paralleling calculations

  • OldETC

    Hi, guys,

    I have been working with ngspice for the last year or so... And love

    it!!! Thanks..
    However I am moving up in my simulations by quantity and complexity of
    circuits. As I do so, the calculations are taking quite long
    (understandably of course), but I was wondering about the process of
    calculation and wondering if using a graphics card for processing the
    nodes or components might be faster. I know from experience that each
    node depends on the previous nodes in the signal path and that there may
    be several signal paths, so I was thinking something like:

    processor step0 step1 step2 step3 step4
    1 T0 T1 T2 T3
    2 wres1 T0 T1 T2
    3 wres2 wres2 T0 T3
    4 wres3 wres3 wres3 T4
    where T# represents the time strobe increment
    wres means waiting for results of the prior node
    Completion would be when the time x (end time or greater) passes through
    the pipeline. I haven't thought of how to handle the case where nodes
    or devices are greater than the number of shaders, but you guys already
    handle that, so maybe you have a partitioning algorithm already in

    Since a video board has up to 2048 shaders, it might be possible to get
    up to 2000 times the through put, -minus the DC setup time.

    Maybe you guys have already thought of this, if so, sorry for the noise.