You have to do some timing ! Check the time (eg. of the benchmark on the wien2k page) with OMP_NUM_THREAD =1,2,4 and maybe 8. You will see, that the gain with 4 is already very small and thus more or less useless.
Many people think that using more cores makes a calculation automatically faster. This is not true. Very often it can happen that more cores need LONGER time. (This is also true for k-parallel wien2k calculations or mpi-runs, where you have to find out by timings (in the dayfile or stdout) how far you can go). At least at the beginning you have to find out what the optimal number of cores is for your specific case. This number will vary from case to case (maybe only 2 cores for fcc Al using OMP_NUM_TREAD), to 2000 cores (mpi) for a supercell with 1000 atoms ....), and after some time you will have enough experience to select the proper resources and method (OMP, k-parallel, mpi, often a combination of 2 of the methods (OMP+k-point or k-point+mpi) for a new case. PS: My experience with these huge shared memory machines (more than 16 cores) is NOT very good. Their performance is usually quite lousy, probably because of memory-bus limitations. Maybe core-binding helps. Am 05.06.2013 01:37, schrieb Jose Alfredo Camargo Martinez:
Hi. We install Wien2k (serial mode) in a server with 64 processors (Workstation AMD Opteron 6328 4X16 ). In bashrc we defined OMP_NUM_THREAD=58. When we run "run_lapw" (Kmax=7 and N° of k-points=2000), only 33 processors running. Why it doesn't run in 58 processors? J. Camargo. _______________________________________________ Wien mailing list [email protected] http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien SEARCH the MAILING-LIST at: http://www.mail-archive.com/[email protected]/index.html
-- ----------------------------------------- Peter Blaha Inst. Materials Chemistry, TU Vienna Getreidemarkt 9, A-1060 Vienna, Austria Tel: +43-1-5880115671 Fax: +43-1-5880115698 email: [email protected] ----------------------------------------- _______________________________________________ Wien mailing list [email protected] http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien SEARCH the MAILING-LIST at: http://www.mail-archive.com/[email protected]/index.html
