[Pw_forum] Computational Speed for pw.x

[Lorenzo Paulatto]

On 11/12/17 17:53, Amar Singh wrote:
> ​1. I found somewhere that open-mpi may provide parallelization upto few 
> processors, but not for a large number of them. If so, could MPI be a 
> better option.

QE can use both, if you cannot scale over a certain number of CPUs with 
MPI you can take care of compiling with OpenMP support and see if 
setting OMP_NUM_THREADS to something large than one gives you a bit more 
scaling. You have to set it by doing
   export OMP_NUM_THREADS=N
(N is the number of OpenMP threads, I would try 2 or 4, hardly more)
before running pw.x, and take care that the product of N times the 
number of MPI processes is less than the number of CPUs on your computer.


> ​2. I do not have Intel MKL on my system, could this be the reason for 
> slow processing speed with open-mpi.

Even if Intel compiler is very expensive these days, I think it is still 
worth buying it, as it is likely still much cheaper than the 40 CPUs 
computer, and the performance gain with respect to gfortran can be 
important. Even with respect to other proprietary compilers I usually 
found ifort to be faster, but I did not test it much.


hth

> thanks
> ​Amar
> 
> 
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
> 

-- 
Lorenzo Paulatto - Paris