[gmx-users] Why it is run so slow in gromacs 2019.3 compared to VERSION 5.1.1?

[Justin Lemkul]

On 1/16/20 11:57 AM, ZHANG Cheng wrote:
> I have re-run the MD, with only 2 ps in both version. Indeed, the Version 2019 is far slower than the Version 5.1.1.
>
>
> Version 2019 needs 114.111 hours to complete 1 ns, while Version 5.1.1 only needs 6.362 hours to complete 1 ns. 
>
>
> The new logs files are at
> https://github.com/lanselibai/gromacs-20200116
>
>
>
> The MD is run on our HPC cluster. So I personally could not compile it. Is there still some way to get equivalent performance for the Version 2019?
>

Try assigning the same number of PP and PME ranks in each run (30/6 in 
5.1.1 and 27/9 in 2019.3). You're losing a lot of performance in 
communicating PME forces in the 2019.3 run. You may also want to try 
version 2020 to see if automatic detection has improved.

-Justin

>
>
> ------------------ Original ------------------
> From:&nbsp;"ZHANG Cheng"<272699575 at qq.com&gt;;
> Date:&nbsp;Thu, Jan 16, 2020 07:58 AM
> To:&nbsp;"ZHANG Cheng"<272699575 at qq.com&gt;;"gromacs.org_gmx-users"<gromacs.org_gmx-users at maillist.sys.kth.se&gt;;
>
> Subject:&nbsp;Re:Why it is run so slow in gromacs 2019.3 compared to VERSION 5.1.1?
>
>
>
> Hi Justin, what kind of information should I look at in the log files? They are too big to paste here. Would it be possible if you can see them at https://github.com/lanselibai/gromacs-20200115&nbsp;?
>
>
> Thank you!
>
>
> ------------------ Original ------------------
> From:&nbsp;"ZHANG Cheng"<272699575 at qq.com&gt;;
> Date:&nbsp;Thu, Jan 16, 2020 03:38 AM
> To:&nbsp;"gromacs.org_gmx-users"<gromacs.org_gmx-users at maillist.sys.kth.se&gt;;
> Cc:&nbsp;"ZHANG Cheng"<272699575 at qq.com&gt;;
> Subject:&nbsp;Why it is run so slow in gromacs 2019.3 compared to VERSION 5.1.1?
>
>
>
> I have a nearly identical run using the "VERSION 2019.3" compared to my previous "VERSION 5.1.1". Everything during the preparation is the same except "-r" needs to be added in the "VERSION 2019.3". So "-r em.gro", "-r nvt.gro" and "-r npt.gro" are added in the grompp commands for NVT, NPT and production run, respectively.
>
>
> Using "-pe mpi 12" (i.e. 12 nodes) per hour, only 5-7 ps can be processed in the "VERSION 2019.3", while 200 ps can be achieved in the "VERSION 5.1.1". Can I ask is this normal? Is there some configuration I can do in the "VERSION 2019.3" so as to accelerate the MD?
>
>
> I follow Justin's tutorial at http://www.mdtutorials.com/gmx/lysozyme/index.html
> The commands I used are listed below.
>
>
> gmx pdb2gmx&nbsp; -f Fab.pdb -o Fab_processed.gro -water spce&nbsp; -inter&nbsp; -ignh -merge interactive
> gmx editconf -f Fab_processed.gro -o Fab_newbox.gro -c -d 1.0 -bt cubic
> gmx solvate -cp Fab_newbox.gro -cs spc216.gro -o Fab_solv.gro -p topol.top
> gmx grompp -f ions.mdp -c Fab_solv.gro -p topol.top -o ions.tpr -maxwarn 1
> gmx genion -s ions.tpr -o Fab_solv_ions.gro -p topol.top -pname NA -nname CL -nn 31 -neutral -conc 0.05
> (Afterwards, using the HPC to run)
> gmx grompp -f minim.mdp -c Fab_solv_ions.gro -p topol.top -o em.tpr
> gerun mdrun_mpi -v -deffnm em
> gmx grompp -f nvt.mdp -c em.gro -p topol.top -o nvt.tpr -r em.gro
> gerun mdrun_mpi -deffnm nvt
> gmx grompp -f npt.mdp -c nvt.gro -t nvt.cpt -p topol.top -o npt.tpr -r nvt.gro
> gerun mdrun_mpi -deffnm npt
> gmx grompp -f md.mdp -c npt.gro -t npt.cpt -p topol.top -o md_0_1.tpr -r npt.gro
> gmx mdrun -deffnm md_0_1 -cpi -append

-- 
==================================================

Justin A. Lemkul, Ph.D.
Assistant Professor
Office: 301 Fralin Hall
Lab: 303 Engel Hall

Virginia Tech Department of Biochemistry
340 West Campus Dr.
Blacksburg, VA 24061

jalemkul at vt.edu | (540) 231-3129
http://www.thelemkullab.com

==================================================