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

I switched from an AMD 5850 OC to an NV760 OC 4GB last year and was immediately shocked by the low quality SS with the NV card. The AMD looked much better with rF2 without any fiddling. Maybe nVidia SS just isn't that great?

 

Can somebody please tell me which track Spinelli is using in his screen shots?

 

Palm Beach

 

Thanks Ari

 

Palm beach, main straight facing backwards.

Chase cam is the default chase cam (just click it once to go to chase cam), and then not adjusted in any way except zoomed/moved all the way back (using the default chase-cam adjustment keys which are defaulted to the number pad keys).

I really hope you're right about this an there is an issue with MY PC and TS @ 0, because that amount of down-the-track-surface blurring is terrible, just atrocious.

The thing is, I've noticed this since the texture sharpening feature first became available. The TS option first became available quite a few rF2 builds, rF2 full uninstall-reinstalls, Nvdia driver builds, and full Nvidia driver uninstall-reinstalls ago. Not to mention I've had 3 GeForce 780 Tis over that time period. So I would be extremely shocked if the apparent issue is on my end.

1. Apparently TS @ 2 (+1) is too sharp on my PC compared to others and therefore showing too much track detail and aliasing, it should be more blurred and also therefore show less aliasing than it does on my PC.

2. Apparently TS @ 0 is way too blurred with regards to further-down-the-road track detail compared to everyone elses PCs, which would also mean that TS @ 0 should be showing more aliasing than it actually does on my PC since a sharper picture can introduce more aliasing.

Hope you're right TechAde.

The only issue I can see is that I use 16xCSAA instead of 8xQ (8xMSAA). Since CSAA does, both, a better job of antialiasing and offers better framerates than MSAA (this may be different when using supersampling, and most likely different when using sparsegrid supersampling whers you should use MSAA not CSAA [and also at the same sample amount for each] but I use triple-screens so no supersampling for me, just CSAA and transparency multisampling as i pointed out in this threads original post). Maybe the TS settings have some sort of issue with CSAA??.... Other than this, I just can't think of any other setting I have that may be affecting this, everything is pretty much standard stuff, heck, I even disabled all texture filtering optimizations ("Texture filtering - Trilinear optimization = Off, Texture filtering - Quality = High quality).

 

Just a quickie, which 16xCSAA are you using?

I see two different 16xCSAA options in Inspector:

16x [16x CSAA (4 color + 12 cv samples)]
and
16xQ [16x CSAA (8 color + 8 cv samples)]

 

Jesus, thanks for noticing this (need to update my first post). Since 8xQ is just 8xMSAA, I thought 16xQ was also 16xMSAA, I never even noticed the writing next to it saying it's a CSAA version. Anyways though, I'm just using 16xCSAA (4 colour + 12 coverage samples). The one on top of 8xQ, not below it (I'm pretty sure this corresponds to Level 5 if doing it from rF2, but I do it from Inspector anyways).

 

Thanks.

I've tried a few things whilst on lunch break and so far it's looking like the issue might not be with your installation as such, but Palm Beach itself.

Even at TS=1 (LOD bias of +2) at Palm Beach the trees in the distance don't get blurry.

Could you do me a favour and try e.g. Tigermoth at TS=0 and TS=1 and confirm that it works as expected (i.e. TS=1 is a lot more blurry than TS=0)?

 

THANK GOD, lol

Defintely, I'll try Tiger Moth right now.

P.S. If the TS at anything other than 5 (auto) is a "locked-in" global setting, then how can an individual track affect and/or override that??...

 

Not sure yet, one possibility is that the materials don't have any mip maps.

I'll work it out, might take a while though as I can only spend a few minutes here and there on it.

Sent from my SM-G900F using Tapatalk

 

Ok, TechAde, I have 2 new interesting results. One acts exactly how you describe it should, and one acts how the Palm Beach test acts.


1. Tiger Moth (also checked with TS @ 3 [-1] and 5 [auto]) - It works exactly how you describe it's supposed to work. TS @ 2 (+1) is blurrier everywhere, no "ifs", "ands", or "buts". TS @ 0 (off) is way better here, PERIOD.

TS @ 5 (auto) was almost identical to TS @ 0 (off). At first I thought they were indeed identical, but I had to double check my pics extremely closely, they are just so slightly different, but it's extremely difficult to notice in direct comparison pics, let alone while playing.

2. Melbourne by RaceRFactor (also checked with TS @ 3 [-1] and 5 [auto]) - It works as a hybrid of how you say it's supposed to work (like in the Tiger Moth test), and how it works in the Palm Beach tests. Let me explain....The background, like the stands/seats, objects, buildings, etc. work just like in the Tiger Moth tests - they are blurrier with TS @ 2 (+1) than with TS @ 0 (off), however, when it comes to the actual track surface then TS @ 2 (+1) is sharper than TS @ 0 (off), and there is also less blurriness and therefore more aliasing with TS @ 2 (+1). Just like Palm Beach, the further-down-the-road track surface textures just get absolutely destroyed with TS @ 0 (off) unlike with TS @ 2 (+1).

TS @ 5 (auto) is quite different, here, than TS @ 0 (off).


Conclusion: At Palm Beach it works exactly how I have been saying it does, at Tiger Moth it works exactly how you have been saying it does (and most likely the proper way), and at Melbourne (by RaceRFactor) it is a hybrid of both. Basically it's all track dependent. I have a feeling that something from the track makers/modders is probably overriding/ignoring/screwing with the rF2 player-file TS setting; I can't figure out anyother variable besides the track itself



P.S. I deleted all cbash and shader folder files for each TS setting this time; I didn't for my Palm Beach tests, but obviously it doesn't make a difference as my Melbourne test proves.

 

Very interesting, thanks. I've not had any more time to look at this and I doubt I will have today sorry.

 

While you are at it, TechAde, I think your suspicion that SS was not working on my rig is correct. Or, it may be that there is some trick to applying it. I decided to revert to using rF2 in-game AA and AF and set the NCP to application-controlled for both AA and AF. Works fine. Every level from None to 8 in rF2 gives me a different FPS result.

This is the tricky part. I thought that one could apply SS to an app in the NCP even when application controlled was selected for AA and AF. If this is not the case, you can stop me here!

Whether the AA Mode is set to "Application controlled" or "Enhance the application setting," no SS gets applied. My prior post also showed next to nothing happening when various SS levels were applied to settings when the "Override any application setting" mode is used, though SGSS gets applied when you do this via Inspector.

Is there some very precise combination of settings to get NCP to actually work with rF2, or how on Earth could my computer just ignore the SS settings when all else seems to respond normally? By the way, I am sure in-game AA with just a little help from some SS or SGSS would look great if I could figure-out how to get it to activate! No exotic settings for AA required.

Thanks in advance for your help. Sorry we are magnetically attracted as soon as someone who knows what they are talking about appears here

 

Marc, it's not just you, mine is doing the same with the Transparency Supersampling setting at 2x/4x/8x Supersampling.

I'm fairly sure those settings used to work so my hunch is some sort of driver bug.

Edit:

Try
Mode = Application Controlled
Setting = 1x2/2x1/2x2/3x3/4x4 Supersampling (D3D only)

I have a feeling the transparency supersampling mentioned in first paragraph is only working in OpenGL, not DirectX.

 

Driver as in only the latest version (344.65)? And your recommended SS above are only available in Inspector... I'll try and report back, though I was hoping to not use Inspector going forward! Those are also heavy-duty SS, so will like kill frame rates. Thanks!

 

OK, tested. As expected, when Application Controlled is selected, nothing happens. rF2 ignores the settings.

Switched to Enhance and yes the frame rate dropped like a stone with 4x4 on. However, the same flickering shadows and textures were still there (that solidify when you get close enough to them). So seems to activate, but adds nothing to IQ and completely kills performance.

 

Sorry, brain fart, I didn't mean app-controlled, I meant override!


Unfortunately I think it's going to be very tricky, if not impossible, to come up with a single set of settings that will work everywhere, due to the way different tracks have been modelled.

I'll know more once I've had chance to analyse what's going on at e.g. Palm Beach where attempting to globally force a zero bias doesn't work.

 

I think you'll find the railings along the pit lane at Portugal may also be impervious to any settings.

However, your tests proved that standard SS is not being applied under Enhance or Override. The 4x4 class of SS (whatever that's called) does apply under Enhance and Override, but doesn't do a heck of a lot!

Plain old Level 3 in rF2 with AFx16 looks pretty good aside from the flickering and very poorly aliased shadows and a few other areas like the railing in Portugal pits. So far, nothing I can force, including settings that reduce the frame rate by 90% (!!!!!!!) fixes it.

 

Also, I wanted to delete and re-create my NCP profile for rF2. Just in case and because for some odd reason it doesn't pick-up the icon, but shows as a generic Windows rectangle. The Remove option is greyed-out. I can find no documentation that tells me where the actual configuration files are stored in case I could manually delete it. There is no config file with the executable in the Program Files folder.

 

OK, I think I've found the problem at Palm Beach.

Take the treeline for example, that's one part of the scene that refuses to blur even when forcing a +2 bias globally.

The reason appears to be that the treeline material has been set up to use chroma transparency. This appears to kill mip mapping entirely, meaning it uses the full resolution texture map no matter how far away the trees are.

If I tweak the treeline material to use alpha blending instead of chroma it then 'listens' to the global bias setting and blurs as it should when a positive bias is used.

I think we can be confident that this is the same everywhere that refuses to accept a positive (or even neutral) bias.

Long story short is there's nothing we can do about it as users.

There's also a major drawback to track creators using alpha rather than chroma for everything, sorting issues. This is where textures render in-front of other textures, when they should be behind them.

Using chroma rather than alpha is the answer to the sorting issue, but as we can see if comes with it's own issues.