Dear Paul, On Jan 22, 2014, at 8:05 PM, Paul.Caheny at uk.fujitsu.com wrote: > Any other information about the purpose/applicability/trade-offs involved in > this option would also be very useful.
It works if you run K-point calculations and each pool has a size of 1 (meaning "-npool N" where N is equal of the number of MPI processes). In this way you can collapse FFT calculations to be totally local within a MPI process. In such a way, z-stick are distributed in a serial fashion and you can call any 3D FFT serial library. More k-point you have, more N can grow. But it is obvious where the limit is. There are trade-off for sure, I noticed that by reducing communication intra_pool the communication between pools becomes more intensive. I have no numbers to show but I can easily run few tests using QE-GPU. I am not aware about other special cases. HTH F -- Mr. Filippo SPIGA, M.Sc. Quantum ESPRESSO Foundation http://www.linkedin.com/in/filippospiga ~ skype: filippo.spiga ?Nobody will drive us out of Cantor's paradise.? ~ David Hilbert ***** Disclaimer: "Please note this message and any attachments are CONFIDENTIAL and may be privileged or otherwise protected from disclosure. The contents are not to be disclosed to anyone other than the addressee. Unauthorized recipients are requested to preserve this confidentiality and to advise the sender immediately of any error in transmission." -------------- next part -------------- An HTML attachment was scrubbed... URL: http://pwscf.org/pipermail/pw_forum/attachments/20140125/5993b704/attachment.html
