[gmx-users] (no subject)

[Szilárd Páll]

Please do not post questions to the list "owner" (=admin) address. Post
your questions to the users' list instead.

Are you sure you are using the MPI-enabled GROMACS installation? Look in
the mdrun log header (posting the link to the whole log uploaded to some
sharing service might also help identifying the issue).

--
Szilárd

On Sat, May 20, 2017 at 1:09 AM, Li, Zhixia <zhixia2 at illinois.edu> wrote:

> Hi all,
>
> Recently, I want to try hybrid parallelization on two nodes with 12 cores
> per node.  I want it to run 1 mpi per node and 12 openmp per mpi. The
> command I use is:
> mpirun -np 2 gmx_mpi mdrun -ntomp 12
>
> But I got the output like the following. It seems it only uses one node.
> Does anyone know how to realize it? And is it the issue of compiling? Thank
> you.
>
> Number of logical cores detected (12) does not match the number reported
> by OpenMP (6).
> Consider setting the launch configuration manually!
>
> Running on 1 node with total 12 cores, 12 logical cores
> Hardware detected on host taub217 (the node of MPI rank 0):
>   CPU info:
>     Vendor: GenuineIntel
>     Brand:  Intel(R) Xeon(R) CPU           X5650  @ 2.67GHz
>     SIMD instructions most likely to fit this hardware: SSE4.1
>     SIMD instructions selected at GROMACS compile time: SSE4.1
>
> Reading file em4.tpr, VERSION 5.1.4 (single precision)
> The number of OpenMP threads was set by environment variable
> OMP_NUM_THREADS to 12 (and the command-line setting agreed with that)
> Using 2 MPI processes
> Using 12 OpenMP threads per MPI process
>
>
> WARNING: Oversubscribing the available 12 logical CPU cores with 24
> threads.
>          This will cause considerable performance loss!
>
> NOTE: Your choice of number of MPI ranks and amount of resources results
> in using 12 OpenMP threads per rank, which is most likely inefficient. The
> optimum i$
>
>
> Non-default thread affinity set probably by the OpenMP library,
> disabling internal thread affinity
>
>
> Best,
> Zhixia Li
>