Marty McFly's Ambient Obscurance (MXAO) with IL

Oh well. Not sure if it's planed at some point but no rush, HDR could prove to be just another short-lived gimmick like stereoscopic 3D.

Helluw,

I have a problem with MXAO on Final Fantasy XIV in Directx 11. This is the error: pastebin.com/e1Nfy5i2 . This error only occurs in performance mode. No idea why.

The MXAO effect also seems to disappear and come back a second later for no apparent reason. I don't really know what to do with it. duckduckgo.com doesn't seem to hold any answers.

EDIT: It also disappears when in configuration mode.

Hello Marty,

I'm currently working on optimal settings for Warframe, just read the pdf for MXAO but ingame, MXAO is sometimes turning off. Like when I'm in my Ship, its alright, but when I join a match MXAO is starting to flicker, like turning on and off and sometimes its completly off. I couldnt find anything in the pdf or the shader file.

Do you know this problem and if so, is there a work around or do I need to turn MXAO off for this game?

Thanks a lot for your hard work

LandsHeer wrote: but when I join a match MXAO is starting to flicker, like turning on and off and sometimes its completly off

ReShade disables depth effects in multiplayer to prevent cheating. It does so depending on network activity, so the block may sometimes trigger for brief moments only and sometimes block the effect for longer times.

Good news for all MXAO IL fans who can't use it in regular game play due to its performance: I've found a really cheap way to get 75% of the visual quality with little additional fps cost. Instead of sampling occluder color and normal per offset, I use a heuristic based on AO data to determine which occluder color is most likely predominant and read it only once. For 16 samples, current system is like this:

16 AO samples + 16 color samples (+ 16 normal samples)

With the cheap approximation it's

16 AO samples + 1 color sample (+1 normal sample)

This means you can get faked IL almost for free with little visual loss compared to full blown IL.

Will post comparison pictures and more later this week.

Marty McFly wrote: Good news for all MXAO IL fans who can't use it in regular game play due to its performance: I've found a really cheap way to get 75% of the visual quality with little additional fps cost. Instead of sampling occluder color and normal per offset, I use a heuristic based on AO data to determine which occluder color is most likely predominant and read it only once. For 16 samples, current system is like this:

16 AO samples + 16 color samples (+ 16 normal samples)

With the cheap approximation it's

16 AO samples + 1 color sample (+1 normal sample)

This means you can get faked IL almost for free with little visual loss compared to full blown IL.

Will post comparison pictures and more later this week.

*eyes glaze over* I don't know what you said, but I want it!

Also Marty, is the latest version of MXAO in the Reshade 3 github repositories?

Yes, the github.com repo contains MXAO 2.0 which is newest.

About the IL, it's basically reducing the IL samples to 1 so you have 2 additional samples instead of tripling the sample count.

Superawesome. Thanks for working on performance I use IL in very few games. Look forward to the comparison shoots.

Here's the results. MXAO w/ IL and backface check enabled, 14 samples. One problem of this new hack is that its quality almost doesn't benefit from sample count so MXAO with 8 samples with and without hack looks almost same but there are worlds between MXAO 64 samples with and without hack. But I reckon only ppl with slower systems will use it so they'll use low sample counts anyways.

Regular IL:
abload.de/img/gta_sa2017-06-0118-102nu9h.png
IL hack:
abload.de/img/gta_sa2017-06-0118-10ecui7.png

There *is* some false coloring visible but it's minor.

But there's more! I managed to boost the regular IL performance in general as well, like a LOT (no hack, no visual difference to before), thanks to a tip Boris Vorontsov gave me.
MXAO with IL and 255 samples:

AO without IL: 132 fps
AO with IL + backface, stock MXAO 2.0: 72 fps
AO with modified IL (same visuals as MXAO 2.0): 114 fps
AO with IL hack: 125 fps

Marty McFly wrote: AO without IL: 132 fps
AO with IL + backface, stock MXAO 2.0: 72 fps
AO with modified IL (same visuals as MXAO 2.0): 114 fps
AO with IL hack: 125 fps

That's fantastic news since MXAO is one of my favorite shaders but even on a 1080Ti, the performance penalty can be severe, especially on higher resolutions like 4K or 5K.

Promising results. I think my GTX 960 is about 25% of your 1080. I mostly use between 8-32 samples. This will for sure make IL a more viable option.

I really look forward to get my hands on this. Love the whole shader concept that just matures and getting better. I just know how awesome this will look in the Hitman game

I think I'll post a preliminary release build here later today. Betatester with a much slower video card measured 50 vs 20 fps with 255 samples for AO only, AO and IL 45 vs 12. These are just general performance updates,, he didn't try out the IL hack. Of course the fps aren't viable but it's a slower card and a crazy amount of samples so with a moderate count (8 to 24/32) they'll be obviously so much higher. I was a bit surprised by this improvement because the fps differences on my end were marginal compared to this.

Zarathustra wrote: That's fantastic news since MXAO is one of my favorite shaders but even on a 1080Ti, the performance penalty can be severe, especially on higher resolutions like 4K or 5K.

Run 1440p with some mild AA. Save 4k for really old/simple games.
My OC'd 1070 gets basically no performance drop, even with non-optimized IL.

I see now if you use such big resolutions, you need some sort of downsampling functionality...well, that complicates things...anyways, testing time!

pastebin.com/j1s3NHFG

It doesn't yet contain the modified IL but it has some serious general code changes. Should run faster by a margin. Biggest difference is that the AO estimator uses different (simpler) math now so same sampling radius settings produce smaller AO radius, hence I raised the maximum AO raidus to 20.0 to match visuals somewhat of older range.

Marty McFly wrote: I see now if you use such big resolutions, you need some sort of downsampling functionality...well, that complicates things...anyways, testing time!

pastebin.com/j1s3NHFG

It doesn't yet contain the modified IL but it has some serious general code changes. Should run faster by a margin. Biggest difference is that the AO estimator uses different (simpler) math now so same sampling radius settings produce smaller AO radius, hence I raised the maximum AO raidus to 20.0 to match visuals somewhat of older range.

Hi, Marty. I'm trying to use the new code, but my game hangs when begin to load the shaders.

Ah I'm stupid.

pastebin.com/j1s3NHFG

Hi Marty...

First of all....thanks for this great tool.

I'm having some problem with Euro Truck Simulator 2.

1) ETS2 has options for internal rendering at higher resolutions (200% , 400%, etc), before scene is sent to screen (at 1920 x 1080 in my case). MXAO only works at 100% Not good because demmands strong anti-aliases (maybe reshade fault).

2) There is a anoying black-blinking rectangle on bottom left corner of screen. Only happens when in game HDR is enabled. (probably game engine fault).

.

peplegal wrote: Hi Marty...

First of all....thanks for this great tool.

I'm having some problem with Euro Truck Simulator 2.

1) ETS2 has options for internal rendering at higher resolutions (200% , 400%, etc), before scene is sent to screen (at 1920 x 1080 in my case). MXAO only works at 100% Not good because demmands strong anti-aliases (maybe reshade fault).

2) There is a anoying black-blinking rectangle on bottom left corner of screen. Only happens when in game HDR is enabled. (probably game engine fault).

.

It does not really work with internal downsampling. At least it has been so for me in some games.

Would advise to use Nvidia DSR or if you are using AMD, VSR. It works fine with all kinds of shaders. Though the drawback is many games doesn't scale the graphics interface. So you end up with tiny icons and text.

About that rectangle glitch. It just happens that sometimes games have really weird depth information. I have game that have weird box under my character. Other game there is a big rectangle around the mini map where there is no depth at all

Hm, I expected some feedback for the new version...

Checked the new version on STALKER SoC, looks good, works fine, but I can't see any performance improvements. I was using 20 samples on older version, tried to set a higher value, but fps dropped so I got back to 20 samples.

I have gtx1080 and I play this game with DSR turned on.