how to improve CP2K scaling for small systems

Matt W MattWa... at gmail.com
Thu Mar 20 21:12:39 UTC 2014


Hi,

Final question, again dumb l(ike the memory/swapping earlier - you've 
definitely not linked to threaded libraries? If your MKL or FFTW3 are 
threaded then it could saturate the box...

Other than that, apart from the GAPW suggestion which, maybe could bring 
the data transfer below saturation point on your system. I think you need 
more processors...

Matt

On Thursday, March 20, 2014 12:15:06 PM UTC, labicia wrote:
>
> Hi!
>
> 1) Reducing the cutoff improves the speed but the scaling stays the same.
>
> 2) As you suggested, we will try the GAPW method.
>
> 3) We are using FFTs in double precision, is this a waste?
>
>  

> Thank you very much,
> Marco and Andrea
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140320/2c8c1402/attachment.htm>


More information about the CP2K-user mailing list