[AMBER] Memory usage of NMR center of mass restreaints on GPUs

From: paul <paul_westphaelinger.ewetel.net>
Date: Tue, 21 Feb 2017 14:33:15 +0100

Hello dear members of the AMBER mailing list,


I have recently been forced to move from AMBER 14.12 to AMBER 16.

My simulations started crashing and I soon realized that a dramatic
increase in memory consumption was at fault.

I am using the NMR center of mass distance restraints and after the
version change a four window replica system jumped in

memory usage from ca. 900MB to 2800MB, no other parameters were changed.
I suspect this is due to a change that happened in AMBER 14.13. as I
think I had these

same problems about a year ago after that update without realizing the
problem was the increase in memory usage.


A colleague did further testing and compared the NMR restraint
simulation to one using no restraints and one using the new NFE
restraints with the following results:

pmd
only pmd with CoM-Distance and CoM-Dihedral: 755MiB memory

md
no restraints at all: 755MiB memory

bbmd
pmd and bbmd in 2x2 replica: 839MiB memory/replica

NMR
only CoM-Distance with disang file: 1642MiB memory


The difference from using CoM and diehedral restraints (755MiB) to the
NMR (1642MiB) is a mystery to me. I would like to use the NMR restraints
because of other drawbacks of the new NFE restraints.

If you have any idea what is going on or if this might be a bug, I would
love to talk to you.


Greetings,

Paul Westphälinger


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Feb 21 2017 - 06:00:03 PST
Custom Search