I am doing the computation of a perovskite system of 20 atoms in a supercell. I know that options like NPAR and KPAR, and the numbers of cores used for computations can greatly affect the efficiency, but are there some general principles for the optimal setting? (e.g. for what kinds of systems, how will KPAR affect the computational efficiency and how do I optimize all of them)

  • $\begingroup$ It depends on computing machine. General rule of thumb, use NCORE instead of NPAR. Set value to 4,8,12 up to Max number of process to see any increase in performance. KPAR is kpoint parallelization. If running on multiple node use KPAR equal to number of node. $\endgroup$ Jun 8 at 15:47


You must log in to answer this question.