Page 1 of 2 1 2 >
Topic Options
#37111 - 08/23/18 05:19 AM Does the free charmm version runs in parallel
Chamikara Offline
Forum Member

Registered: 06/04/18
Posts: 5
Hi,

I installed charmm software from your website and it doesn't run in parallel unfortunately in my PC.Does the free charmm version runs in parallel or in one core?

If it runs in a single core, how to make it run in 8 cores?

thanks


Edited by Chamikara (08/23/18 05:32 AM)

Top
#37112 - 08/23/18 07:17 AM Re: Does the free charmm version runs in parallel [Re: Chamikara]
lennart Online   content

Forum Member

Registered: 09/25/03
Posts: 4742
Loc: ~ 59N, 15E
It does run in parallel using MPI if you add the flag M to the install command, eg:
install.com gnu M

Then run it using mpirun:
mpirun -np 8 charmm
_________________________
Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden

Top
#37322 - 02/10/19 11:31 PM Re: Does the free charmm version runs in parallel [Re: lennart]
SNOW001 Offline
Forum Member

Registered: 11/12/18
Posts: 16
I use free charmm and can run, but get very different results on one core and 4 cores. I use eef1 membrane model and Langevin dynamics, does anyone have some opinions on this kind of problem.

Top
#37323 - 02/11/19 03:45 AM Re: Does the free charmm version runs in parallel [Re: Chamikara]
lennart Online   content

Forum Member

Registered: 09/25/03
Posts: 4742
Loc: ~ 59N, 15E
Some difference is expected, in particluar if you run long dynamics simulations. Do the eef1 testcases run and produce similar results on one and 4 cores?
_________________________
Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden

Top
#37324 - 02/11/19 06:11 AM Re: Does the free charmm version runs in parallel [Re: lennart]
SNOW001 Offline
Forum Member

Registered: 11/12/18
Posts: 16
I have a transmembrane helix contains 23 residues, and placed it perpendicular to the membrane. When I use 4 cores, running for about 5 ns, the helix has local unfolding and run out of the membrane.When I use 1 cores, running for about 60 ns, the helix still maintains its transmembrane state.
Free charmm do not contain DOMDEC, is it communicating all particle information between different cores when calculate force or potential energy?
Part of my input.inp file is:
eef1 setup temp 298.15 unit 93 name solvpar.inp membrane -
slvt water slv2 chex nsmth 10 width 27 aemp 0.85 -
nbond ctonnb 7.0 ctofnb 9.0 cutnb 11.0 group rdie

set fbeta = 5.0
scalar fbeta set @fbeta
scalar fbeta set 0.0 select type H* end
shake bonh param fast
DYNAMICS LANGEVIN start nstep 5000000 timestp 0.002




Edited by SNOW001 (02/11/19 06:14 AM)

Top
#37325 - 02/11/19 12:05 PM Re: Does the free charmm version runs in parallel [Re: Chamikara]
rmv Online   content

Forum Member

Registered: 09/17/03
Posts: 8379
Loc: 39 03 48 N, 77 06 54 W
Standard parallel CHARMM works fine for systems with explicit solvent, and was used as a reference when developing the newer DOMDEC engine.

However, I do not make much use of implicit solvent models myself, and have not evaluated EEF1.

Have you run the test cases?
_________________________
Rick Venable
computational chemist


Top
#37326 - 02/11/19 10:10 PM Re: Does the free charmm version runs in parallel [Re: rmv]
SNOW001 Offline
Forum Member

Registered: 11/12/18
Posts: 16
I run the test cases,
./test.com gnu_M
The last few lines of screen print are:
/home/me/tools/charmm/exec/gnu_M/charmm -input /home/me/tools/charmm/test/c43test/soft14benz.inp -prevclcg >& output/soft14benz.out
/home/me/tools/charmm/exec/gnu_M/charmm -input /home/me/tools/charmm/test/c43test/testimm1-curv.inp -prevclcg >& output/testimm1-curv.out
/bin/rm: No match.
Summary of charmm testresults. Tue Feb 12 10:39:31 CST 2019
chmost gnu_M using /home/me/tools/charmm/exec/gnu_M
outputs in directory output
=============================================================
Testcases that fail:
=============================================================
Testcases that do not finish:
=============================================================
Testcases that finish abnormally:
output/dhdgb_mini.out
=============================================================
Number of testcases that have not been run:
168
=============================================================

I check the testcase about eef1,
eef1_2ndderiv.out (c29test)
testeef1imag.out (c27test)
testeef1.out (c27test)
and no error in them.

I will check if my charmm has other errors.


Edited by SNOW001 (02/11/19 10:24 PM)

Top
#37328 - 02/12/19 07:56 AM Re: Does the free charmm version runs in parallel [Re: Chamikara]
lennart Online   content

Forum Member

Registered: 09/25/03
Posts: 4742
Loc: ~ 59N, 15E
Check if the eef1 testcases give the same results when you with 1 core and with 4 cores.
_________________________
Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden

Top
#37330 - 02/12/19 09:08 AM Re: Does the free charmm version runs in parallel [Re: lennart]
SNOW001 Offline
Forum Member

Registered: 11/12/18
Posts: 16
Their output energy are different.

Other phenomenon:
If use [ vacuum + DYNA VERLET ], 1core and 4cores get same energy within 10000 integral steps, then different gradually.
If use [ eef1 setup temp 298.15 unit 93 name solvpar.inp + DYNA VERLET ], 1core and 4cores get same energy within 100 integral steps, then different gradually.
If use [eef1 setup temp 298.15 unit 93 name solvpar.inp membrane slvt water slv2 chex nsmth 10 width 27 aemp 0.85
+ DYNA VERLET], 1core and 4cores get same energy within 1-2 integral steps, then different gradually.
If use [ DYNA Langevin ], no matter vacuum or implicit membrane, 1core and 4cores get different energy in 1 integral steps.
(the 10000,100 and 1-2 steps are just example values)

Top
#37338 - 02/18/19 03:34 AM Re: Does the free charmm version runs in parallel [Re: lennart]
SNOW001 Offline
Forum Member

Registered: 11/12/18
Posts: 16
Hello, I want to know if eef1/imm1 can do parallel . I know eef1 implicit water model can do parallel ,but how about the eef1 implicit membrane . I think I get something wrong when use it.

Top
Page 1 of 2 1 2 >

Moderator:  John Legato, lennart