[Wien] WIEN2k benchmark

[Peter Blaha]

> I have run the benchmark on an SGI ALTIX (4 Itanium2 1.4 GHz processors)
> machine which we are planning to buy. It has ifc+mkl and for the
> sequential run, it gave a time of 126 secs (x lapw1 -c). Now we wish to

Was that compiled such that it really runs only on ONE processor ? or
does it automatically use more ?

> run the benchmark in parallel mode. The code has been compiled with the
> mpi libraries + ScaLapack and PBlas. I want to know whether it will run
> properly by giving the command (x lapw1_mpi -p -c) or we have to do the
> run init_lapw, followed by run_lapw -p -c for 1 iteration?

a) the executable is called   lapw1c_mpi
b) you need a file:   lapw1.def    (make it once with x lapw1 -c -d)

you can run this "by hand" using the appropriate command to execute
a mpi executable for this system (the company should know).

Often this is simply    mpirun -np 4 lapw1c_mpi lapw1.def
maybe you need a special "hosts" file.

c) Or, if WIEN is installed and properly configured (also here you must
tell siteconfig how to run an mpi job) you can use simply

x lapw1 -c -p     and put into the .machines files lines like

1:hostname:4       where hostname is the name of the machine you are running

Regards





                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------