Molcas Forum

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 62

Memory requirements of SEWARD

Dear Users,

Is there a simple rule of a thumb (e.g. "10MB per basis function") to estimate peak memory requirements of SEWARD?

With such, there would be no trial and error search for optimal MOLCASMEM.

In some cases, if the memory limit is very low, the code does complain from the start.
However, sometimes it happens late in the calculations, resulting in a lot of wasted time.
(Some of my SEWARD runs take a day.)

Also, is there a precise way to do this estimate from the start, like, do all of the memory allocation calls before any actual work?

Thank you.
Andrew

valera

Administrator

Registered: 2015-11-03

Posts: 124

Re: Memory requirements of SEWARD

*if* you use conventional integrals, yes, such rule can be suggested. But [unless something has been changed recently] for CD and RI  the code trying to figure out the max available memory, and use all of it. This is why the routine computing MAXMEM cheat, and return a bit smaller value. And I can imagine how F90 allocates helps to use the memory outside the molcas memory manager.
So, I would suggest a very simple trick - locate the place where MAXMEM is computed, and reduce the number of available memory. CD/RI will consume a bit less, it will (not significantly) reduce the speed, but it will leave more memory for other use, so the code stop crashing.

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 62

Re: Memory requirements of SEWARD

Thank you very much. I'll try to implement such a hack. Also, sorry for the delayed answer.

Last edited by andrewshyichuk (2023-09-12 15:02:47)