[Pw_forum] parallel for single k point calculation

[sbraccia carlo]

Dear Mahmoud Payami,

of course you can.

If you run the parallel version of the program on N cpus, the code will
distribute the plane waves among the processors (this is called R&G
parallelization) so that both the memory and workload are distributed.
This is done automatically.

When you have k points and you want to parallelize also on k points, you
must run your job with the option -npool NK, where NK is some number
smaller than the total number of k-points and smaller than the number of
cpus.

Here some exaples, where I suppose that your job has 10 k-points. When
you have just 1 k-point (for example gamma), the case 1 is the only
possibility. Whenever possible a balance of the two parallelizations
(case 2) gives the best speedup.

1) without k-points (only R&G):

  mpirun -np 10 pw.x -in input > output


2) with k-points (10) and R&G :

  mpirun -np 10 pw.x -npool 2 -in input > output


3) with k-points (10) only (no R&G):

  mpirun -np 10 pw.x -npool 10 -in input > output


Best,

carlo sbraccia


Mahmoud Payami wrote:
> Dear All,
>  
> How can I run pw.x for a single k point in parallel? Is it possible?
> Thanks in advance.
>  
> Mahmoud Payami
>