[CP2K-user] [CP2K:20931] Re: Slow AIMD Calculations and Memory Issues

[Frederick Stein]

Dear Jawad,
I checked your timings. The time per SCF iteration is acceptable to me. The 
issue is that your system converges quite slowly in the very first step. To 
improve convergence, you can
- reduce MAX_SCF in the inner convergence loop. This rebuilds the 
preconditioner more often and may help with convergence.
- try FULL_ALL for preconditioning
- try a different minimizer (DIIS or BROYDEN instead of CG) 
- adjust ENERGY_GAP to a value lower than your actual gap. Too low value 
slow down convergence, too high values prevent convergence.
Did you check whether the numerical stress tensors are actually faster than 
analytical stress tensors?
Is there a reason why you work with CP2K 8.2?
Regarding you MPI error: This seems to be related to your MPI 
implementation, NOT to CP2K. What MPI distribution do you use (library 
name+version)?
Best,
Frederick


Jawad Elhamdaoui schrieb am Donnerstag, 28. November 2024 um 11:09:16 UTC+1:

> Dear Frederick,
>
> Thank you for your prompt response.
>
> Please find attached the output file. Regarding the input, I used the same 
> file I uploaded earlier, but I commented out the cell part to restart the 
> calculation and add the restart part.
>
>
> Please let me know if you need any further details.
>
> Best regards
>
> Jawad 
> On Thu, Nov 28, 2024 at 11:02 AM Frederick Stein <f.s... at hzdr.de> wrote:
>
>> There is no output file. You can remove all steps except from one. In 
>> doubt, compress it or share a link to the output file.
>>
>> Jawad Elhamdaoui schrieb am Donnerstag, 28. November 2024 um 10:59:46 
>> UTC+1:
>>
>>>  Dear Frederick,
>>>
>>> Thank you for your response.
>>>
>>> I hope that the output file is uploaded correctly. It says that it 
>>> exceeds the 8 Mb limit, so I removed some in the middle.
>>>
>>>
>>> Please let me know if you need any further details.
>>>
>>> Best regards
>>>
>>> Jawad
>>>
>>> On Wed, Nov 27, 2024 at 3:13 PM Jawad Elhamdaoui <jawadelh... at gmail.com> 
>>> wrote:
>>>
>>>> Dear Frederick,
>>>>
>>>> Thank you for your prompt response.
>>>>
>>>> Please find attached the output file. Regarding the input, I used the 
>>>> same file I uploaded earlier, but I commented out the cell part to restart 
>>>> the calculationa and add the restart part.
>>>>
>>>> Please let me know if you need any further details.
>>>>
>>>> Best regards
>>>>
>>>> Jawad 
>>>>
>>>> On Wed, Nov 27, 2024 at 2:52 PM Frederick Stein <f.s... at hzdr.de> wrote:
>>>>
>>>>> Dear Jawad,
>>>>> Without any output file, it is hard to guess what the issue is. Could 
>>>>> you maybe run a short MD (a few frames) and provide the output file? I am 
>>>>> especially interested in the statistics part at the end of the output file 
>>>>> (Timing report, "MEMORY| Estimated peak process memory [MiB]"). Please 
>>>>> provide the full input file you used (not necessarily the geometry files) 
>>>>> because you commented out the cell definition in your input file.
>>>>> Best,
>>>>> Frederick
>>>>>
>>>>> Jawad Elhamdaoui schrieb am Mittwoch, 27. November 2024 um 14:26:48 
>>>>> UTC+1:
>>>>>
>>>>>> Dear CP2K Community,
>>>>>>
>>>>>> I am facing issues with my AIMD calculations. My *system size is *162 
>>>>>> atoms, and I run on an *HPC setup of* 500 CPUs. Each step takes ~15 
>>>>>> minutes, and I also encounter out-of-memory errors.
>>>>>>
>>>>>> I suspect the issues might be due to input settings or 
>>>>>> parallelization inefficiencies. Could you please advise how to optimize 
>>>>>> performance and memory usage for a system of this size?
>>>>>>
>>>>>> I appreciate your help!
>>>>>>
>>>>>>   Best regards,
>>>>>>
>>>>>> Jawad
>>>>>>
>>>>> -- 
>>>>> 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 visit 
>>>>> https://groups.google.com/d/msgid/cp2k/463afbde-75aa-4338-9eeb-3f4d64cd4227n%40googlegroups.com 
>>>>> <https://groups.google.com/d/msgid/cp2k/463afbde-75aa-4338-9eeb-3f4d64cd4227n%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 visit 
>> https://groups.google.com/d/msgid/cp2k/ade71ae0-4617-4824-adc9-4c782057c7f6n%40googlegroups.com 
>> <https://groups.google.com/d/msgid/cp2k/ade71ae0-4617-4824-adc9-4c782057c7f6n%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 visit https://groups.google.com/d/msgid/cp2k/377953f3-58ba-4a92-8765-2d2c276a0212n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20241128/7bf41808/attachment.htm>