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

Absolutely outstanding detective work there TechAde. Thanks for taking the time to confirm that you and I hadn't lost our minds.

 

Thanks. It actually took me longer to compose the forum post than it did to do the detective work!

Sent from my SM-G900F using Tapatalk

 

Can someone tell me what the Level 1 to 8 AA-Settings in rF2 correspond to? Because i can use Level 8 with nearly no hit in Framerate compared to Level 4...

 

Maybe you have a problem, too? Join the crowd. Not sure if it will be helpful, but here are my results (only useful for the proportions) for a medium-heavy scene:

None - 120 FPS
LEVEL 1 - 112 FPS
LEVEL 2 - 116 FPS
LEVEL 3 - 109 FPS
LEVEL 4 - 107 FPS
LEVEL 5 - 106 FPS
LEVEL 6 - 93 FPS
LEVEL 7 - 86 FPS
LEVEL 8 - 84 FPS

This is with all else turned off in the NCP and no Inspector. Yes, LEVEL 2 was higher than LEVEL 1! Please post your results of a consistent replicable spot (e.g., in the pit garage with fixed weather and time ... I usually use Portugal), because my SS may not be working properly. Would be interesting to compare proportions.

 

Marc, I tried to replicate your findings regarding super sampling (SS not SGSS) not working as expected using Inspector. Bear in mind that my GFX card is only a 760.

METHODOLOGY

For each SS setting, the track was loaded a minimum of 4 times and FPS recorded while sitting on start grid. An average of those values is displayed in the test results. There was never a variance of more than 1 FPS within each data set.

NVIDIA INSPECTOR

1) Version: 1.9.7.3
2) "Antialiasing - Mode" set to "Override any application setting"

RACE SETTINGS

1) Track = Brianza 10k
2) Car = DW12
3) Grid position = 4 (locked)
4) AI = 14

IN-GAME GFX SETTINGS

1) Anti-aliasing = none
2) Texture Filtering = X16 Anistropic
3) Shadow Blur = Quality
4) Road Reflection = Low
5) All other settings on max

TEST #1: 4xMS [4x Multisampling]

2xSS: 45 fps
4xSS: 43 fps
8xSS: 43 fps

TEST #2: 16xQ [16xCSAA (8 color + 8 cy samples)]

2xSS: 43 fps
4xSS: 40 fps
8xSS: 36 fps

TEST #3: 32x [32xCSAA (8 color + 24 cy samples)]

2xSS: 43 fps
4xSS: 40 fps
8xSS: 36 fps

CONCLUSION

1) 8xSS should absolutely kill the NV 760, but it didn't.
2) Can't explain why test 1 result is 43 fps for 4xSS and 8xSS.
3) Can't explain why test 2 & 3 results are identical.
4) Results are inconclusive. Don't know whether this confirms Marc's findings or not, but is unexpected.

Whazzz Up?

 

Kind of OT:
On my config (GTX670) I never was able to choose Level 1 to Level 3. It always reset to 'no AA'. Only Level 4 stay in the config window of the launcher or in game. That's true from the first day I took rF2 (Build 69 or 89 can't remember). That's why I stopped using those choice and replace them in NV Control Panel. / End of OT

 

Ah YoLolo69 ... Another mystery

Like you, I always set AA to "none" in Config Mgr.

 

Could you report that to the ISI team?!
TIA

 

CONCLUSION: Something is out of whack on both of our systems. When I applied SS systematically it was using 32xCSAA. That's when I got effectively no difference between 2x and 8x. Your test at least showed some drop, though it seems that if you have 4xMS you are effectively not getting SS applied. This is all highly irritating. The settings themselves are complex and annoying enough to fiddle with when they work properly, but when they don't it is really unbearable.

At this point all I need is to get rid of the damned shadow flickering and edge-related temporal aliasing and I would be happy with the IQ on several of the built-in settings (LEVELS). All I should have to do is apply some SS and be good to go. But that apparently is too easy!

 

Well, if you want to see your card start to cry, try 16xQ with it's matching setting 4xSGSS. Your jaw will drop at the beauty of the image and at the low frame rate. Seriously--let us know what the FPS is for the same test as above.

 

Ok, i just did a short test, GTR sitting in the Pits in Portugal, 15 AI, Weather fixed.

rF2 AA Settings none to 8:
78
78
77
75
75
73
80
79
80

After that short series of test i assumed there must be something wrong... Then i let NVI set AA:

16xQ 75
32x 78
2x2 79

IQ was definatly improving, but Frames are still staying the same... Is NVI working at all? Let's try...

4x4 34

Ok, it works... Did some more short tests:

2x2+2sgss 77
4x4+8sgss 34
8xQ+8sgss 58 (this is by this short test the way best image quality that could just be used to race)
4x+4sgss 75 (what i use atm)

I think my resolution (temporary only 1280x1080) and my system are just about on the limits. But still, very strange scaling of all the settings i just tried...

 

Marc, at this point I'm inclined to believe that there may be a problem with the bitwise operations (NOT, AND, OR, XOR) for certain settings in Inspector itself. A coding issue unrelated to hardware or the nVidia driver (except for clamp). There is an option to "restore current profile to nVidia defaults", but I don't think we should have to do that between each test.

Yeh, you're right - this is "highly irritating". At least SGSS seems to work consistently and as expected.

 

whiplash - Yep, I'm currently using 4xMS and 4xSGSS on my 760. That's about all she wrote with the 760 - anything more and it seems to be beyond the capability of the card.

 

It sucks that we also get a way sharper road with TS @ 2 (+1) than with TS @ 0 (off) at other tracks as well, not just Palm Beach. I only tested one other track (Melbourne by RaceRFactor) and, bam! Same issue, the road gets destroyed at TS @ 0 (off).

Why do things only get messed up at TS @ 0 (off)? Why is this issue not happening with TS @ 2 (+1). If they're the same thing just one LOD apart (0 and +1), then why is this only happening on TS @ 0? Why isn't the same issue showing up at TS @ 1, 2, 3, or 4? It only happens at TS @ 0, so there must be something special about that one particular setting....

Instead of using the in-game "level" AA, would you be so kind as to use the AA from inspector so that we all know exactly what AA you're using (eg. 4xMSAA, 8xCSAA, 16xCSAA, etc.)?

I found CSAA easily gave the best framerates and IQ (as reported in my OP). My tests were when not using transparency supersampling though, ans I only enable transparency multisampling since I'm on triples. With regards to SGSSAA you're best with matching the regular MSAA sample amounts with the SGSSAA transparency supersampling amounts as that is how they work (for eg. 2xMSAA + 2xSGSSAA, 2xMSAA + 4xSGSSAA, etc.)

 

All good questions that I don't have answers for, yet.

There has to be a reason for it thought so I should be able to find it when I have time to dust off the detective hat.

Sent from my SM-G900F using Tapatalk

 

Thats exactly what i did, although i have not yet tested 32xCSAA in "real" conditions.

 

Honestly, this is depressing. There is no rhyme nor reason to it, which there isn't for some of my results, either, but some of yours are opposite!?!?!?!?!

However, be lucky you have IQ improvements with no loss of performance instead of the reverse situation that I have.

And like me, you have concluded that 8xQ with 2xSGSS or 4x with 4xSGSS is the best IQ and performance balance even though there should be a much "lighter" method to get a decent image. As I mentioned, rF2 built-in LEVEL 3 is actually perfectly acceptable save for the shadow and seams temporal aliasing issues. Arrrgggghhhh!

 

Marc, I gave the following a shot as you suggested using the same criteria as above:

TEST: 16xQ [16xCSAA (8 color + 8 cy samples)]

4xSGSS: 31 fps

But, it ain't all bad. Surprisingly, it's not a slide-show. In fact, I get no screen tearing and no microstutters. I have to play with it some more to determine if I'm getting input lag (more than likely I am).

GFX card runs at 100% and all three fans on the card rev up to 1950 rpm (50%). Temp never exceeds 122F. I do have an additional 120mm turbo fan about an inch from the card to provide ample air to the fans on the card though. Dead quiet (at least to my ears - I'm almost legally deaf).

As far as IQ goes, shadows seem better and so do fences - there's less shimmer and z fighting (compared to 4xMS). There's probably a bit more aliasing, but if there is, I don't see it while driving. Looks good to me, too.

Silverstone still has some minimal shimmering, but I don't think any setting can cure that track 100% without using blurring to get rid of it. I know you're like me and value image sharpness more.

One thing I did a few days ago in Inspector. I disabled:
1) Texture filtering - Anisotropic filter optimization
2) Texture filtering - Anisotropic sample optimization

Doing so seems to increase sharpness, but also increases shimmering slightly. Couldn't really detect any FPS difference so I might leave it that way for a bit.

 

OK, try 8xSGSS

I had already disabled the optimizations, though extra sharpness is not something I need. Just to get rid of the shimmering and flickering, which you may be right is worse with them disabled. But I am trying not to use Inspector if I can get away with it... If i could just get simple SS to work with an rF2 in-sim AA level, I think my problems would be solved.

 

"8xQ with 2xSGSS or 4x with 4xSGSS is the best IQ and performance balance"

I'm also still in agreement with that. I would prefer an in-game solution that does equally as well with IQ simply because FPS isn't optimal using the above.