[Wien] problem in parallel calculation

[Peter Blaha]

I'm only familiar with the Intel compiler and the mkl-library:

a) ifort+mkl will also on an AMD machine be much faster

b) parallelization via OMP_NUM_THREADS (on Intel mkl) works only for 2 
cores (maybe 4) in a reasonable ways. It is often a wrong believe that 
"more cores" are faster; but often even the opposite can be true. If 
your blas-labrary is good, 2 cores should reduce the time to about 1.5 
hours. One should in every parallelization make tests using 
1,2,4,8,16,... cores and compare the timing to find the optimal strategy.

c) You will have to use k-parallelism or mpi-parallelism, if you want to 
use your machine (see the UG about parallelization). However, usually 
such 48-core machines are VERY unbalanced between memory access, 
cpu-speed and sometimes disk-access and I would not expect too much.


Am 11.01.2014 17:50, schrieb hassan jamnejad:
> Dear Prof. Blaha
>
>    we have configured a multi-core Authentic AMD processor(48 CPUs),
> with a CentOS 6.4 installed on it and we are going to use it as an
> standalone machine and not in a cluster nor on a networked parallel
> machines. we have used to use wien2k installed on Core i7 cpus without
> any problem not in compiling nor on running it.
>
> surprisingly on this powerful multi-core machine when we try running the
> SCF cycle for any structure, although we have set the environment
> variable OMP_NUM_THREADS to 48 there is less LAPW1 threads created on
> the system and more importantly it will go through an unkown infinite
> loop on the first cycle even we have waited for a whole week and the
> LAPW1 on the first cycle did not end.
>
>    As a blind try to understand where the bug is, we set
> OMP_NUM_THREADS  to 1 and change nothing else the complete SCF cycle
> took 2 hours and we didn't have any problem.
>
>    The softwares installed are:
>
>        WIEN2k 13.1, ACML 5.3 ( we have tried with OpenBLAS v0.2.8-0 too)
> , GCC 4.9.0,
>
>    the compiler options that have been used for compiling WIEN2k is as
> bellow:
>
>                  Current settings:
>                      O  Compiler options:                -ffree-form
> -ffree-line-length-none -m64 -mfma4 -mavx2 -fopenmp -w
>                      L  Linker Flags:                        $(FOPT)
> -L/opt/acml5.3.1/gfortran64_
> fma4_mp/lib
>                      P  Preprocessor flags            '-DParallel'
>                      R  R_LIB (LAPACK+BLAS):    -lacml_mp
>
> I would appreciate if you can help me on finding and solving the problem.
>
> kindly regards
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at:  http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>

-- 
Peter Blaha
Inst.Materials Chemistry
TU Vienna
Getreidemarkt 9
A-1060 Vienna
Austria
+43-1-5880115671