visit

  • About: Open Source, interactive, scalable, visualization, animation and analysis tool (pre-compiled)
  • Category: Graphics
  • URL: https://visit.llnl.gov/
module (bold = default) build dependencies
visit/2.10.1/b1
visit/2.7.2/b2 pre-compiled xterm
visit/2.7.3 pre-compiled xterm openmpi/1.6.5/b1
visit/2.8.2/b4 pre-compiled xterm

Running a serial version of visit on a visual node

To get full rendering capabilities, you will need to run the visit client on a gpu-enabled node in the visual queue. Instructions for starting and connecting to a remote visualization node can be found here.

You can launch visit through the drop down menu, or you can run visit in parallel by using the terminal inside of the TurboVNC desktop and typing

module load virtualgl
module load visit/2.7.3
vglrun visit

Running a parallel version of visit on a single visual node

When you start a remote desktop server using vnc_start, the entire node is allocated to your job. You can launch a parallel engine on the localhost by opening a terminal inside of the TurboVNC desktop and typing

module load virtualgl
module load visit/2.7.3
vglrun visit -np $[ $SLURM_CPUS_ON_NODE + 1 ]

Running visit and launching a parallel compute engine as a separate job in the cluster

After starting a remote desktop server using vnc_start, open a terminal inside of the TurboVNC desktop and launch visit using

module load virtualgl
module load visit/2.7.3
vglrun visit -par -forcestatic -noloopback -sshtunneling -launchengine bluehive.circ.rochester.edu -l srun -np 44 -p debug -t 60

This will start the client on the remote visualization node, start the engine server on bluehive.circ.rochester.edu, and fire up a parallel compute engine within a slurm job (in this case 44 cores in the debug partition for 60 minutes). You will need to disable scalable rendering with this approach.

Running visit and a parallel compute engine as part of the same job

If you want to use more than one node for the parallel compute engine, but without having to submit two separate jobs, you can ask for multiple nodes when you run vnc_start (here we are asking for 2 nodes for 60 minutes with 23GB of memory per node)

vnc_start -g 1920x1080 -t 60 -N 2 --mem=23g

After connecting to the turboVNC server running on the first node in the allocation, you can open a terminal and run

module load virtualgl
module load visit/2.7.3
vglrun visit -par -forcestatic -noloopback -sshtunneling -launchengine bluehive.circ.rochester.edu -l srun -np $SLURM_NTASKS -la --jobid=$SLURM_JOBID

This will launch the visit client and connect it to an engine server running on bluehive and then start a parallel job within the same allocation as your remote desktop session. You will need to disable scalable rendering with this approach.

Starting a parallel visit compute engine on BlueHive from a client running on your local workstation

You can start parallel compute engines on bluehive from your own local copy of visit once you setup a host profile.

You can download the host profile for bluehive host_bluehive.xml and place it in the .visit/hosts subdirectory of your home directory.

After you start visit on your local machine, you will need to update the host profile with your netid. Click on Options->Host Profiles, select bluehive and then enter your netid in the Username field. Then save your changes by clicking Options->Save Settings

Please note that this only currently works for visit/2.7.3 and you will need to place the following lines in your .bashrc file in your home directory on bluehive.

source /software/modules/init/bash
module load visit/2.7.3

or if your default shell is tcsh, then the following should be added to your .tcshrc file

source /software/modules/init/tcsh
module load visit/2.7.3

If you are not sure what your shell is, you can find out by running

echo $SHELL

Please note that at this time, scalable rendering will not work, so be sure to set scalable rendering to Never under the Rendering Preferences

also see Using Software.