TURBOMOLE Users Forum

TURBOMOLE Modules => Ricc2 => Topic started by: evgeniy on October 05, 2015, 03:32:25 PM

Title: Questions on large CCSD/EOM-CCSD calculations with TM
Post by: evgeniy on October 05, 2015, 03:32:25 PM

Hello,

I am interested in CCSD and EOM-CCSD calculations of a large systeme (~1800 MOs) with TM.
Is this theoretically possible with the newst version of TM? I guess, at present there is only
an SMP parallelization for CCSD/EOM-CCSD and only Abelian point groups can be used,
am I right?
Finally, what would be the best configuratrion for such calculations regarding the
maxcor memory? Are there some other important keywords/options for CCSD/EOM-CCSD
calculations with TM?

Many thanks!

Best regards,
Evgeniy

Title: Re: Questions on large CCSD/EOM-CCSD calculations with TM
Post by: christof.haettig on October 06, 2015, 03:26:14 PM

First of all, TM does not use the outdated EOM-CC formalism (EOM-CC is NOT size-consistent for transition moments and unrelaxed properties)!

All CCSD functionalities are only parallelized for SMP systems.
$maxcor should be set to something around 75% of the physical memory available in total for the CPU cores used for the computation. (Not the memory per thread or per core). If you want to use >> 16 GByte it is highly recommended to use
the "huge" version of TURBOMOLE (at least for the ccsdf12 program). There is essentially no performance difference between
the standard and the huge version for CCSD, but only the huge version uses 64bit addressing and can work with individual
arrays > 16 GByte.

Regards,
Christof

Note: you can now follow TURBOMOLE on Facebook at: https://www.facebook.com/TURBOMOLE

Title: Re: Questions on large CCSD/EOM-CCSD calculations with TM
Post by: evgeniy on October 16, 2015, 11:09:19 AM

Dear Christof,

Many thanks for your response and helpful points!

Best regards,
Evgeniy