multiple screens performance


    Oct 17 2006 | 7:40 pm
    hi list
    just a quick ping about performance and multiple screens:
    we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X slots.
    we need to run one screen and 4 projections from that machine and
    apparently the only way to go seems to be using relatively slow PCI
    graphics cards.
    question: how much do the cards influence the overall performance.
    will stuff be unbearably slow once something is on one of the slow
    card's screens (as recently experienced with openGL in processing) or
    are the processors most important part here?
    cheers
    sascha
    www.blinksandbuttons.net
    blog.plugimi.com
    AIM + skype: plugimi

    • Oct 17 2006 | 9:38 pm
      On Oct 17, 2006, at 12:40 PM, sascha pohflepp wrote:
      > we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X
      > slots. we need to run one screen and 4 projections from that
      > machine and apparently the only way to go seems to be using
      > relatively slow PCI graphics cards.
      >
      > question: how much do the cards influence the overall performance.
      > will stuff be unbearably slow once something is on one of the slow
      > card's screens (as recently experienced with openGL in processing)
      > or are the processors most important part here?
      Sending large amounts of data (e.g. textures and huge vertex arrays)
      across the 1x PCI bus as opposed to the faster AGP bus will be slow,
      but the rendering itself should exhibit decent performance. In
      Jitter, I'd recommend using display lists whenever possible for such
      a setup, and don't use monitor spanning windows unless they are both
      driven by the the same GPU. Another thing to keep in mind is that VBL
      sync will degrade your framerate by an integer multiple, one for each
      screen. So either turn off sync or use something like the Matrox
      triple head (this will also let you connect up to 6 displays from one
      dual head AGP card that has enough VRAM). Link to matrox triple head
      below (in the instance of the triple head you really only have two
      monitors, it gets cut up after the fact, and you're better off using
      one large window for each "triple monitor" output).
      More info on this is in the Jitter forum archives. search on "matrox"
      or "triplehead"
      -Joshua
    • Oct 27 2006 | 3:59 pm
      today i tried out the setup with two ATI radeon 9200 but it's pretty
      slow so far, currently trying to optimize. the size of the matrices
      seems to have the most impact, so it's kinda smooth with 320 x 240,
      which is not really great. any more ideas?
      what are display lists?
      cheers
      sascha
      >> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X
      >> slots. we need to run one screen and 4 projections from that
      >> machine and apparently the only way to go seems to be using
      >> relatively slow PCI graphics cards.
      >>
      >> question: how much do the cards influence the overall performance.
      >> will stuff be unbearably slow once something is on one of the slow
      >> card's screens (as recently experienced with openGL in processing)
      >> or are the processors most important part here?
      >
      > Sending large amounts of data (e.g. textures and huge vertex
      > arrays) across the 1x PCI bus as opposed to the faster AGP bus will
      > be slow, but the rendering itself should exhibit decent
      > performance. In Jitter, I'd recommend using display lists whenever
      > possible for such a setup, and don't use monitor spanning windows
      > unless they are both driven by the the same GPU. Another thing to
      > keep in mind is that VBL sync will degrade your framerate by an
      > integer multiple, one for each screen. So either turn off sync or
      > use something like the Matrox triple head (this will also let you
      > connect up to 6 displays from one dual head AGP card that has
      > enough VRAM). Link to matrox triple head below (in the instance of
      > the triple head you really only have two monitors, it gets cut up
      > after the fact, and you're better off using one large window for
      > each "triple monitor" output).
      www.blinksandbuttons.net
      blog.plugimi.com
      AIM + skype: plugimi
    • Oct 27 2006 | 6:16 pm
      I have a setup that drives 3 screens from one graphics card using the Matrox Triple Head 2Go
      splitter.
      I create one wide matrix (3072 x 768) that I populate within Jitter and then feed to the Triple
      Head and it splits it into three 1024 x 768 vga monitor feeds. Works great and allows you to move
      items accross the large internal matrix with impunity and still have them stitch together on the 3
      external screens.
      Keith
      --- sascha pohflepp wrote:
      > today i tried out the setup with two ATI radeon 9200 but it's pretty
      > slow so far, currently trying to optimize. the size of the matrices
      > seems to have the most impact, so it's kinda smooth with 320 x 240,
      > which is not really great. any more ideas?
      >
      > what are display lists?
      >
      >
      > cheers
      > sascha
      >
      >
      >
      > >> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X
      > >> slots. we need to run one screen and 4 projections from that
      > >> machine and apparently the only way to go seems to be using
      > >> relatively slow PCI graphics cards.
      > >>
      > >> question: how much do the cards influence the overall performance.
      > >> will stuff be unbearably slow once something is on one of the slow
      > >> card's screens (as recently experienced with openGL in processing)
      > >> or are the processors most important part here?
      > >
      > > Sending large amounts of data (e.g. textures and huge vertex
      > > arrays) across the 1x PCI bus as opposed to the faster AGP bus will
      > > be slow, but the rendering itself should exhibit decent
      > > performance. In Jitter, I'd recommend using display lists whenever
      > > possible for such a setup, and don't use monitor spanning windows
      > > unless they are both driven by the the same GPU. Another thing to
      > > keep in mind is that VBL sync will degrade your framerate by an
      > > integer multiple, one for each screen. So either turn off sync or
      > > use something like the Matrox triple head (this will also let you
      > > connect up to 6 displays from one dual head AGP card that has
      > > enough VRAM). Link to matrox triple head below (in the instance of
      > > the triple head you really only have two monitors, it gets cut up
      > > after the fact, and you're better off using one large window for
      > > each "triple monitor" output).
      >
      >
      >
      >
      >
      >
      > www.blinksandbuttons.net
      > blog.plugimi.com
      > AIM + skype: plugimi
      >
      >
      >
      Keith McMillen
      BEAM Foundation
      http://www.beamfoundation.org/
      510.502.5310
    • Jan 05 2007 | 5:56 am
      Hi all,
      I'm pretty sure it has come up a few times before but couldn't find it.
      I have a patch that needs to be split to two beamers, each on a resolution of 1280x768 making a 32/9 screen I don't have a triplehead from matrox and when I order it will take some time to get here do that's not an option at the moment.
      I was wondering if my plain G5 with build-in AGP card (64mb I'm afraid) would allow spanning to two screens. I know you shouldn't span on diffrent graphic cards and trying on a macbook (maybe dumb to do since it hasn't got any real GPU) it doesn't work. I can't test it on the G5 but need to know if this will work. Can anyone lead me to an aproach that will work?
      Best Regards
      pieter