The interactive usage is intended for debugging and profiling. Run production and benchmark jobs in the batch queue. For interactive jobs first log in to g8 from vuori with ssh, load required environment and initialize the dynamically linked Gromacs:
ssh g8Now you can run interactively serial Gromacs-GPGPU jobs without the queuing system:
<give password>
module swap PrgEnv-pgi PrgEnv-gnu
module load fftw cuda
module load gromacs/4.5.4-gpu
mdrun-gpu -s topol.tpr -o md-gpu.log
Batch jobs
To run batch jobs in the GPGPU-compute nodes, use a script of the following type (you can also load the module before submitting the job and then omit it from the script):
At the moment only serial runs (one GPGPU-card) are possible. This means, that often more performance (if cores are available) can be gained by using many cores with the traditional cpu-code although certain jobs/algorithms are much more cost efficient on GPGPU cards. A serial batch job would use:
#!/bin/csh
#SBATCH -J gpu-test
#SBATCH -e my_output_err_%j
#SBATCH -o my_output_%j
#SBATCH --mem-per-cpu=1000
#SBATCH -t 04:02:00
#SBATCH -n 1
#SBATCH -p gpu
#SBATCH --gres=gpu:1
module swap PrgEnv-pgi PrgEnv-gnu
module load fftw cuda
module load gromacs/4.5.4-gpu
# note: it is important to set gres=gpu:1 also here
srun --gres=gpu:1 mdrun-gpu -s topol.tpr
The script is submitted with:
sbatch script_name
More information
It is a good idea to check the advantages of GPGPU-version from e.g. here. A lot of useful information can be found also in the Gromacs home page www.gromacs.org and its mailing lists.Back to CSC Gromacs page.