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 |
Cronstedtite.scf
Description: Binary data
Cronstedtite.scf.8P16Ecore.out
Description: Binary data
Cronstedtite.scf.8Pcore.out
Description: Binary data
Cronstedtite.scf.16Ecore.out
Description: Binary data
CalculationSpeed.ods
Description: application/vnd.oasis.opendocument.spreadsheet
_______________________________________________ The Quantum ESPRESSO community stands by the Ukrainian people and expresses its concerns about the devastating effects that the Russian military offensive has on their country and on the free and peaceful scientific, cultural, and economic cooperation amongst peoples _______________________________________________ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users