NVIDIA 344.48 WHQL GPU Drivers

I'm pretty sure LOD bias works regardless of rF2's texture sharpening setting, but only with SGSSAA enabled. If you don't use SGSSAA (most people don't), then I think the LOD adjustments are still "broken".

I need to do a bunch of testing tonight. I'm also going to do apples-to-apples (no LOD adjustments) comparisons of different MSAA and SGSSAA combinations (including each at different sample amounts).

Also, if you're using a real high quality form of AA like SGSSAA or DSR, then shouldn't people NOT be using texture sharpening @ "0"? All you're doing then is blurring-off details into the distance for almost no gain in reduced aliasing since DSR and SGSSAA should already be taking care of aliasing, no??...
 
I've just tried forcing LOD bias via Inspector and it looks like Spinelli is correct, the override only works when an SGSAA mode is selected. With MSAA or SSAA the LOD bias override doesn't work, even with texture sharpening at zero.
 
Yeh, I hadn't yet tried LOD Bias without SGSAA enabled. Thanks for taking the time to test it. Odd though, don't you think?
 
Thanks for confirming that TechAde. Yes, it's been known to be broken for years now on the past 3 or 4 generations of GPUs. It's even listed in the NVIDIA driver notes (PDF file they release with every new driver) as broken/disabled. There's been many complaints about this for years because players of non-rF2 ISI based sims heavily relied on the LOD adjustments to get rid of jaggies and fix broken and crawling lines without having to use supersampling forms of AA.
 
Spinelli, are you still going to test all the different modes in Inspector/game settings? I haven't messed with different settings in quite awhile and would be interested in your findings.
 
Yes Carlo, I got busy with school-work, I will have results in about 3-4 hours.

For my tests, should I leave texture sharpening at the default of 5? If the supersampling AA does a good job of removing jaggies then we shouldn't have to deal with the blurring - and even the full disappearance - of textures in the distance that texture sharpening @ 0 induces...

So, texture sharpening at 5, no LOD BIAS adjustments, comparison of different combinations of different amounts of MSAA (2x, 4x, and 8x) and different amounts of SGSSAA (2x, 4x, and 8x). Sounds good??...
 
School work is more important! Psst - how did you obtain "2x NVIDIA GTX 780 Ti" while in school? Buying food was a priority when I was working on my degrees.

For SGSAA tests, I think you should disable texture sharpening and set the appropriate LOD Bias in Inspector (or try different settings for comparison testing). For all others, leave it texture sharpening at the default of 5.

If you could rank results of IQ with various settings that would be good. I know results are subjective, but you might discover something that surprises us.
 
Ok, texture sharpening in the player file set to "4" (very sharp), and transparency AA set to "false" (off). So as to get the most aliased picture.

INDY
2xMSAA + 2xSGSSAA http://postimg.org/image/lf4punksf/full/
4xMSAA + 2xSGSSAAhttp://postimg.org/image/vdpoh4u7z/full/
8xMSAA + 2xSGSSAAhttp://postimg.org/image/5ptkug3cv/full/

2xMSAA + 4xSGSSAA http://postimg.org/image/uln0hxq0v/full/
4xMSAA + 4xSGSSAA http://postimg.org/image/qwxbigfzz/full/
8xMSAA + 4xSGSSAA http://postimg.org/image/hrtk2c1z3/full/

2xMSAA + 8xSGSSAA http://postimg.org/image/4oxxj2br3/full/
4xMSAA + 8xSGSSAA http://postimg.org/image/cwzv423nj/full/
8xMSAA + 8xSGSSAA http://postimg.org/image/ry25lbmdb/full/



SEBRING
2xMSAA + 2xSGSSAA http://postimg.org/image/ntwlf311b/full/
4xMSAA + 2xSGSSAA http://postimg.org/image/rl661eepr/full/
8xMSAA + 2xSGSSAA http://postimg.org/image/iau1xv3zz/full/

2xMSAA + 4xSGSSAA http://postimg.org/image/6zre9hx4v/full/
4xMSAA + 4xSGSSAA http://postimg.org/image/w3d278k5b/full/
8xMSAA + 4xSGSSAA http://postimg.org/image/oktwylssf/full/

2xMSAA + 8xSGSSAA http://postimg.org/image/9fxt134lb/full/
4xMSAA + 8xSGSSAA http://postimg.org/image/sq9ey9fy7/full/
8xMSAA + 8xSGSSAA http://postimg.org/image/o2yrir37j/full/


I suggest right clicking each link to open in a new tab so you can switch between different ones to get the best comparison. Also, make sure the pics are full sized, some browsers shrink the images slightly to fully fit your screen.


CONCLUSION: I think we're both sort of correct...

8xMSAA looks best - regardless of it matching or not matching the SGSSAA sample amount - on certain parts of Sebring. Look at the walls/barriers especially the white barrier almost under the bridge (just a bit past it) 8xMSAA always looks better there for any given amount of SGSSAA, even when non-matching.

However, look at all the Powerlines at Sebring, look at the white track-boundary line on the ground at Sebring, and look at pretty much the entire scene at Indy - the matching MSAA and SGSSAA looks better in all three of those cases than 8xMSAA + a non-matching SGSSAA does. You can see the powerlines at Sebring look better at 2xMSAA + 2xSGSSAA than at 8xMSAA + 2xSGSSAA, and again better at 4xMSAA + 4xSGSSAA than at 8xMSAA + 4xSGSSAA. Look at the white line on the ground at Sebring (the track-limits line before the bridge) - at 4xMSAA + 4xSGSSAA it is nice and smooth, at a non-matching 8xMSAA + 4xSGSSAA it is all broken and split apart. Look at the fences and beams at Indy - again you can see that the matching MSAA + SGSSAA easily gives the best result for a given SGSSAA amount, rather than 8xMSAA + 2/4xSGSSAA. It's not even a contest.

So, 8xMSAA looks better on certain things regardless of it matching the SGSSAA sample amount or not, but a matching MSAA and SGSSAA sample amount (even at just 2x or 4x) will look better than 8x non-matching MSAA on many other objects.

Overall, at least in these 2 scenes, I would easily give the winner of IQ to a matching MSAA + SGSSAA sample amount, although you are correct in that there are some instances/objects where 8xMSAA + 2x or 4x SGSSAA does beat a matching 2x or 4x MSAA + SGSSAA.
 
Last edited by a moderator:
Gee Spinelli, I stopped looking when I saw 164fps!

Seriously, thanks for taking the time to post your results. I've just spent more than an hour scrutinizing the images. At Sebring, clearly there is a difference looking at the power lines. On the other hand, looking at the clarity of the text of the sign on the right, the leaves on the trees, and the smoothing of roof lines in the distance vary as well. None of examples did particularly well with the chain link fence on the right.

I can't say one way or another whether the test is valid, but methodology seems logical since the same baseline was used. I'm also not sure if setting the correct LOD bias would change the outcome in any way. It isn't clear to me how "Texture Sharpening" does its' thing. If "clamp" worked on nVidia cards, then only those parts of the scene where the track creator used a negative LOD bias would be impacted. I don't know if "Texture Sharpening" works in a similar fashion.

If "Texture Sharpening" is set to zero, then the correct LOD bias for the various SGSS modes can be applied through Inspector (hence the reason why I'm currently using "-0.5" with 2xSGSS). As an aside, I've come across differing opinions on whether or not recent versions of Inspector take care of LOD bias settings automatically when SGSS is used. One thing is certain, even your GFX setup gets a massive hit with 8xSGSS and would obviously be impossible with mine.

As a result of your tests, I am going to try some different settings now. I became tired of experimenting with each new ISI build/nVidia driver revision and thus stopped awhile back. Never had to do this when I was using AMD cards with rF2 and I still think AMD provides a better image without having to go through all this tinkering (sigh).
 
I have no idea what the technical definition of 8xQ is, but it performs completely differently for me than 8xMSAA. 8xMSAA looks worse and 8xQ plays nicely with any level of SGSS, with very obvious quality improvements offset by performance declines. I am going to try 16xQ just for fun, although I already know FPS is too low to actually use it.
 
8xQ IS 8xMSAA, look at NVIDIA Inspector, it outlines it all for you there. Regular 8xAA in the NVIDIA control panel is actually 8xCSAA (8xCSAA uses 4 colour + 4 cv samples), while 8xQ is pure, full-on 8xMSAA.

I did an apples to apples comparo. No touching LOD stuff, in-game texture sharpening to nice and sharp. It's pretty clear that the matching MSAA and SGSSAA amount (wether it be 2x, 4x, or 8x) is the easy winner in the entire Indy scene, and on most - but not all - of the Sebring scene.

Make sure your browser isn't slightly resizing the images.

Other than certain areas of the Sebring track (mostly the walls/barriers), there are hardly, if at all, any other areas where a non-matching MSAA + SGSSAA looks better than a matching MSAA + SGSSAA at a given SGSSAA amount.
 
Last edited by a moderator:
I use only 8x AA in NV control panel and Texture Sharpening=0 in PLR, Is it possible to improve the graphics without losing fps with NV inspector ?
 
You must log in or register to reply here.
Back
Top