CSAA, Transparency AA (Multisampling), & Texture Sharpening (MIP LOD Bias) (w/pics)

CSAA is a win-win over MSAA

Using CSAA instead of MSAA (Nvidia) seems to give me a less aliased picture + even better framerates. If you're using any supersampling AA modes (eg. 2x2, 12xS, etc.) or supersampling transparency aa modes (eg. sparse grid supersampling) then this may not apply to you, however if you're just using regular MSAA without supersampling then give CSAA a shot. 16xCSAA (known as "16x [16x CSAA (4 color + 12 cv samples)]" in Nvidia Inspector) gives me a cleaner picture (less aliasing AKA "jaggies") AND slightly better framerates than 8xMSAA (8xQ). If you like to set your AA from rF2 itself rather than the GPU control panel, then I'm pretty sure 16xCSAA is Level 5 for most users, but don't quote me on that. It really is a win-win situation . If people have the same results as me, then they should definitely be going with 16xCSAA instead of 8xMSAA (8xQ).

If you're right on the limit of acceptable framerates and are therefore currently using 4xMSAA, then give 8xCSAA a shot. If it's the same case as above (16xCSAA VS 8xMSAA), then 8xCSAA should give a slightly nicer picture while offering - at worse - the same framerates (if not, even higher framerates.) than 4xMSAA.

[HR][/HR]
Enable GPU control panel Transparency AA, even if you're already using "Transparency AA" in the rF2 player file

Enable Transparency Multisampling in Nvidia Inspector or, I believe, the Nvidia control panel as well. Transparency Supersampling is obviously superior, but I'm talking about low demand, non-supersampling, basic AA modes here. The Transparency Multisampling pretty much does the same thing as "Transparency AA" in the rF2 player.json file, however, the NVidia version does a better job, having said that, you can enable both for an even teeny, tiny further improvement because it doesn't seem to effect framerates, so I leave both on. However, if you were going to just pick one, then enable the transparency multisampling from the NVidia control panel (or Nvidia Inspector) and set it to "false" in your player.json file (again, leaving both enabled gives a further teeny, tiny, IQ improvement over having just the NVIDIA one enabled so you can just enable them both).

Note: Even though I have Transparency Multisampling enabled through Nvidia Inspector, when I look at my rF2 profile settings in the official Nvidia control panel, it DOES NOT say that I have transparency multisampling enabled, but rather still on the default "global" setting. This is odd as most other changes in NVidia Inspector get reflected in the official NVidia control panel since they should be "tied" to the exact same thing in the "back-end". However, rest assured that enabling Transparency Multisampling from Inspector definitely, 100% works, and that is where I enabled it for all my comparison tests.

NOTE 2: To have even smoother foliage and stuff, you can even enable FXAA in the rF2 graphics options, however I don't like post-processing AA because it blurs textures, technically adds input-lag (since the processing happens after the image is already done being rendered and ready to be sent to the monitor - the FXAA processing is an extra step; in FXAA's defense, though, the input lag is so tiny that it makes no difference, probably not even to hardcore simracing-drivers), and finally because FXAA dropped my framerates massively (odd, since FXAA's framerate hit is usually less than 5%). My framerate was at around 90 fps with 16xCSAA; when I enabled FXAA - with 16xCSAA still enabled - my framerate dropped to around 70 fps.

[HR][/HR]
RFactor 2 player file "Texture Sharpening" results

THIS IS ABOUT OVERALL IMAGE QUALITY OF YOU'RE ENTIRE GRAPHICS, THE ENTIRE SCENE IN GENERAL, NOT JUST ALIASING/JAGGIES. THERE IS A LOT MORE GOING ON HERE THAN JUST DIFFERENCES IN ALIASING

This is about the following setting in your player.json file:

"Texture Sharpening":x,
"Texture Sharpening#":"Sharpen textures using MIP LOD bias - 0=Off, 1=+2.0(very blurry), 2=+1.0(blurry), 3=-1.0(sharp), 4=-2.0(very sharp), 5=0.0(auto)",

FOCUS ON THE CENTRE OF THE IMAGE WITH THE CORVETTE AT THE CENTRE OF YOUR SCREENl. You want to be looking at the following:
- track pavement, grid-lines, and curb, as the track nears the end of the straight
- fence alongside the left of the car
- grass and foliage alongside the left and right of the car
- track pavement behind the car
- trees far-off, way in the distance


Texture Sharpening: 4
MIP LOD bias = -2.0 (very sharp)
external
cockpit


Texture Sharpening: 3
MIP LOD bias = -1.0 (sharp)
external
cockpit


Texture Sharpening: 5
MIP LOD bias = 0.0 (auto)
external
cockpit


Texture Sharpening: 2
MIP LOD bias = +1.0 (blurry)
external
cockpit


Texture Sharpening: 1
MIP LOD bias = +2.0 (very blurry)
external
cockpit


Texture Sharpening: 0
MIP LOD bias = Off
external
external B (the track gets even worse here the further down the straight)
external C (and worse, still)
cockpit


Observations and thoughts


TS @ "4" (MIP LOD bias = -2.0 [very sharp]) and TS @ "3" MIP LOD bias = -1.0 [sharp])

- texture sharpening (TS) at "4", and "3", are the only ones that have a really nice road surface, the other TS settings all make the ground look like a blurred Nintendo 64 game ("5" is slightly better than "2", "1", and "0", though, which all look terribly low resolution)
- TS @ "4" and "3" have really nice and detailed white painted lines. You can see all the variation, cracks, imperfections and general detail that the track creators spend so much time and effort making
- there is just way too much aliasing with TS at "4" and "3", so "4" and "3" are effectively out of consideration for me. This may be different if using fancy, demanding forms of AA like supersampling, sparse grid supersampling, dynamic super resolution, etc. but I don't use those intense kinds of AA modes (with triple screens) so I don't know.


TS @ "5" - MIP LOD bias = 0.0 (auto)

- again, has slightly less foliage (as if it's just not rendered at all and completely missing) than TS @ "3" and "4", however it still has much more foliage than TS @ "0", "1", and "2" which all utterly destroy the amount of trackside foliage. Seriously, look at how much foliage is just completely removed from the scene when using TS @ "0", "1", and "2" - compare them to TS at "3" and "4".
- other than this, TS @ "5" isn't too bad, but it still has too much aliasing, especially when in motion. So you loose the nice sharpness and detail of TS @ "3" and "4" but with not much gain in aliasing-removal, which, to me, means TS @ 5 is also out of the question.


TS @ "2" - MIP LOD bias = +1.0 (blurry)

- - better foliage in terms of sharpness and the amount of foliage, it looks equal to TS @ "0" (however it still looks sparse compared to TS @ "4", "3", and even "5"
- less overall blurring - than TS @ "1" - of near objects like the general flat-grass textures lining the track, the lighter patch of grass on the left (seems about even with TS @ "0")
- slightly better detail and sharpness - of the track, grid lines, and the curb on the left, as you look further down the straight - than TS @ "1", and MUCH better than TS @ "0"
- brings back some sharpness and detail to the white painted lines going down the track that TS @ "0" kills
- fixes trees far off into the distance that TS @ "1" destroys
- All seems good doesn't it? It seems to be a decent compromise everywhere The thing is, TS @ "2" does introduce more aliasing but it's nothing like the aliasing in some of the other settings. TS @ "2" just may be my new preferred choice


TS @ "1" - MIP LOD bias = +2.0 (very blurry)

- fixes some of the further-down-the-road graphics that TS @ "0" completely destroys
- TS @ "1" kills all the foliage. It literally looks like some of the foliage itself was not just blurred, but completely missing from the graphics - completely disappeared. The other TS modes (other than "3" and "4") also do this, but TS @ "1" does it the most.
- some of the nearer track surface gets more blurred with TS @ "1" than with TS @ "0". Look at all the flat grass textures and the lighter, small patch of grass on the left - looks like a smudged game from 2002.
- brings back some sharpness and detail to the white painted lines going down the track that TS @ "0" kills
- absolutely destroys all the far-off trees in the distance, they look like someone smudged Vaseline all over my monitor
- basically TS @ "1" gets rid of the abomination that TS @ "0" does to the track further down as you look to the end of the straight, but it seems most other things become even more blurred with TS @ "1" which, to me, means TS @ "1" is out of the question - just way too much low IQ and blurring. Looks like a game from 2002.


TS @ "0" - MIP LOD bias = Off

- looks like there is literally no fence on the left side as you get to around the 8th or so "section". It literally looks like there is no fence at all modelled, just two beams with nothing in between.
- Foliage looks lessened, like some is missing, but it's not as bad as how much foliage is missing with TS @ "1"
- kills the detail and sharpness of the two white painted lines going down the track
- absolutely destroys the track as you look further down, in-fact it looks so bad that it looks like anisotropic filtering is completely disabled (it's set to 16x for these tests). I mean it's beyond bad, it's utterly atrocious and looks like the game is broken or like the track-maker took a big sh*t on the end of the track.



Overall winner for me - at least according to these pics - is TS @ "2"


UPDATE WITH DIRECT 1:1 AND HIGHLIGHTED COMPARISONS - CLICK HERE

 

Good stuff. It would really help if all the Texture Sharpening screenshots were taken at _exactly_ the same angle, any differences can play havoc when trying to compare aliasing. I'm going to try this myself when I have the time, but use e.g. a race session start from the grid so I can be sure the camera is at exactly the same place every time.

 

Focus on the centre of the screen, with the car in the middle of the screen. You want to be looking at the fence and grass alongside the left of the car, the grass along right of the car, the track pavement behind the car, and especially the track pavement, grid-lines, and curb, as it nears the end of the straight

If you full-size the screen shots and have them all open in separate tabs and line the parts you're comparing, then the slight in-game positioning differences are very minor and you should easily be able to identify any differences.

 

Very helpful for those who can't or don't want to use SGSS, which produces the absolute best quality. I am just experimenting (again!) after months of not fiddling because I think some combos are good enough I will no longer use SGSS. The difference is so small at high resolutions that it's not worth the FPS hit (to me).

One good aspect is to compare NVIDIA settings with the built-in ones (and turning everything off in Inspector/control panel and setting Application controlled).

part 2...

 

Part 2

Very timely as I was just fiddling with all of this yesterday evening. I have been using SGSS for years now and one annoyance that seemed to be caused by it was the heat haze bubble that would appear only in the rear view mirror. I know some mixed modes cause the heat haze bubble in the main view as well. The point of this story is that while experimenting I tried turning everything to App controlled and Off in Inspector and used only the rF2 settings, which I could easily see were working from the dramatic differences in FPS and visuals. BUT THE HEAT HAZE BUBBLES WERE STILL THERE IN THE MIRROR!

This I think confirms something frustrating I have experienced over the years, which is one of two possibilities:

1) Inspector app-based settings don't fully cede control when directed, nor do they fully take over control when directed...there are background/default profile settings still operating.

2) Sometimes you have to reboot for certain dramatic changes to occur.

I eliminated #2 because nothing changed after a cold restart. Ran out of time, but I now know somewhere in the maze of settings, something is not following the surface directions. More fiddling later.

And, *&^%$, this is annoying and frustrating. Just thought I would add that comment!

 

I know all that and stand by what I said, if trying to compare aliasing you need to ensure the shots are all at exactly the same angle otherwise you're not comparing apples with apples.

 

TechAde, I'm currently uploading pics from a replay from an identical spot.

Also, it's not only about aliasing, it's about all image quality differences as I pointed out in my revised original post.

 

+1

Slightly out and you can get misleading results just because of the wrong angle from what should be the slightly better AA solution.

 

Sure, but the fact remains that it's far easier and more accurate to compare two shots that are otherwise exactly the same, no matter what differences you're looking for.

Sent from my SM-G900F using Tapatalk

 

Hi guys, I'm still trying all different settings for RF2 and I'm having problems with the outer tyre edges in the ASR F1 92 mod. It's only this mod I have issues with so it may be a mod problem. Also, have you tried DSR in Nvidia cp yet? What do you think to it. It does affect performance but makes a slight difference to the quality. Cheers

 

I'm looking forward to seeing how well MFAA performs....that is if nvidia ever get round to releasing it.

 

At the moment I use 16csaa and 2x sparsing grid in NvInspector. And Texture Sharpe: 0, The white line of track is better than "2" for me.

 

Maybe you're not understanding. The pics are all from a replay, from the exact same spot, camera, angle, etc. THEY ARE EXACTLY THE SAME, lol. For you to even say this makes it sound like you haven't even looked at the pics.

If you tried comparing each of the "external" and "cockpit" pics to one another respectively , then you should be able to clearly see that they're identical. Having said that, a camera view that is just slightly off isn't going to change anything regarding a picture that looks like it's grass has been blurred like vaseline while the other picture has grass that is much sharper. Are your textures and overall image constantly changing quality and going all wack just from moving your camera around a couple degrees? They shluldn't be and no game's does. Your graphics don't change that drastically just from moving your camera 2 degrees, except of course, as you and Dr.R pointed out, possibly with regards to aliasing. However, if you read my "observations and thoughts" or just look at the pictures yourself, you should clearly be able to see that this is about A LOT more than just aliasing.

In my pics (look at "external" pic) you can see that the white lines lining and going down the track (centre of image with the Corvette in the centre of your monitor) loose even more detail with TS @ 0 than TS @ 2. I wonder why it would be better for you the other way around? Maybe something to do with SGSSAA?

 

You've added more pics since I first looked at them, I'm not misunderstanding anything thank you.

Positive bias should be more blurry in the distance then zero bias, if that's not what happens then there's something odd going on.

Sent from my SM-G900F using Tapatalk

 

I added the new, identical pics around 2 or 3 hours before that response of yours and Dr.R's

 

That doesn't invalidate anything I said, does it? Your original pics were at different angles, all I've ever said is that's no good for comparison. That's true now and will always be true.

You obviously agree so I don't know why this stream of conversation is still going on.

Sent from my SM-G900F using Tapatalk

 

Man I wish we could get the sharpness of the nearby track surface of TS @ 4, even 3, while the rest of the entire image quality of even just TS @ 2 would do for me, but with the aliasing (or lack there-of) of TS @ 0...

If I can live with the aliasing of TS @ 2 (judging from these pics, it doesn't seem too bad actually), then TS @ 2 seems to be the overall best-compromise winner. The aliasing of TS @ 2 is the big question though, other than that, it either equals or beats TS @ 0 in every other IQ department. All the other modes (TS @ 3, 4, 5), though, just hopelessly have way too much aliasing, or just destroy and overly blur way too many textures (TS @ 1).

I gotta test TS @ 2 during actual gameplay while in motion. It has the potential - at least according to these pics - to be the overall winner.

 

Like I said earlier if a +1.0 bias (TS=2) is less blurry than 0.0 bias (TS=0) then there is something odd going on.

A positive bias should be more blurry than zero whilst a negative bias should be sharper than zero.

Sent from my SM-G900F using Tapatalk

 

Spin,

You sure that disappearing fence, etc. is not down to anything other ?

Like 3 screens dual cards or settings, memory ?

Because I got to say I never seen anything like that, any fence any track and I actually use O now.

I run 1920@60Hz as Ari tells me I have a blurry low res high latency screen tearing setup anyways. lol

Pavement like Italy pits and road texture quality is never going to bother me, I want to do some nice closeup screens I can turn it on.

From F3 cockpit what road I can see is too far ahead to have good detail.

Everything else signs, posts, fences and trees are very good.

Any LOD level puts these tiny tiny little sawteeth on top edge of my Eve cockpit glass too. lol I won't stand for that.
Also adds shimmer to all suspension arms any sharpening.


What =O does do well imho is give the best motion image, there is no speckles or shimmers just nothing and no jags.

Running down the hill at Belgium if I look to Eau in distance with any sharpening at all the haybales along the wall and other stuff looks really weird , like you can't make out individual bales they get a sort of shimmer to them ?

With =0 they look square and precise and you can count them.

Anything a tad blurry speed will quickly sort out and I do not get the little speckles and shimmers.


So I do agree with you Spin, okay , track texture , pavements and a few others look below par with= 0
But stuff like the buildings , posters, signs etc all look fine to me.

All I saying static screenshots don't tell all the story for mine.


P.S.

TS @ "0" - MIP LOD bias = Off @ 4K

http://postimg.org/image/4zzny7f7v/full/

If the track textures had any more sharpness from that distance they would look faker imo.

But say if you took a shot from that same camera posi but zoomed right in down on 1 or 2 cars then I would need the good LOD

You also notice difference in quality of addons with LOD alot, esp. tracks, look at the skins, look at the car numbers you can see, any stick out better ?


BTW Better LOD would be good maybe high res textures with 2 installs based on PC grunt.


Fix the holes ......in landscape please I rather have lol .

All the Historics are pretty hideous to do a "fly by wire" run full tracks . ughh

and that mountain at Belgium looks like the one from "Close encounters of the third kind" p springs up out of nowhere. lol hehehe

I give up any LOD improvement to get rid of that. loool

 

OK, so how does one get rid of NVIDIA Inspector?

I have turned everything off and to application-controlled in Inspector, using in-game AA only, and I still have the heat haze bubble in the rear view mirror. I think that is caused by using mixed sampling (at least that's what causes it on the main view), so somehow there is still some mixed sampling existing on my system? I used SGSS for a long time, which I thought was the cause of the heat haze in mirror issue.