[Pw_forum] QE pw-gpu help

[Lorenzo Paulatto]

On 03/20/2014 02:11 PM, 薛诗川 wrote:
> then, if i make these two parameter larger, could the process be 
> accelerated?
No, it won't be accelerated, it will just stop earlier.
> but how about its correctness?
It will the same, but you're be more far away from the relaxed atomic 
positions.
> and how about the value:ethr?
No such input variable.

> anyway ,how can i spped up the program with a correct output?

You cannot, if you could we would have done it. Anyway, here are some 
free tips:

When you ask for help, always post the output as well as the input, 
people may not have time to rerun your calculation, especially when it 
takes 12 hours!

You could try to pre-relax the system with less k-points and/or a lower 
cutoff, then refine the minima with full accuracy, but honestly I rarely 
found this approach to save any time.

Try to use only one pool, it's supposed to be better on desktop CPUs 
that do not have much RAM bandwidth.

Do you really need 300 electronic steps? If yes, there may be a problem 
with your system.

You can try to use TF or local-TF mixing, or increase mixing_ndim, 
especially if electronic minimization has some trouble converging.

Try without pot and wfc extrapolations, those are supposed to be used 
for dynamics and are not guaranteed to work for BFGS optimization.

> -- 
> Dr. Lorenzo Paulatto
> IdR @ IMPMC -- CNRS & Université Paris 6
> +33 (0)1 44 275 084 / skype: paulatz
> http://www-int.impmc.upmc.fr/~paulatto/
> 23-24/4é16 Boîte courrier 115, 4 place Jussieu 75252 Paris Cédex 05