CPU scaling for DMFT calculations

[Joshua Gray]

Hi everyone,

Has anyone come across issues with calculation time scaling with number of CPUs used? When I perform a calculation (such as the example calculation for SrVO3), the time taken to complete DMFT + CTQMC loops does not change much whether I use 4 CPUs or 48. Using 4 CPUs allows me to finish 6 DMFT iterations in ~ 5.5 hours, and using 48 CPUs completes the same number of loops in ~ 5 hours. I was expecting faster completion for 48 CPUs. Does there appear to be an issue with my parallelisation or is this expected behaviour for the types of systems we study? I have confirmed that it is using the correct number of CPUs as specified in the param_com.dat file.

Any help would be greatly appreciated.

Best,

Josh