[QE-users] Number of cores and number of steps to convergence

[Zack Gainsforth]

Hi Folks,

I installed a new i9-13900k and I was just curious to see how the P cores and E cores compared.  I did an SCF of a phyllosilicate using each and I find that it converges nicely in all cases but that the number of iterations to reach convergence is different.  I’m just curious why.

Run on the 16 E cores only:

mpirun —cpu-set 8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23 —bind-to core -n 16 pw.x < Cronstedtite.scf | tee Cronstedtite.scf.16Ecore.out

Run on the 8 P cores only:

mpirun —cpu-set 0,1,2,3,4,5,6,7 —bind-to core -n 8 pw.x < Cronstedtite.scf | tee Cronstedtite.scf.8Pcore.out

Run on all 24 cores:

mpirun —bind-to core -n 24 pw.x < Cronstedtite.scf | tee Cronstedtite.scf.8P16Ecore.out

I find 1 hour for the E cores, 53 mins for the P cores, and 29 mins for all cores — which is practically linear scaling.

However, 22 iterations to converge on E cores, 26 to converge on P cores, and 15 to converge on all cores.  It looks like the more cores the fewer the number of convergence steps?  

(Input and output files and a spreadsheet attached if anyone is interested.)

Thanks,

Zack Gainsforth
Space Sciences Laboratory
University of California, Berkeley


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Cronstedtite.scf
Type: application/octet-stream
Size: 2597 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment.obj>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Cronstedtite.scf.8P16Ecore.out
Type: application/octet-stream
Size: 17561 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0001.obj>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0002.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Cronstedtite.scf.8Pcore.out
Type: application/octet-stream
Size: 20983 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0002.obj>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0003.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Cronstedtite.scf.16Ecore.out
Type: application/octet-stream
Size: 19723 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0003.obj>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0004.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: CalculationSpeed.ods
Type: application/vnd.oasis.opendocument.spreadsheet
Size: 13539 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment.ods>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20230301/07c37b88/attachment-0005.html>