Comparison between CP2K and Quantum Espresso
tfran... at gmail.com
tfran... at gmail.com
Wed Feb 8 10:28:48 UTC 2017
first of all thanks a lot for your very useful comments.
I will benchmark my system in CP2K with a lower basis sets, in order to
test again the parameters and to have better reference with respect to QE.
Regarding the compilation of the program in our cluster, I think that it is
the pitfall and that our technician missed something. He told me ( because
I am not allowed to install anything, but only the technician ) that he
compiled the precompiled version of CP2K. But I am pretty sure he missed
some of the libraries you listed above. Indeed I am insisting in a more
"serious" compilation of the program, that we wanna massively exploit for
Regarding the single point shift from the gamma point in QE, that was
suggested by one of my old Professors. At that time ( almost three years
ago) he told us that the shift is recommended.
I always used that setting, without honestly taking care too much about the
change in performances. By the way, thanks a lot for the really useful
hints. Now I will re-benchemark the new machine testing several levels of
theory as you suggested.
Il giorno mercoledì 8 febbraio 2017 11:09:16 UTC+1, David T ha scritto:
> Hi Tommaso
> as you I am more expert on QE and only recently moved on CP2K.
> On my experience CP2K is much more quicker and allows to work with bigger
> systems that planewave code could not afford. On the other hand it is true
> that this is really system depended so the fact that for nano-porous
> material CP2K is more efficient could not be true for other systems.
> After having read the inputs a comment I can give to you is that you are
> probably not making a fair comparison.
> I've see you use ultrasoft pseudo which allows a very low cut-off and can
> be "though" as minimal basis-set. So you should probably use some
> corresponding low basis-set in CP2K (for instance a dobule zera instead of
> triple zeta).
> Probably more important another thing I can tell you is that I've seen
> that if CP2K is not properly compiled, its performance can be slow. For
> instance my own version with mpi, mkl, libxsmm and elpa is about 30% faster
> than the standard one I found on our cluster.
> Moreover the speediness of the code can be further boosted if you use also
> GPU and hybrid openMP-MPI.
> P.S. a curiosity, in QE why are you using a single point shifted from
> gamma? if there is not a major reason using the gamma algorithm on QE can
> accelerate you calc up to 30%
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the CP2K-user