Dear Bogdan, Different results actually imply something that differs by say, ~ 0.1 Ryd at convergence, not 4e-8 ... Your results are perfectly fine! You should expect differences when you change number of cpus, or the computer architecture.
It'd be great if the codes could always produce the same numerical answer with respect to the double precision accuracy, but I am afraid that'll never happen. Cheers, Kostya --- "B. Yanchitsky" <yan at im.imag.kiev.ua> wrote: > Dear users, > > There is a strange thing in parallel version of pw.x. > Parallel version running on 2 processors (cores) (dual-core > pentium-d, mpich, npool=2) > and the same pw.x code running on single processor (core) produce > different results. > As example > First iteration for 2 processors > total energy = -154.98769308 ryd > First iteration for 1 processor > total energy = -154.98786918 ryd > > > 2 processors > ! total energy = -155.02553035 ryd > 1 processor > ! total energy = -155.02553039 ryd > > Forces are also different > 2 processors > atom 61 type 1 force = -0.00042175 -0.00024350 > 0.00000000 > 1 processor > atom 61 type 1 force = -0.00042242 -0.00024388 > 0.00000000 > > > My configuration is ifort 9.0 + mkl 8.0.2. > Could you give any ideas/propositions? > Any additional info will be sent on your request from my side. > > Regards, > Bogdan Yanchitsky > > Institute of Magnetism > Vernadsky Blvd., 36-b > 03142 Kiev > UKRAINE > > Tel. (+380-44) 444 34 20 > Fax. (+380-44) 444 10 20 > _______________________________________________ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum > __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
