[Pw_forum] QE pw-gpu help
akohlmey at gmail.com
Fri Mar 21 13:42:25 CET 2014
On Fri, Mar 21, 2014 at 8:35 AM, 薛诗川 <nudt.xueshichuan at gmail.com> wrote:
> To sum up:
> there are some parameters that i can adjust to help the convergence process.
> 1. etot_conv_thr , forc_conv_thr .
> they are threshold values. so if they are larger, the process is easy
> to converge?
> 2. ecutwfc, ecutrho
> they are cutoff values. what are their functions? is it better when
> they are smaller?
> 3. electron_maxstep
> it means the max number of iteration. 100 is enough.
> 4. conv_thr
> what does this mean?is it the same with etot_conv_thr ,
> forc_conv_thr, the larger the faster to stop?
> 5. npool
> and dear pro, i am a freshman actually. and i don't know the detail of its
> calculation, or to say i don't need to know the details but i should have a
> fundamental understanding of its calculation. i should have known about its
> physic principles, its heaviest calculation part so i can change these
> important values to make it faster.
> looking forward to your reply.
very simple. don't use a GPU unless you can answer all of the
questions above by yourself.
compile a CPU (that is "CeePeeYouu") version of the executable and
work through the significant available tutorial material and start
working on improving your understanding.
people understand that you are new to this, but please keep in mind
that a mailing list is not replacement for an adviser and learning
understanding the basics first. insisting to ask the same questions
over and over until people will satisfy your personal expectations
will only lead to you frustrating others and using up time that could
be spend more productive on *both* sides.
> thank u again.
> 2014-03-21 17:17 GMT+08:00 薛诗川 <nudt.xueshichuan at gmail.com>:
>> Thank you very very much for your reply.
>> i still have some problems.
>> i have tested an input file from "espresso-4.3.2-examples.tar" which
>> contains a lot of input files and ref files.Specifically, i tested the
>> following input file with my pw-gpu.x/
>> however it still ran a long time without an ending so i killed it.
>> but when i check the ref file, i found that test time is only a few
>> i am totally confused? is the test machine's performance so good?
>> here is my test file and ref file:
> Pw_forum mailing list
> Pw_forum at pwscf.org
Dr. Axel Kohlmeyer akohlmey at gmail.com http://goo.gl/1wk0
College of Science & Technology, Temple University, Philadelphia PA, USA
International Centre for Theoretical Physics, Trieste. Italy.
More information about the users