[gmx-users] Force (%)" value in md.log file

[David van der Spoel]

Den 2019-12-19 kl. 12:21, skrev Pragati Sharma:
> Hello all,
> 
> I am simulating a polymer system (30000 atoms, 2fs time step, PME) on a
> workstation with CPU: 2X6 cores (24 logical) and 2 2080 RTX GPUs.
> 
> I am running the 2 separate simulations using 10 threads and 1 GPU each
> using command:
> 
> *gmx_tmpi mdrun -v -deffnm md2 -nb gpu -gpu_id 1 -nt 10*
> 
> I found that the force values (%) are very high in md.log file. *Is this
> unusual ? If yes, how to decrease this value?*
MD is all about computing forces, the higher fraction of computer time 
goes to this, the less overhead. The benchmark you refer to below seems 
the strange one instead.

> 
> *Excerpts from md.log file*
> 
> "On 1 MPI rank, each using 10 OpenMP threads
> 
>   Computing:          Num      Num         Call      Wall time
> Giga-Cycles
>                            Ranks    Threads     Count      (s)
> total sum    %
> -----------------------------------------------------------------------------
>   Neighbor search             1   10       1001       4.724
> 160.233              3.4
>   Launch GPU ops.           1   10     200002      10.971        372.134
>                 7.9
> * Force                            1   10     100001      80.756
> 2739.287              57.9*
>   Wait PME GPU gather    1   10     100001       0.472         16.007
>             0.3
>   Reduce GPU PME F       1   10     100001       2.185         74.127
>             1.6
>   Wait GPU NB local         1   10     100001       0.591          20.042
>                 0.4
>   NB X/F buffer ops.          1   10     199001      12.043          408.514
>                8.6
>   Write traj.                       1   10         11          0.337
>    11.429                 0.2
>   Update                           1   10     100001       4.054
> 137.520                2.9
>   Constraints                     1   10     100003      19.981
> 677.776                14.3
>   Rest                                                            3.326
>         112.803                  2.4
> -----------------------------------------------------------------------------
>   Total                                        139.440       4729.873
> 100.0 "
> 
> I am comparing with a published benchmark having values for "*Force=
> 14.9%".*
> 


-- 
David van der Spoel, Ph.D., Professor of Biology
Head of Department, Cell & Molecular Biology, Uppsala University.
Box 596, SE-75124 Uppsala, Sweden. Phone: +46184714205.
http://www.icm.uu.se