[CP2K:8709] Libinit compilation
hut... at chem.uzh.ch
hut... at chem.uzh.ch
Fri Feb 17 17:33:21 UTC 2017
Hi
integrals are distributed over processors by CP2K. Each processor
calls libint to calculate the integrals.
(lack of) Parallelization of libint is not causing your problem.
regards
Juerg
--------------------------------------------------------------
Juerg Hutter Phone : ++41 44 635 4491
Institut für Chemie C FAX : ++41 44 635 6838
Universität Zürich E-mail: hut... at chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------
-----cp... at googlegroups.com wrote: -----To: cp2k <cp... at googlegroups.com>
From: Reza
Sent by: cp... at googlegroups.com
Date: 02/17/2017 06:21PM
Subject: Re: [CP2K:8709] Libinit compilation
Very thanks for time devoted to my discussion.
But my main question was about parallel compiling of Libint. Is it possible? May be my very bad performance of hybrid method is due to the serial compiling of Libint.
On Friday, February 17, 2017 at 8:47:55 PM UTC+3:30, Reza wrote:
Hi again,
I have set parameters to obtain the same total energy to the Gaussian, as could as possible. the change of total energy in the last SCF converged in Gaussian is about 5*10(-9) and for CP2K is 8*10(-9).
the grid of Gaussian is 75 for radial and 302 for angular point. Setting lower or higher cutoff of grids in CP2K changes total energy in CP2K.
total energy of Gaussian is -9143.41735963 and for CP2K is -9142.9538484165.
About memory, you are right. I forgot to increase it in CP2K.
You can see the output of Gaussian as attached file.
On Friday, February 17, 2017 at 2:25:42 PM UTC+3:30, jgh wrote:Hi
from your output one can see that your settings (no input to confirm)
are far from optimal. I'm also sure you didn't use the same tight settings
in your Gaussian calculations.
three obvious points:
- no screening of integrals (10^-16)
- recalculation of almost all integrals (not enough memory allocated)
- very tight SCF convergence and non-optimal OT settings (your BLYP
calculation could converge 10 times faster)
regards
Juerg
--------------------------------------------------------------
Juerg Hutter Phone : ++41 44 635 4491
Institut für Chemie C FAX : ++41 44 635 6838
Universität Zürich E-mail: hut... at chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------
-----cp... at googlegroups.com wrote: -----To: cp2k <cp... at googlegroups.com>
From: Reza
Sent by: cp... at googlegroups.com
Date: 02/17/2017 10:48AM
Subject: Re: [CP2K:8706] Libinit compilation
Dear Juerg;
Thank you for your answer. I calculated nearly the same job for Gaussian09 and CP2K for C240 molecule (contains 240 carbon atom).
For BLYP, wall clock times of Gaussian09 and CP2K are similar (about 20min), but for hybrid methods such as PBE0 and B3LYP, the wall clock times are very different. For Gaussian09, PBE0 time is about 40 min, and B3LYP is about 45 min, but for CP2K, PBE0 two cycles of SCF needs 96 min!!!!. It needs at least 10 cycles to SCF to be converged.
Please see the attached files.
I think this is because I didn't use parallel compiling for Libint (googling doesn't show any parallel compiling for Libint).
Thanks again.
On Friday, February 17, 2017 at 12:37:35 PM UTC+3:30, jgh wrote:Hi
without more information I will have to guess:
this is most likely NOT a libint problem. Doing a basic PBE0
calculation will cost you about 10 times more than a PBE calculation.
Now, if you didn't have enough memory (check the output) this will
be even longer. The settings of the HFX part are also crucial for performance
(screening).
regards
Juerg
--------------------------------------------------------------
Juerg Hutter Phone : ++41 44 635 4491
Institut für Chemie C FAX : ++41 44 635 6838
Universität Zürich E-mail: hut... at chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------
-----cp... at googlegroups.com wrote: -----To: cp2k <cp... at googlegroups.com>
From: Reza
Sent by: cp... at googlegroups.com
Date: 02/17/2017 07:44AM
Subject: [CP2K:8704] Libinit compilation
Dear Users,
I have compile libint and linked it in cp2k ssmp and psmp versions. I used ifort, icc, icpc of intel.
speed of GGA BLYP method for fullerene (C240 molecule in 30*30*30 box) is good, but when I use PBE0, a hybrid method, speed is very low.
Is it possible to compile libint with mpicc? I think libint should be very fast for HFX, but my performance is very low.
I used 6-31G** basis set for C atoms in C240.
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+... at googlegroups.com.
To post to this group, send email to cp... at googlegroups.com.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+... at googlegroups.com.
To post to this group, send email to cp... at googlegroups.com.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
[attachment "C240_6_31Gxx_SCF_Conv_blyp.inp" removed by Jürg Hutter/at/UZH]
[attachment "C240_6_31Gxx_SCF_Conv_blyp_cpu40_psmp_mpirun_Good_Perf.out" removed by Jürg Hutter/at/UZH]
[attachment "C240_6_31Gxx_2SCF_40cpu_ssmp_Bad_Perf.out" removed by Jürg Hutter/at/UZH]
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com.
To post to this group, send email to cp... at googlegroups.com.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
More information about the CP2K-user
mailing list