[Pw_forum] pw.1.3.0 on Linux cluster (mpirun)

Goranka Bilalbegovic goranka.bilalbegovic at zg.hinet.hr
Sat Nov 15 13:26:42 CET 2003


The compiler for pw.1.3.0 is PGI 4.1.2 and it was ifc 7.1.008 for pw.1.2.0.
I forgot that I changed  the compiler because of the ifc "compiler internal
error, please report to Intel corporation..." I do not have ifc 6.*, and
would be happy to know which version of ifc 7.1.*  (if any) compiles
pw.1.3.0.

Up to know I have tested PGI compiled pw.1.3.0 with 2 (same node) and 8
processors (4x2). The number of processors jobs use is always correct.

When I use PARA_PREFIX= mpirun -np #, job starts, the files in out_dir are
not formed, calculation goes on (i.e., jobs only eat time), perhaps forever,
no error massages, output files are empty.

For PARA_PREFIX= mpirun -n #,  the file job.save forms, but
(i) for 2 processors on the same node it runs forever, pw does not start to
write
(ii) for 8 processors jobs stops after writing:
----------------------------------------------
Current dimensions of program pwscf are:
ntypx =10   npk =40000  lmax = 3
nchix = 6  ndim = 2000  nbrx = 8 nqfm = 8
-----------------------------------------------
In both cases (with PARA_PREFIX= mpirun -n #) there is a message:
-----------------------------------------------
Warning: Command line arguments for program should be given
after the program name.  Assuming that 2 is a
command line argument for the program.
Unrecognized argument -n ignored.
------------------------------------------------

Unfortunately, I  do not have the same type of test jobs for 2 and 8
processors, and  the cluster is now very busy.

Thanks for help.
Best regards,
Goranka










More information about the users mailing list