TACC Launcher

These docs are heavily based off the TACC User Portal docs.

  • Account on Ganymede or Europa.

  • Logged into Ganymede or Europa.

  • Have software to run.

Don’t use MATLAB with the TACC Launcher

1. Job Scripts

Two scripts are used to run a serial job.

Slurm serial job script.
#SBATCH -J launchergpu              # job name
#SBATCH -o launchergpu.o%j          # output and error file name (%j expands to jobID)
#SBATCH -N 4                        # number of nodes requested
#SBATCH -n 16                       # total number of tasks to run in parallel
#SBATCH -p normal                      # queue (partition)
#SBATCH -t 02:30:00                 # run time (hh:mm:ss)

module load launcher

export LAUNCHER_WORKDIR=/home/nbc170001/TACCLauncher
export LAUNCHER_JOB_FILE=jobfile.sh


LAUNCHER_JOB_FILE points to the script executing software. For example to run a c++ executable:

Example LAUNCHER_JOB_FILE=jobfile.sh

echo "Hello, World!"

./Main 3
./Main 4
./Main 10
./Main 15
To run your jobscript on compute nodes, use sbatch
$ sbatch runserial.sh
To check on job status
[nbc170001@ganymede TACCLauncher]$ squeue

450558    normal launcher nbc17000  R       0:53      4 compute-1-8-7,compute-7-3-[20-22]

The process of running multithreaded jobs is similar to running serial jobs. .Slurm multithreaded job script

#SBATCH -I launcher-test	# job name
#SBATCH -o launcher.o%j		# output and error file name
#SBATCH -N 2			# number of nodes request
#SBATCH -n 16			# total number of tasks to run in parallel
#SBATCH -p developent		# queue (partition)
#SBATCH -t 00:30:00		# run time (hh:mm:ss)

module load launcher


export LAUNCHER_WORKDIR=/scractch1/01234/joe/tests
export LAUNCHER_JOB_FILE=hello_openmp


2. Ensure Tasks Per Node Are Accurate

To specify the number of tasks per node the following is used in the job script:

#SBATCH -n 4
Ensure the correct number of CPU per node using
$ sinfo --Node --long