AMD vs. Nvidia filtering IQ

Discussion in 'Videocards - AMD Radeon' started by yasamoka, Mar 9, 2013.

  1. yasamoka

    yasamoka Ancient Guru

    Messages:
    4,336
    Likes Received:
    4
    GPU:
    EVGA GTX 1080Ti SC
    I have opened this topic after Memorian requested 7xxx series cards users provide him with a 3Dcenter screenshot.

    3Dcenter link: http://www.3dcenter.org/download/3dcenter-filter-tester

    I'm interested in comparing the filtering IQ in AMD 7xxx, 6xxx, 5xxx, 4xxx cards, and on the Nvidia side, 2xx, 4xx, 5xx, 6xx, and Titan cards.

    Please mention the drivers you are using. Some filtering problems may have been / may be fixed by drivers.

    Settings used are:
    Display Mode: Split
    ALU Filtering Mode: Perfect
    Max Anisotropy: 512x
    LOD Bias: 0.0
    Texture Scale: 1.0x
    Texture Movement X: 0.0
    Texture Movement Y: 0.0

    You can use File - Screenshot. Make sure you upload lossless PNGs and do not convert them to lossy JPG.

    One screenshot of normal mode, and one screenshot of tunnel. View - Show Tunnel to see the tunnel.

    If these are suitable settings, I hope everyone can use them in at least one test run.

    Please use spoilers, as the images are rather large.

    I suppose Memorian used these settings as well in his screenshot. Can you please confirm, Memorian? Thank you.

    This is Memorian's screenshot off of his GTX670 FTW:
    [​IMG]

    This is my R7970:
    [​IMG]

    For a straight comparison, save both pics to your PC, and shift between the two. You'll see differences. At this stage, I won't say which looks better. I need confirmation of the settings used by Memorian, and I need further data from the same card, and from different, older cards, Nvidia and AMD.

    Hopefully, I'll add a poll when we have something solid, eventually.

    Cheers.
     
    Last edited: Mar 10, 2013
  2. eclap

    eclap Banned

    Messages:
    31,511
    Likes Received:
    2
    GPU:
    Palit GR 1080 2000/11000
    the hd 7970 is the clear winner here. just place your mouse cursor to the left of center, where the iq starts deteriorating on the gtx 670 pic, then scroll down and you'll see the hd 7970 is still fine at that point. No contest, really.
     
  3. Fox2232

    Fox2232 Ancient Guru

    Messages:
    5,583
    Likes Received:
    5
    GPU:
    Fury X - XL2420T(Z)@144Hz
    Image rendered by GTX670 is closer to specification, but that HD7970 feels better from subjective point of view.

    And to add one thing here, comparison of static images does not matter that much. What matters is shimmering seen once you move texture along axis.
    At center where you see gray pixels should really be gray pixels because it's weight average of black and white squares at size smaller than 1 pixel.

    Further you go off center higher chance black/white square will cover whole pixel and then it should be black/white.
    If you see too black or too white pixel close to center then it was incorrectly calculated. (or filter type nearest_mipmap_nearest/linear_mipmap_nearest)

    There is thing to think about. And that is how far away from center is 1st pixel which can (therefore should) be covered by whole black or white square and if such pixel is totally white or black once it is fully covered.

    Anything from this breakpoint in direction to center should be closer and closer to gray.

    If you go from this breakpoint to the edge of screen some pixels should be black, some white and some which are sitting on edge should be gray.

    Once you load some texture you may notice real level of shimmering.
    It's much more noticeable on both nV/AMD compared to reference 512x aniso because it's kind of unfair since both cards limit themselves to 16x.
    I think that with 4k2k TVs we should bump aniso to 32x.
     
    Last edited: Mar 10, 2013
  4. yasamoka

    yasamoka Ancient Guru

    Messages:
    4,336
    Likes Received:
    4
    GPU:
    EVGA GTX 1080Ti SC
    One of the reasons I opened this thread was because I was unable to make anything objective out of this. Subjectively, yes the 7970 results do look better, but what specification are we talking about?

    If subjectively, straying away from specifications looks better, then what's the importance of sticking to spec? What disadvantages come from straying out of spec? And what about tweaks AMD, or AMD & Nvidia has / have done to improve / reduce the performance overhead of filtering?

    I wish you could explain your point further. Thank you.
     

  5. Fox2232

    Fox2232 Ancient Guru

    Messages:
    5,583
    Likes Received:
    5
    GPU:
    Fury X - XL2420T(Z)@144Hz
    Then there is another way to look at what is displayed to you. And if you use that as base you realize How good/bad LCD you have in front of your face :)

    If you look at black/white checkerboard which sits exactly on pixels then you will from close range see distinct black and white pixels, but once you look on such matrix from lets say 3 meters, screen will appear gray.

    Now you have this tunnel and at edges squares are big and from 3 meters there is no problem to distinct if they are black/white. But those closer to center will cover smaller area of your eye nervous tissue and will become grey.

    And if everything was calculated correctly from that threshold to the center everything should be exactly same gray.
    If it's not then it is incorrectly calculated or you have incorrectly calibrated lcd.

    Well I took liberty of measuring lightness value of each zone and it is indeed calculated correctly, but on my 120Hz TN panel center area looks darker than rest of it from far away. Even while it has lightness 50 (BGR 127,127,127) which is correct.

    I calibrated my lcd to have correct color representation, but gamma curve is another story since I do not use any special software for calibration anymore. (maybe I should again)
     
  6. yasamoka

    yasamoka Ancient Guru

    Messages:
    4,336
    Likes Received:
    4
    GPU:
    EVGA GTX 1080Ti SC
    Hmmm, my monitor is calibrated by a colorimeter, and I believe gamma has been calibrated to 2.2, not sure though (still learning). I'll take a look at this phenomenon tomorrow.

    Thanks for the interesting explanation!
     
  7. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    9,843
    Likes Received:
    77
    GPU:
    1080Ti H20
    here's mine. pic is rather large.

    [​IMG]
     
  8. -Tj-

    -Tj- Ancient Guru

    Messages:
    13,873
    Likes Received:
    9
    GPU:
    ZOTAC GX980Ti Amp!Extreme
    by the looks of it 7970 clearly has less shimmering in the distance.

    This is with 4xaa + Traa 4xSS and yet there is still some huge shimmering going on -_-

    [​IMG]

    [​IMG]

    worst case scenario :S
    [​IMG]
     
  9. Fox2232

    Fox2232 Ancient Guru

    Messages:
    5,583
    Likes Received:
    5
    GPU:
    Fury X - XL2420T(Z)@144Hz
    Shimmering is caused because filter is too sharp and decides to favor one of zones which cover pixel on screen instead of doing proper average.
    In case of single texture which goes to infinity it has to be totally gray at infinity (if source was B/W checkerboard).

    That is reason why I wrote gtx670 is closer to reference, but HD7970 is giving nicer feel. Since reference does not mean perfect.
    Titan AF is pretty impressive from accuracy point of view compared to reference, but that is what will in end cause shimmering on flat textures.

    @-TJ-: your shimmering is different than texture one, yours is geometry shimmering and that needs quite a lot of compute power to get rid off.
    And sometimes its better to turn off MSAA/SSAA all together and use just postprocess filter like SMAA/FXAA/MLAA because their edge detect works best on sharp edges (AA reduces contrast).

    And here is mine HD7970 13.2b7 driver (only area which matters):
    [​IMG]
     
    Last edited: Mar 10, 2013
  10. yasamoka

    yasamoka Ancient Guru

    Messages:
    4,336
    Likes Received:
    4
    GPU:
    EVGA GTX 1080Ti SC
    Thank you for the input so far, guys.

    Would the filtering quality change with drivers, or is it mostly dependent on hardware?

    -Tj-, may you please provide a 3DCenter shot, seeing as you have a GTX570?
     

  11. warlord

    warlord Master Guru

    Messages:
    695
    Likes Received:
    0
    GPU:
    R9 390X MSI(1080/1500)
    AMD since always use hardware optimizations vs nvidia. Nvidia will always provide better IQ unless AMD drops the hardware optimizations. Look for example 3d Center, whatever settings you will never match alu. Nvidia on max quality settings will even surpass ALU. Also as far as i remember i could match Nvidia only with RP because i turn off optimizations and using negative lod bias. Without RadeonPro, AMD is so, you know i do not need to say it, weak.
     
  12. -Tj-

    -Tj- Ancient Guru

    Messages:
    13,873
    Likes Received:
    9
    GPU:
    ZOTAC GX980Ti Amp!Extreme
    The thing about that UT3 shot, i see it in all games that use fence like floors

    I used 4xaa + transparency AA 4x super sampling, and its still so bad -_-, so you say its not AF side effect fault? It happens if i move the camera by certain angles.


    Btw can you please post a link for this 3dcenter AF test tool? I will check it out when I get home. Im also at quadro driver and this tends to have a little sharper AF. :]
     
    Last edited: Mar 10, 2013
  13. yasamoka

    yasamoka Ancient Guru

    Messages:
    4,336
    Likes Received:
    4
    GPU:
    EVGA GTX 1080Ti SC
    The whole reason of opening this topic is to provide comparisons, not draw conclusions from the start. Have you tried all AMD & Nvidia cards?

    @-Tj-: http://www.3dcenter.org/download/3dcenter-filter-tester
     
  14. -Tj-

    -Tj- Ancient Guru

    Messages:
    13,873
    Likes Received:
    9
    GPU:
    ZOTAC GX980Ti Amp!Extreme
    Ok thanks, will post in a few hrs. :)

    One more thing about my issue with UT3 or other games with such floor textures, does AMD have similar affect like that?
     
  15. kapu

    kapu Ancient Guru

    Messages:
    3,722
    Likes Received:
    0
    GPU:
    MSI Geforce 1060 6gb
    Too me they look close to same, i'd rather look at some game tests.

    if i switch fast between them, i think 7970 is better to my eye , but i doubt anyone can see that during gaming .
     

  16. teleguy

    teleguy Maha Guru

    Messages:
    1,213
    Likes Received:
    4
    GPU:
    GTX 1070/290X
    HD 7970
    [​IMG]

    GTX 560 ti 448
    [​IMG]

    HD 7970
    [​IMG]

    GTX 560 ti 448
    [​IMG]
     
  17. kapu

    kapu Ancient Guru

    Messages:
    3,722
    Likes Received:
    0
    GPU:
    MSI Geforce 1060 6gb
    Why there is soo big difference here ? on the last images.
     
  18. teleguy

    teleguy Maha Guru

    Messages:
    1,213
    Likes Received:
    4
    GPU:
    GTX 1070/290X
    AMD put in a lot of effort to make their AF angle independant, Nvidia didn't.
     
  19. WhiteLightning

    WhiteLightning Don Illuminati Staff Member

    Messages:
    26,669
    Likes Received:
    36
    GPU:
    Inno3d GTX 1070 iChill x4
    Here are my results 4870x2 and GTX580
    i put them in a spoiler because the images are 1,5MB each

    4870x2
    [​IMG]

    GTX580
    [​IMG]
     
    Last edited: Mar 10, 2013
  20. Anarion

    Anarion Ancient Guru

    Messages:
    12,922
    Likes Received:
    2
    GPU:
    EVGA GeForce GTX 1070 ACX
    My rig:
    [​IMG]
    [​IMG]
     
    Last edited: Mar 10, 2013

Share This Page