[CP2K-user] [CP2K:19930] Hybrids and double hybrids on LUMI
Sigbjorn Bore
sigbjorn.loland.bore at gmail.com
Mon Feb 19 08:55:50 UTC 2024
Thank both of you! I saw the 4c did better in one of the papers indeed.
Would you happen to have specific inputs I can try out? Are the ones in
benchmark (GitHub) and supplementary materials up to date?
PS. I have initiated a ticket to update cp2k LUMI.
On Monday, February 19, 2024 at 9:23:46 AM UTC+1 Augustin Bussy wrote:
> Hi Sigbjørn,
>
> I will add to Frederick's answer concerning the HFX part of your
> calculations. RI-HFX is only more performant for molecules and medium size
> solid state systems. For water, my experience has shown that using the
> initial 4-center implementation is always faster, especially if ADMM is
> also used.
>
> Best,
>
> Augustin
> On 2/18/24 17:47, Frederick Stein wrote:
>
> From out last tests, the high-scaling algorithm has a better performance
> up to 128 water molecules. For slightly larger systems, you should try
> both, for significantly larger systems (512 water molecules), the
> low-scaling algorithm is preferrable.
>
> PS: Soon, there will be an announcement for a workshop on these methods in
> a few weeks.
>
> Sigbjorn Bore schrieb am Sonntag, 18. Februar 2024 um 16:07:43 UTC+1:
>
>> Okay! Will try to reach out to lumi about the version, seems like gpu
>> part is changing rapidly, should not be too hard to update it.
>>
>> Definitely interested, I will try any specific suggestion and report back
>> on performance! I was thinking 64 water molecules for a start.
>>
>> On Sunday, February 18, 2024, Frederick Stein <f.s... at hzdr.de> wrote:
>>
>>> Dear Sigbjørn,
>>> From the manual, I can tell that the available version is not up-to-date
>>> (The latest release is 2024.1). If I were you, I would just test which
>>> setup is better. The k-point implementation is not compatible with
>>> post-Hartree-Fock methods. It will depend on your systems whether the
>>> low-scaling or the high-scaling implementation of RPA or SOS-MP2 is faster
>>> and whether ADMM and/or RI-HFX is more suitable for your system (You may
>>> combine ADMM with RI-HFX if seems reasonable to you). I can give you
>>> recommendations regarding the high-scaling methods, in case of the
>>> low-scaling methods and RI-HFX, you should consult Augustin Bussy.
>>> Best,
>>> Frederick
>>>
>>> Sigbjorn Bore schrieb am Sonntag, 18. Februar 2024 um 10:33:30 UTC+1:
>>>
>>>> Hi, first, thank you for this awesome software!
>>>>
>>>> I am planning to run a lot of water single point calculations (forces
>>>> and energies) for water on LUMI-G. I am currently using this installation
>>>> https://lumi-supercomputer.github.io/LUMI-EasyBuild-docs/c/CP2K/#user-documentation.
>>>> This recommends a slightly different setup (6 openMP vs 3 in
>>>> https://pubs.aip.org/aip/jcp/article/158/16/164109/2886896, which I
>>>> understand was done during pilot phase). Is this installation up to date,
>>>> and what is generally the best jobsetup currently?
>>>>
>>>> Then, to get best performance for hybrids and double hybrids on LUMI,
>>>> do you have any specific recommendations? From
>>>> https://pubs.aip.org/aip/jcp/article/160/6/064116/3263749,
>>>> https://pubs.aip.org/aip/jcp/article/160/6/064116/3263749,
>>>> https://pubs.aip.org/aip/jcp/article/158/16/164109/2886896, and rapid
>>>> developments I am little confused on what I should be running to get
>>>> reasonable numerical accuracy (at each level of theory) for lowest amount
>>>> of cost. So far it seems standard input file (4c) for hybrid functionals
>>>> and use the SOS of
>>>> https://pubs.aip.org/aip/jcp/article/158/16/164109/2886896 for RI-MP2.
>>>>
>>>>
>>>> All the best,
>>>> Sigbjørn
>>>>
>>>>
>>> --
>>> You received this message because you are subscribed to a topic in the
>>> Google Groups "cp2k" group.
>>> To unsubscribe from this topic, visit
>>> https://groups.google.com/d/topic/cp2k/CHsAz_QXXkE/unsubscribe.
>>> To unsubscribe from this group and all its topics, send an email to
>>> cp2k+uns... at googlegroups.com.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/cp2k/d5734839-2288-4c48-b3b9-50dbcfceee22n%40googlegroups.com
>>> <https://groups.google.com/d/msgid/cp2k/d5734839-2288-4c48-b3b9-50dbcfceee22n%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> .
>>>
>> --
>
> 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
> https://groups.google.com/d/msgid/cp2k/02fb4998-47e5-4267-ac99-9f1ba0ad2f34n%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/02fb4998-47e5-4267-ac99-9f1ba0ad2f34n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
>
--
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/3531d9fa-9875-4071-81d9-d2c03ea37869n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240219/825831e5/attachment.htm>
More information about the CP2K-user
mailing list