Dear Marko Babić

Thanks for your reply.

I know Gromacs commands very well. I have been working for years. Also, I know input and output files. But, this is the first time I am using cluster to run my simulations.

However, my colleagues use LAMMPS and use the mentioned command. Unfortunately, they don't know how to run Gromacs on the cluster.

The administrator told me they installed the Gromacs in this path:

/nfsexports/SOFTWARE/gromacs-2024.3/build/bin/gmx_mpi

But, I don't know how to run my .tpr file in mdrun command. I mean what is the exact command on this cluster?

does It should be something like :??

srun --propagate=NONE -p parallel -n 30 /nfsexports/SOFTWARE/gromacs-2024.3/build/bin/gmx_mpi mdrun -deffnm npt -v &>SC.out &

Thanks

Dear Marko Babić

Thanks again for your reply. This is the error:

run: job 489788 queued and waiting for resources

srun: job 489788 has been allocated resources

/nfsexports/SOFTWARE/gromacs-2024.3/build/bin/gmx_mpi: error while loading shared libraries: libmkl_intel_lp64.so.1: cannot open shared object file: No such file or directory

Marko Babić

Thanks for your efforts.

Hi Jaffri. Re

First of all, are you using a sbatch file script? Where are you running this command? The admins of your HPC should have a sbatch script file archetype of some sort that you can modify to meet your needs. Please do realize that your question has no real general answer, since it largely depends on the GROMACS installation/HPC hardware and environment, which are specific of your institution.

The error you get is related to the fact that GROMACS does not find some libraries to which it was linked during installation. First, rather than calling the executable, I'd suggest you to source it. You have to find GMXRC file which is usually in the directory where you installed GROMACS, something like

source /nfsexports/SOFTWARE/gromacs-2024.3/install/bin/GMXRC

If you do not find it, then yes, stick to using directly full path to gmx_mpi. To fix the library error you get, you will have to load the necessary libraries that were used to compile GROMACS. How to load them depends on the HPC environment, and is something you can ask to your HPC admins/guides online of your institution. It is most likely that they use module load, and will likely have to load some GCC compiler, the OpenMPI libraries, some GPU libraries (e.g. CUDA if it was compiler against nvidia GPUs), etc. If this is the case, you will have to load the libraries and then source GROMACS/use gmx_mpi, something that - within the sbatch submit file - will look like this

module load GCC

module load OpenMPI

module load CUDA

source .../..../GMXRC

GROMACS is highly parallelizable and the performances you will get depend a lot on the resources you give and how you split them. You will have to find a reasonable amount of nodes, GPUs, CPUs, and consequently a given number of tasks/tasks-per-gpu/cpus-per-task etc which you can fine tune by using the mdrun flags of GROMACS and requesting the same resources to the HPC nodes via slurm. For example, I never saw these flags you are using "--propagate=NONE -p parallel", so they might be okay for LAMPPS but not for GROMACS. The way you are running the sims, that is mdrun -deffnm npt -v, is the bare minimum. GROMACS might be able to auto-fit a good approximation of the best combination of num GPUs/CPUs/tasks etc based on the hardware you provide, but I am not sure it will work, especially with an OpenMPI installation. To get better performances, you will need to fine tune it anyway, take a look at the manual

for mdrun flags

https://manual.gromacs.org/current/onlinehelp/gmx-mdrun.html

how to run in HPC/fine tune your runs

https://manual.gromacs.org/current/user-guide/mdrun-performance.html#getting-good-performance-from-mdrun

and finally SLURM

https://hpc-unibe-ch.github.io/slurm/submission.html

To run GROMACS on an HPC cluster:

all of which can we easily done at mdsim360.com, a new platform that lets you run MD simulations entirely online without local installation.