multiple videoPlanes, videoCards and performance issues


    May 22 2006 | 1:56 am
    Hi list,
    I'm currently trying to build a gl patch that could drive three 800 X 600
    windows (each on a separate 6600 NVidia card stacked in a quad 2.5 G5). The
    tricky part about this project is that I have to subdivize each window in 12
    smaller blocks/videoScreens and have a gl.shader available to do real-time
    treatment on all the video ouputs.
    I'm currently reading the texture files (800 X 600, photoJpg, 15 fps) from
    the ram, slicing them up with jit.scissors and tying them to independent
    gl.texture and videoplane objects. I'm using a separate rendering context
    for each window (seems to be HW).
    I'm able to get 14 fps (if i can trust the fpsgui object...) but the video
    seems to" clog-up" at moments and the computer eventually crashes after 15
    minutes. I was wondering If there would be better ways to structure this
    project.
    I've read previous threads concerning multiple video Outputs and openGL +
    other ones relating to texturing issues. I found there valuable information
    (concerning rendering context, colorspace, window syncing and such) but
    wanted to ask if anyone on the list tried something similar and what kind of
    performance I should expect out of my set-up.
    Thanks,
    Etienne

    • May 23 2006 | 8:46 am
      There was just a post on May 20th called "Using jit.gl.videoplane as
      jit.glue" that might be of interest to you, for a different technique
      with a single texture.
      Otherwise, I use a similar method, but using multiple.jit.submatrix
      objects instead of jit.scissors objects, each with it's own
      jit.texture. I have over 96 of them on a dual-core 2GHz G5 (nVidia
      6600) with no problem (50fps, easy). Perhaps the multiple windows is
      what's bogging you down.
      best,
      evan
    • May 26 2006 | 9:57 pm
      Sorry, forgot to add two necessary subObjects.
      E.
      ----
      max v2;
      --------
      max v2;