B3D's AMD R6xx image quality investigation

Discussion in 'Beyond3D Articles' started by B3D News, Aug 30, 2007.

  1. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    Could you by chance post bigger screenies?Your effort is much appreciated, but the screenies are a tad too small to discern the differences between modes:)
     
  2. Xmas

    Xmas Porous
    Veteran Subscriber

    Joined:
    Feb 6, 2002
    Messages:
    3,298
    Likes Received:
    137
    Location:
    On the path to wisdom
    What the edge detect mode does is compare the colors of all samples per pixel. If they are identical it is considered an interior pixel, and that color value is used for the final pixel (you could say that's a box filter).
    However if the samples differ you are at a geometry edge, and in this case a tent filter is applied to smoothen the edge.

    Of course some pixels along the edge of a triangle have all their subsamples covered by that triangle so the edge detect mode won't consider them edge pixels. You can see that in the screenshots, too. You get a nice gradient, but it ends in a slight "jump".
     
  3. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    Thanks for clearing that up, it was a murky area for me. But the question remains, how would one get the tent-filter+ED combination?
     
  4. ERK

    ERK
    Regular

    Joined:
    Mar 31, 2004
    Messages:
    287
    Likes Received:
    10
    Location:
    SoCal
    As he said AFAIK...
     
  5. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    But then it is a different tent-filter from the narrow/wide implementations that are transparently available...considering the current ATi nomenclature(tripling of base-AA level) it must be a very-wide tent filter(as wide "doubles" the base AA level in CCC terms and narrow multiplies it by 1.5).

    The fact that every reviewer under the sun has mentioned tent and ED as separate entities doesn't help. Nor the fact that in spite of their opening up to the ppl, AMD doesn't detail this thing, which IMHO is one of the best features R600 has brought. Dave, would you kindly give more info on the topic?:D
     
  6. Xmas

    Xmas Porous
    Veteran Subscriber

    Joined:
    Feb 6, 2002
    Messages:
    3,298
    Likes Received:
    137
    Location:
    On the path to wisdom
    I wouldn't give a lot on the nomenclature, but judging from no-X's screenshots it might indeed be a significantly wider filter. It might also be a different filter shape than tent.
     
  7. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    Well, in my experience, sample density is highly improved in 24x ED when compared to the base 8X-a simple black square on white background, slightly tilted shows this quite nicely, so it's probably something else...gah, I hate picking at straws:D
     
  8. no-X

    Veteran

    Joined:
    May 28, 2005
    Messages:
    2,297
    Likes Received:
    247
    I agree with Xmas. It's a wider filter. It's visible on 4x / 8x WT / 12x ED differences (8x zoomed):

    4x (box)
    [​IMG]

    8x (WT)
    [​IMG]

    12x (ED)
    [​IMG]

    More samples per-pixel would cause more continuous transition on this angle (using a good pattern). This two-pixel steps are supposedly caused by significantly wider filter. It works on other angles, it works on 24x ED (because higher sample density of MSAA 8x cause nearer location of these 2-pixel steps), but on 12x ED and this angle, ED isn't perfect and due to this unperfection we can judge how it's done.
     
  9. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    Thanks for the zoomed-in pictures. Yup, you seem to be correct...interesting.
     
  10. no-X

    Veteran

    Joined:
    May 28, 2005
    Messages:
    2,297
    Likes Received:
    247
    You're welcome. As I said, other angles are OK and almost perfectly smooth (zoom 7,5x):

    detail 4x box:
    [​IMG]

    detail 12x ED:
    [​IMG]


    24x ED is OK (almost perfect) for all angles (thanks to sufficient level of per-pixel samples even for very low angles; zoom 8x):

    8x box
    [​IMG]

    24x ED
    [​IMG]

    the only thing which could be improved (to make me happier) would be the transition between last AAed pixel and black area, which isn't perfectly smooth because of box filter usage for this pixel, as Xmas mentioned. The step is even a bit amplified by ATis' gamma correction. I think 2.20 is a bit much, even for my Diamondtron CRT. It would be nice if ATi let users to set their own level, or at least prepare some basic presets (CRT, LCD...). I played a bit with a graphical editor and ~ 1.80 semms to be best for my eyes and my CRT :)
     
  11. ERK

    ERK
    Regular

    Joined:
    Mar 31, 2004
    Messages:
    287
    Likes Received:
    10
    Location:
    SoCal
    To me that doesn't look like a wider filter for ED, but just a finer grid/more samples within the box. The ED pattern shows much more in common with the box than it does with the wide tent. In a way I almost prefer the wide tent because it offers a consistent transition to the edges, as opposed to sometimes there is an abrupt change from one color to another, and sometimes there is an intermediate blended pixel at the edge. With the tent there is always the blended edge=no jaggies (but slightly more blurring perhaps).

    Thanks for the excellent pictures and examples.

    ERK
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...