather just super-sample everything on a rotated grid, just do RGSSAA, but the cost of that is not practical in many cases. So TXAA does the next best thing, and uses samples across time. So in motion, when the temporal aliasing is bad with MSAA, you get some non-conventional form of SSAA with TXAA.

Note at some point the LOD bias could also be applied to TXAA just like RGSSAA. This is something I'm looking to improve upon on later TXAA releases once I get the jitter to the point where I like it (then you'd have 2 shaded samples/pixel on still images too, so LOD bias starts to make more sense).

 

[MrK6]

I don't disagree with that but offering a traditional super-sampled may of translated into totally unplayable at x2 or x4 in this specific title.

This is why I enjoyed this quote:

Why super sample? Just run a higher sampling of MSAA (8x or 16x) with transparency if applicable. If that's performance prohibited, upscale or don't run AA. IMO, blurring the heck out of an image and losing fidelity just so you don't have aliasing is counterintuitive; AA in its mission is an IQ enhancement.

 

[SirPauly]

Because of the temporal aliasing?

 

[f1sherman]

How do you like them apples

 

[ZeroRift]

I have to wonder: would it be possible to remove the wide tent pixel sampling, but keep the temporal components?

It seems like, if TXAA was ONLY a temporal AA solution, you could combine it with other AA methods (anything from FXAA to SGSSAA) to get a sharp(er) image AND cinema-like motion....