IS NVIDIA Image Quality A Joke, Or Is It Just Me

Discussion in 'General Discussion' started by Spinelli, Mar 8, 2014.

  1. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    32
    Guys I have finally gone back to Nvidia since my first ever GPU over 10 years ago (FX 5600). I got 2 780Tis but just installed one for now. I have having some issues, and really frustrated and a little down as I spent a good amount of money on these cards.

    Ok what I want to do is 2 things...

    1) Most importantly I need help with Image quality. With AMD cards (HD 6870, 6950, 6970, 6990, 7950, 7970) in RFactor 1 and 2 I could make my Image quality ABSOLUTELY PERFECT with one simple step. Simply disable RFactor AA and set the control panel to override application settings and simply choose "AA TYPE = SUPERSAMPLING" and the amount of samples = 4X. Well I don't see any supersampling options in NVidia all I see is 2x, 4x, 8x CSAA, 8x, 16x CSAA, 16xQ CSAA, 32xCSAA. Also, there is a separate AA setting for transparent objects (trees, fences, foliage, etc) and only under there do I see multisample, 2x (supersample), 4x (supersample). 8x (supersample). I also tried Negative LOD Bias to "Clamp" as I remember that was very popular with RFactor 1 based games, but nope, still jaggies especially in motion.

    I do not want Supersampling on just the transparent objects, I want just regular, old-school, inefficient, FPS killing, ENTIRE SCENE, SUPERSAMPLING AA.

    I tried just maxing all my options out anyways and put regular AA to 32xCSAA and the transparent AA to 8x (supersample), but I still saw aliasing on posts, fences, lines on the ground. It wasn't bad but still very distracting because I am so used to a flawless full-scene supersampled image. Also I was getting over 400 FPS in RFactor, and with 8X proper true full-scene Supersampling I highly doubt I should be getting anywhere near to 400 FPS, even with a 780Ti.

    Please guys help me find good old, FPS murdering, full-scene supersampling as even at 4x with AMD cards it gave me an almost flawless image (heavy FPS cost though, couldn't use triple screens at over 100 FPS at 4x SSAA in RFactor 1, not even with an HD 7970 / R9 280X).

    and....

    2) I would like to make my driver settings exactly equivalent (or as equivalent as possible) to my AMD driver settings in order to do a ton of benchmark comparisons. There are so many options in NVidia what the? LOD clamp? Having 2 separate settings and amounts for regular AA and Transparency AA? Gamma AA? Cuda - GPUs setting??? Threaded Optimiations?????....


    Man talk about being bummed out after spending a bunch of money on the 2nd best video card in the world.
     
    Last edited by a moderator: Mar 8, 2014
  2. paarma

    paarma Registered

    Joined:
    Mar 20, 2012
    Messages:
    24
    Likes Received:
    0
    Download Nvidia Inspector for full control of the graphics settings.
     
  3. Ari Antero

    Ari Antero Registered

    Joined:
    Jul 27, 2012
    Messages:
    1,882
    Likes Received:
    829
  4. Prodigy

    Prodigy Registered

    Joined:
    Nov 2, 2012
    Messages:
    949
    Likes Received:
    64
    I am interested in this too, as I am considering to switch on nVidia and now I'm also using Supersampling on AMD for the nicest and clean image quality.

    So Spinelli, bring me some good news..
     
  5. Lazza

    Lazza Registered

    Joined:
    Oct 5, 2010
    Messages:
    12,346
    Likes Received:
    6,572
    If you are planning to have them both installed later, for use with rF2, I hope you read some of the threads regarding relatively poor performance (and smoothness) of nVidia SLI before you made your purchase.
     
  6. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    32
    Prodigy good news brotha, I found the best settings. I just spent 6 hours straight of non-stop video setting combinations using Nvidia inspector going back and forth trying every combination imaginable with it and RFactor 1....

    I found good old hardcore supersampling, the samples are different from AMD. Instead of 2x, 4x, 8x, they have 2x1, 1x2, 2x2, 3x3, 4x4. The bottom line is that the Nvidia supersampling doesn't work as beautiful as the AMD one. In RFactor 1 there is still some very distracing and annoying shimmering that drives me insane, especially in motion, even in 4x4 supersampling (which brought frames down to less than 120 FPS and that is in RF1 w/ 1 car only lol)....

    So, seeing that full-on supersampling apparently applies to the entire image and yet I still wasn't completely satisfied, I was thinking all hope was lost because nothing can be more complete than a full scene, no compromise supersampling. I looked up LOD clamp, adjusted those with a million different values, "allowed", "clamp", tons of different number combos, (including the advised ones on the link Ari sent, thanks by the way Ari). I tried messing with the transparency AA modes in combination with AA modes, nothing, everything was worse than full-on supersampling. I even tried the highly praised Sparse Grid Supersampling. I tried different amounts for that (2x, 4x, 8x) in combination with different amounts of AA (CSAA, multisample, everything). Things were starting to approach AMD's full full-on supersample quality but still not quite and still with tons of shimmer, I even tried 8X Sparse Grid with 32x CSAA and with 16x CSAA, still shimmering and broken jagged lines when in motion. I did some research and a guy wrote that when you use Sparse Grid Supersampling to use the regular AA mode with it, not CSAA or anything else, just regular multisampling, AND use the same amount for each. So, 2x Sparse Grid / 2x Multisample, 4x Spare Grid / 4x Mutisample, ETC., I tried this and BAM! IQ just about as "perfect" as AMD's full-on supersampling. GREAT!

    So basically to make a long story short, if you're looking for AMD's full-on supersampling image quality then do not use NVidia's supersampling AA type, use their regular Multisampling in combination with Sparse Grid Supersampling, and make sure they are both set at the same sample amounts. Like with AMD's supersampling, 4x is pretty close to 8x but with much less FPS hit.

    So with NVidia, for ultimate image quality in all RFactor / ISI based sims, it really comes down to 3 choices....2x Multisampling w/ 2x Sparse Grid Supersampling, 4x Multisampling w/ 4x Sparse Grid Supersampling, 8x multisampling w/ 8x Sparse Grid Supersampling.

    I even used 16xCSAA with 8x Sparse Grid, and even 32xCSAA with 8x Sparse Grid, but nope, as per the recommendation 8x Multisample still looked best when using 8X Sparse Grid.
     
    Last edited by a moderator: Aug 3, 2014
  7. peterchen

    peterchen Registered

    Joined:
    Oct 25, 2011
    Messages:
    2,099
    Likes Received:
    287
    Spinelli:
    Instead of the second 780ti you should get you a proper monitor with higher resolution if you have problems with aliasing!
    Can´t understand that people spend hundreds and thousands of dollars/Euros for cards and stuff
    but run with 1080 p...

    Said this rF2 has indeed some problems regarding aliasing, but it can be reduced to a non-
    noticable minimum.

    Greets
    Pete
     
  8. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    32
    1440P monitors don't have 120 Hz, they don't have 2D lightboost, they don't have 1ms response times, they don't have very low input lag (compared to the top end gaming monitors). Also, I have 3 monitors, 5760 x 1080 is a lot of pixels to push. There is a the Asus Swift monitor and a Phillips that are 1440P w/ 120 Hz but I don't think the Asus one can do 3D/2D lightboost (not sure about the Phillips), not to mention 7680 x 1440 is extremely demanding (even 5760 x 1080 is on new games), and the Asus one is around $700 for 1, nevermind 3.

    That is my future monitor goal though, to go to 3 27 inch 1440P monitors, as long as they are the Asus VG248QE / Benq XL2411T equivalent (1 ms pixel response, 3D/2D Lightboost, 120 Hz, very low tested input lag my VG248QE is reportedly around 3.5 ms)

    Yup, and I found the settings for Nvidia and AMD. Supersampling for AMD, or a combination of regular multisampling & Sparse Grid Supersampling (both at the same sample amount) for Nvidia :)
     
    Last edited by a moderator: Mar 8, 2014
  9. TechAde

    TechAde Registered

    Joined:
    Oct 13, 2010
    Messages:
    606
    Likes Received:
    38
    I'd suggest there are a couple of other options worth considering too: 16xS and 32xS. These are combined multisampling and supersampling modes available through nVidia Inspector.

    I find 32xS in particular is comparable to 8xMSAA + 8xSGSSAA but gives (much) better frame rates.

    One caveat with 16xS and 32xS is that the rF2 heat haze effect is buggy when they are in use. It's easy enough to disable the heat haze in the PLR with:

    Heat FX Fade Speed="0.00000" // Speed at which exhaust heat effects reduce by half (0 to completely disable)

    Another trick that helps with aliasing is to hit Alt-M to disable the track creator specified MipMap LOD Bias settings. You do need to make sure you're running 16xAF for this otherwise you'll get very blurry textures in the distance. This is particularly effective with 16xS and 32xS as these special nVidia modes do their own MipMop LOD Bias tweaking. I actually prefer the slight distance blur that this gives, I think it helps to create a realistic looking scene and minimises the 'cartoon' effect.

    It does rely on the track creators making sure they have well done MipMaps in their textures, so might not work well with all tracks.

    In theory setting Texture Filtering - Negative LOD Bias to Clamp should be the same as using Alt-M (i.e. disables any negative MipMap LOD Bias settings), however I find this doesn't seem to work for me lately.
     
  10. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    32
    Tried them all Techade, the method I mentioned was still the best, 2nd best was just pure full on supersampling (2x2, 3x3, 4x4).

    Yup the Negative LOD Bias stuff doesn't work for me either. After researching, I read people saying it stopped working for them as far back as 2 years ago. Are you saying that the alt-M method still works though, even though messing with the Negative LOD Bias directly in the control panel doesn't work???...If so I'll give that a shot, it might allow me to use something less demanding than 4x/4x Multisample/Sparse Grid (things can get heavy when using 5760x1080, multiview, high AA, & aiming for minimum of 90 FPS, even with a 780Ti or 2).
     
  11. peterchen

    peterchen Registered

    Joined:
    Oct 25, 2011
    Messages:
    2,099
    Likes Received:
    287
  12. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    32
    Those specs you pointed to aren't impressive at all.
     
  13. Murtaya

    Murtaya Registered

    Joined:
    Jan 25, 2012
    Messages:
    531
    Likes Received:
    0
    To peterchen, for me I cannot understand people trying to make a simulator with one flat screen in front of them no matter what the resolution! I run 720p so I can get 144+ fps with 3 screens. Feel and response beats graphical resolution every time.

    So Spinelli I don't need to let you have my first born then? Happy with the 780Ti?
     
  14. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,294
    Likes Received:
    36
  15. peterchen

    peterchen Registered

    Joined:
    Oct 25, 2011
    Messages:
    2,099
    Likes Received:
    287
    Mh, a matter of taste I think....
    I for one would prefer more res vs. 3 screens.
    But I surely understand the advantage of a 3 screen setup!
    Maybe when I should have it I don´t want to go back too. Don´t know...

    At Spinelli:

    What specs do you mean?
    If you refer to input lag, you should know that the measuring method and it´s numbers that came
    from that are all relative!
    So if you compare you have to compare with same method.

    Greets
    Pete
     
  16. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    32
    Haha yes Murtaya I am happy.

    Not 100% though. 2 issues (I have to live with one, but the other I should be able to solve :) ).

    Issue 1 - I don't know if it was a driver thing, the specific card (Asus HD 7970 Matrix Platinum) or what but with vsync off I NEVER had tearing. I know you probably won't believe me but in every single game, at EVERY & ANY FPS, there was no tearing, ever, ever, ever. 108 fps, 35, 135, 220, 85, 61, 119, didn't matter, not a trace, not even when turning real fast horizontally in a sim. It was as if there was no such thing as tearing. It was glorious. Now i'm back to the "try to get your fps locked to your monitors refresh rate to minimize, but still get a bit of, tearing" days, and it sucks. I was so spoilt with those 8 or 9 months with that "magical" card, but I guess I am back to reality now :(

    Issue 2 - I should be able to fix this one with some searching. I was playing with vsync on for some tests, previous to vsync on I was averaging 250 - 350 fps around the track (RFactor 1 with only 1 car), never an instant below 200 fps. Then I enabled vsync (120 fps/hz) and since I never had a single instant under 200 fps without vsync I was then expecting to understandably never drop below 120 fpsduring vsync, however at 2 or 3 points on the track I get a random big stutter and the frames drop to like 70 fps then a second or 2 later back to 120. Very big stutter, can definitely cause a crash. I found out that the stutters happened EXACTLY whenever that stupid dynamic boost clock would change the clocks. It didn't matter if it went up or down, even by a tiny 20 MHz, it would still stutter BAD and drop massive frames for a second or 2. So I need to figure out how to disable this dynamically variable boost clock crap and have a fixed/constant clock (and voltage).
     
  17. Robert Gödicke

    Robert Gödicke Registered

    Joined:
    Oct 5, 2010
    Messages:
    409
    Likes Received:
    11
    You sir, ARE MY HERO!
    I never heard of this key shortcut! Doing this finally fixes alpha textures for me! All white lines on the track surface, finally they are (more or less) anti-aliased! I don't mind the very little blur in the distance, the whole picture finally looks much better without all those ugly Moiré patterns! Once again, thank you!

    Here are some pics using Multisampling only to demonstrate the huge effect of this setting:

    Enabled vs. Disabled:
    [​IMG] [​IMG]

    Enabled vs. Disabled:
    [​IMG] [​IMG]


    And btw, it seems that rF1 has the very same issue for me. You don't know a solution for rF1 as well by chance? It's the same "over-sharpening" issue I have with my GTX770 which causes all those jaggies on textures.

    See here:
    http://isiforums.net/f/showthread.php/16536-Antialiasing-GTX285-vs-GTX770 (Look at the three pictures! Same issue!)

    Thank you, thank you, thank you!

    Greez,
    a very happy Rob :)
     
  18. TechAde

    TechAde Registered

    Joined:
    Oct 13, 2010
    Messages:
    606
    Likes Received:
    38
    One solution in rF1 would be to regenerate the relevant GMTs with all materials set to a mip bias of 0.0.

    For some reason track creators have been using negative mip bias for years, in order to cure texture blur in the distance, even though this creates major aliasing.

    When everything in the track is at 0.0 it also gives users the ability to override the bias using the GPU control panel. It also increases frame rate in a lot of situations.

    Sent from my GT-I9300 using Tapatalk
     
  19. Robert Gödicke

    Robert Gödicke Registered

    Joined:
    Oct 5, 2010
    Messages:
    409
    Likes Received:
    11
    So rF1 doesn't have an option to disable the custom set LOD settings? Damn.
    Since I upgraded my GFX card, the visuals in rF1 are just really bad. Too much Aliasing and Moiré patterns all around.
     
  20. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,294
    Likes Received:
    36
    I just tried 4xMSAA with and without neg lod bias to clamp but no visual difference between the two.

    I currently use 2xMSAA with 2xSGSSAA and -0.5 lod bias as it looks better than 4xMSAA with only a 10% fps loss. I would love to use 4xMSAA with 4xSGSSAA as it is really clean and sharp to me. But, it does look as if using clamp (or alt+m ingame which doesn't seem to make any different for me either) gives a similar result. Wish i could get it to work.
     

Share This Page