# [CP2K-user] [CP2K:17739] Need help to increase compute speed

Lucas Lodeiro elunicolomo at gmail.com
Thu Sep 22 19:24:30 UTC 2022

```Hi Dev Rana,

You are usign the GPW method, so you need and your are using
pseudopotentials for the atoms. In the KIND section of each atom class, you
select the POTENTIAL i.e. the pseudo potential. The number after the "q"
means the number of valence electrons explicitly computed. So for Cu is
q-11 and for C is q-4: 62*11 + 4 = 686... this information also is printed
in the output, before the SCF cycles, where you can check this,

Regards - Lucas

El jue, 22 sept 2022 a las 15:09, Dev Rana (<dev.rana3 at gmail.com>) escribió:

> Hi lucas,
>
> Thank you for this. This definitely speeds things up.
>
> I feel like I'm missing something in your explanation. You say that 62 Cu
> + 1 C = 686 electrons total. Cu has 29 electrons * 62 Cu atoms = 1798. 1 C
> = 6 electrons. So This should be 1802. With roughly half occupied = 901 and
> 20% of that is still roughly 200. I agree that 200 is too much - but how
> you obtained 70 is a bit confusing to me.
>
> Could you explain how you reached your conclusion?
>
> Thanks!
>
> On Monday, July 25, 2022 at 7:37:56 PM UTC-4 Lucas Lodeiro wrote:
>
>> Hello Rev Rana,
>>
>> From your input, I see 2 thing that could improve your calculation. First
>> you use a CUTOFF of 600 and a REL_CUTOFF of 250. This is a little weird to
>> me... usually REL_CUTOFF is around 60.... using higher values will slow
>> down your calculation without a real improvement over the calculation
>> results. Do you do a convergence study of these parameters?
>> Second, you are adding 200 MOs to your calculation... Obviously you need
>> some virtual orbitals to use the smearing, but the greater the number of
>> these, the slower the calculation is. So, 200 is in my opinion too much.
>> Your system has 62 Cu and 1 C atoms, with 686 electrons in total, this is
>> 343 occupied orbitals (only as a mental exercise). Usually adding 20% more
>> orbitals is enough. so adding just 70 orbitals would be sufficient, and
>> make your calculations faster.
>>
>> Regards - Lucas
>>
>>
>> El lun, 25 jul 2022 a las 20:01, Dev Rana (<dev.... at gmail.com>) escribió:
>>
>>> Hello,
>>>
>>> I'm having some trouble conducting an MD simulation. Each SCF iteration
>>> takes about 100-200 seconds taking about 15-30 iterations per MD step.
>>> Could someone experienced take a look at my input file to see if there is
>>> any input which could be added/removed to increase speed? I've attached my
>>> restart file containing all information.
>>>
>>> I'm using a computing node with openMPI OMP threads set to 2, and 64
>>> cores on a single exclusive node (no multi-node delays).
>>>
>>> Best Regards,
>>> Devyesh Rana
>>>
>>> --
>>>
>> 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 view this discussion on the web visit
>>> .
>>>
>> --
> 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+unsubscribe at googlegroups.com.
> To view this discussion on the web visit
> .
>

--
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+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/CAOFT4PJEW1zTdALMZ29XXrPTQ0zXja2AWRFTiRPTv%3DhHgfamdw%40mail.gmail.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20220922/8d4d5eb4/attachment.htm>
```