High Performance Computing Group

 View Only
Expand all | Collapse all

LSF Community Edition 10.1 GPU AutoConfiguration

  • 1.  LSF Community Edition 10.1 GPU AutoConfiguration

    Posted Mon October 30, 2023 03:26 PM

    I've recently installed LSF Community Edition on three GPU machines running Ubuntu 22.04.  LSF GPU Autoconfig identifies the multiple GPUS on each machine as a single GPU.  It correctly identifies the GPU model on one machine, but reports the GPUs on the other machines as UnknownNVIDIART 

    elim.gpu correctly identifies the number of gpus on each machine.

    I've tried manually configuring the gpu resources, but am told that manual configuration is incompatible with LSF_GPU_AUTOCONFIG which I have verified is not set in any of my lsf/conf files.

    I've tried debugging the lims (lsadmin limdebug -c "LD_ELIM LC2_ELIM"' hostname)  but am still getting minimal log files

    Anyone able to point me to some better debugging documentation or examples?

    -Gaylord



    ------------------------------
    Gaylord Holder
    ------------------------------


  • 2.  RE: LSF Community Edition 10.1 GPU AutoConfiguration

    Posted Mon October 30, 2023 08:26 PM

    LSF Community Edition only supports one GPU per node, this is why one GPU info is shown but not others in your case since there are 3 GPUs on a node. , To get full scale support of LSF GPU integration, you should upgrade to LSF Standard Edition.

    Regards,

    Yi



    ------------------------------
    YI SUN
    ------------------------------

    Original Message