[Pw_forum] Computational Speed for pw.x

[Lorenzo Paulatto]

On 07/12/17 08:15, Amar Singh wrote:
> Dear Friends,
> ​I am trying to vc-relax a 40 atom supercell using a 40 processor/256GB 
> RAM (Dell 7910) computer equipped with open-mpi. Following is the 
> command I used to run the QE
> ​mpirun pw.x -np 40 < XXX.in > XXX.out
> 
> ​​I noticed that the processing speed is slightly better than single 
> processor, but nowhere close to expected 30 - 40 times. Also the dynamic 
> RAM allocated per process is ~ 950 MB (total ~ 39 GB), the rest > 210 GB 
> remains unused.
Dear Amar,
you do not give any information about the kind of calculation you are 
running, size, number of k-points, hybrid functional, etc. You also 
don't say if you did some proper scaling test (i.e. running with 
1,2,4,8,16,32,40 cpus) to see if the code scales properly up to a 
certain size.


Lacking any information, all I can say is: try to increase the number of 
pools.

kind regards

-- 
Lorenzo Paulatto - Paris