[gmx-users] Question about Multi-level parallelization: MPI and OpenMP

[Mark Abraham]

On 28/08/2012 2:40 AM, jesmin jahan wrote:
> Dear Gromacs Users,
>
> I have two questions about the multi-level parallelism of Gromacs.
> http://www.gromacs.org/Documentation/Acceleration_and_parallelization?highlight=verlet#GPU_acceleration
>
> 1. Is this feature is only supported by Gromacs 4.6? Or we can get it
> in 4.5.3 also?

That section of that page is pretty specific about versions. The shiny 
new stuff is due in 4.6.

> 2. In Gromacs 4.5.3, I have used OMP_NUM_THREADS=12 mpirun -np 16
> mdrun_mpi -s imd.tpr command to run 16 gromacs processes each with 12
> threads.
>
> In the log file, I can see
>
>   nodeid: 0  nnodes:  16
>
> Its clear that 16 nodes are being used in work load distribution. But
> its not clear whether 12 thread is being used in each of those
> processes /nodes because there is no mention about the number of
> threads.
>
> Does anyone know about this?

Again from that page: "With GROMACS 4.6 OpenMP multithreading is 
supported..."

Mark