NVIDIA Workshop - September 2020

Note

The compute group compute-workshop and the queues workshop and workshop-interactive are only available to those who partake in the workshop and only for a limited time. If you wish to use compute services beyond the workshops, you will need to sign up for access here.

Quick Start

Run on LSF

export JPORT=<port_number>
LSF_DOCKER_VOLUMES="$HOME:$HOME" LSF_DOCKER_PORTS="$JPORT:$JPORT" bsub -G compute-workshop -M 32GB -Is -R "gpuhost rusage[mem=32GB] select[port$JPORT=1]" -q workshop-interactive -gpu "num=1:gmodel=TeslaV100_SXM2_32GB" -a 'docker(gcr.io/ris-registry-shared/nvidia-workshop-sept-2020)' /opt/conda/bin/entrypoint.sh

  • Please see our documentation for more information on selecting a port.

  • For workshop purposes, placed the script into a scratch1 space that should be accessible to all compute users:

export JPORT=<port_number>
/scratch1/fs1/ris/application/nvidia-workshop/start.sh

  • Point browser to the URL given that starts http://compute1-exec-<host>.ris.wustl.edu:… where <host> is replaced by the exec node the job landed on. This is found in the terminal after the job command as <<Starting on compute1-exec-<host>.ris.wustl.edu>>.

Run on LSF if not using the MedSchool VPN

  • You can also connect to the gui with port forwarding on your local machine. You can find the documentation on port forwarding here.

  • There are some slight differences when using the workshop queues. These are noted below.

    • Use the following ssh command instead of the one in the port fowarding documentation.

    ssh -L $JPORT:compute1-exec-<host>.compute.ris.wustl.edu:$JPORT <wustlkey>@compute1-client-<new-host>.ris.wustl.edu

    • Replace <host> with the exec node where the job landed. This is found in the terminal after the job command as <<Starting on compute1-exec-<host>.ris.wustl.edu>>.

    • Replace <new-host> with a number in the range 204-212.

Seminar Repository