John Carmack on DOOM3 Rendering

Discussion in 'Beyond3D News' started by Dave Baumann, Jun 11, 2002.

Thread Status:
Not open for further replies.
  1. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    [​IMG]Reverend has posted a short interview with John Carmack in which we gain some idea on how Doom III renders and what ramifications this may have on today’s Graphics cards.

    Read the interview here.
     
  2. Rodéric

    Rodéric a.k.a. Ingenu
    Moderator Veteran

    Joined:
    Feb 6, 2002
    Messages:
    3,983
    Likes Received:
    846
    Location:
    Planet Earth.
    Interesting.

    I wonder if the first 'Z filling' pass is still required on PowerVR hardware, anyone can answer this ?
     
  3. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,883
    Likes Received:
    209
    Location:
    Seattle, WA
    There's no way that the first "z-filling" pass would do anything of note on PowerVR hardware, unless it ran in immediate mode, which would certainly make it unable to run DOOM3.

    Basically, if the PowerVR guys show that their Kyro can do halfway-decently in DOOM3, JC may optimize for it. If not, people with Kyro's will be SOL.
     
  4. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    KYRO usually batches all passes together and then renders each pass at a tile level, so I doubt there would be any difference with the Z-fill stage (unless this entirely screws things up on it).
     
  5. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,883
    Likes Received:
    209
    Location:
    Seattle, WA
    Come to think of it, back when I read the paper on how JC does those shadows, the original z-buffer write is imperative. The Kyro had better cache all of the instructions before rendering on a per-tile basis...there's no way it would work otherwise.
     
  6. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    Parhelia

    Will the Parhelia help out?
     
  7. KimB

    Legend

    Joined:
    May 28, 2002
    Messages:
    12,883
    Likes Received:
    209
    Location:
    Seattle, WA
    Parhelia? Help out what?

    The Parhelia should perform well in DOOM3, but I doubt it look nearly as good as the NV30 or R300.
     
  8. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    Kyro should be fine. It can run my shadow volume demo which also uses z-fill at its first pass. Unfortunately, Kyro does not support cube map. With very simple lighting (not per-pixel), Kyro beat GF3 running my demo.
     
  9. pcchen

    pcchen Moderator
    Moderator Veteran Subscriber

    Joined:
    Feb 6, 2002
    Messages:
    2,743
    Likes Received:
    106
    Location:
    Taiwan
    Sorry, forgot to login to write the previous post :eek:
     
  10. Rodéric

    Rodéric a.k.a. Ingenu
    Moderator Veteran

    Joined:
    Feb 6, 2002
    Messages:
    3,983
    Likes Received:
    846
    Location:
    Planet Earth.
    mmmh...
    So is this pass a requirement even on PowerVR hardware or not ?

    If it's not, then pcchen, could you run your demo with and without the first pass on PowerVR hardware in order to see how much it affects performances, please ?


    Is the problem with ATI TRUFORM not to look good (cause shadows would represent non tesselated geometry), or to look wrong in any way ?

    If it's just that it doesn't look too good, I don't care and would like to have an option to enable npatchs anyway, anybody else agrees ?
     
  11. pcchen

    pcchen Moderator
    Moderator Veteran Subscriber

    Joined:
    Feb 6, 2002
    Messages:
    2,743
    Likes Received:
    106
    Location:
    Taiwan
    The first Z-fill pass is mandatory, since you need the Z values for correct shadow volume interaction. This is no different on a Kyro. Anyway, Kyro handles such pass and shadow volumes extremely fast, there is no need to get rid of the first pass.

    Regarding N-patch (Truform), it may create some "holes" in the shadow. The shadow volumes have to be completely closed to avoid artifacts. If the model is convex, N-patch may be fine. However, any model with concave triangles, N-patch will make the shadow volume open and create inappropriate "shadows" at some places.
     
  12. Rodéric

    Rodéric a.k.a. Ingenu
    Moderator Veteran

    Joined:
    Feb 6, 2002
    Messages:
    3,983
    Likes Received:
    846
    Location:
    Planet Earth.
    I see, thanks.
     
  13. LittlePenny

    Regular

    Joined:
    Feb 10, 2002
    Messages:
    276
    Likes Received:
    0
    Location:
    Rolla, Missouri, USA
    Damn shame he is so busy, but it makes it all the more nice that he took the time to answer our questions. I wouldn't mind hearing from Jan Paul Van Waveren about his Physics implementations, but I may be on the wrong board for that.
     
  14. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    That went right over my head. NIce to see that he interacts with the community tho.
     
  15. Doomtrooper

    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    3,328
    Likes Received:
    0
    Location:
    Ontario, Canada
    Grrrr stole my UT 2003 leak thunder :lol:

    Ahh well I'll have a R300 anyways..wife will get a nice new 8500 :)
     
  16. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    About the Parhelia, this is a post I made to another board full of idiots who had nothing to say but "d00d, it'll take a GeForce8!?!!! :0p." Basically, it's me thinking out loud:

    Interesting thing, this new technology is. The blatant abuse of bump mapping I think may come to bite id on the ass this time. Bump mapping works well for screen shots and fine grains (such as grass, gravel, etc.) but when it comes to actually moving around a bump mapped object it ends up looking flat. Of course, most bump mapping we've seen so far IS just a grain of some sort put on a surface with no real attention to what geometry that surface is really supposed to represent, so I'm probably wrong.

    I have to wonder if Carmack has looked into environment mapping (parhelia's marketing gimmick). Of course, adding something like that this late in development would give all your artists aneurisms, but still, it's a compelling technique that would suit doom III almost perfectly.

    Of course, people will say that it's silly to write your game for just one card, but Carmack's a smart guy, I'm sure he could figure out a way to do pretty much the same thing with a grayscale image and a vertex shader. Of course, the issue of speed is still there.

    That's the thing:

    Environment mapping vs. Bump Mapping:

    Of course environment mapping is visually superior, but what about performance? With bump mapping, you have the same poly count with a pixel shader applied to it. The AGP bus has the same amount shoved over it as with a multi-textured polygon, but you get a texture that responds to light as if it existed in 3D space.

    With environment mapping, you have essentially the same thing going on as far as the CPU is concerned, but the GPU has to work an insane amount harder. Is this acceptable with the DOOM III engine and today's hardware?

    Probably not... the shadow volumes used by doom 3 make every scene require an ungodly number of passes to render (I'd guess 5), which means your video card is already damn near maxed out. If done "properly," shadow volumes are calculated almost entirely on the card.

    You could try some optimizations by pushing some of the shadow volume calculations back onto the CPU, but your AGP bus would quickly become saturated, defeating the purpose.

    If you tried to force the card to do environment mapping on top of that, then the geometry the card was calculating shadow volumes off of would become incredibly more complex. The fill rate would triple.

    Environment mapping and shadow volumes don't mix.

    Hmm, makes sense, I don't see any Parhelia screenshots that show an environment mapped object casting a shadow.

    I guess that explains why Carmack never even mentioned the Parhelia in any of his interviews.

    http://www.rit.edu/~dmp9199/pics/yippee.JPG <-- be nice, I'm a rookie
     
  17. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    wondering if the parhelia makes it to the big market at all
     
  18. Nappe1

    Nappe1 lp0 On Fire!
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,529
    Likes Received:
    3
    Location:
    South east finland
    where the hell you pull that thought? :roll:
    some facts to base this comment, pretty please... if you have... But I surely doubt that you have any.
     
  19. Joe DeFuria

    Legend

    Joined:
    Feb 6, 2002
    Messages:
    5,994
    Likes Received:
    70
    Well, I assume you are actully talking about Displacement Mapping vs. bump mapping, NOT "Environment Mapping."

    I believe displacement mapping would wreak havoc with Doom's shadowing system. (Someone correct me if I'm wrong.)

    But the main problem with displacement mapping is: no one supports it "yet." If Carmack built the engine with displacement mapping as the cornerstone (instead of bump-mapping), then there would be relatively few cards able to run it at Doom3's release. And of course, not having any real hardware to develop Doom3 on over the past couple years might be a "minor" issue. ;)

    As an aside, IIRC, Matrox will not be exposing displacement mapping in GL anyway....
     
  20. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    Sheesh! I think somebody needs to read our P10 tech preview... :roll:

    :wink:
     
Loading...
Thread Status:
Not open for further replies.

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...