On Wed, Jun 25, 2025, Silva Carvalho, Felipe via AMBER wrote:
>If I ran the serial version of Sander, the minimization finishes with
>no error or warning messages. However, if I ran the MPI version on 32
>processors I get the following message and the calculation is aborted:
>
> * NB pairs 336 304647 exceeds capacity ( 304964) 17
> SIZE OF NONBOND LIST = 304964
> SANDER BOMB in subroutine nonbond_list
> Non bond list overflow!
> check MAXPR in locmem.F90
>
>If I reduce the number of processors to 16, I get different messages:
>
>However, the calculation finishes and the final structure is the same as
>the one acquired with the serial version that does not display any of these
>messages.
> cut = 12
>The message about the exceeding number of non-bond interactions would
>suggest something wrong with my system, but the serial version does not
>complain about anything. Reducing the number of processors changes the
>message displayed (not complaining about exceeding number of non-bond
>interactions anymore) and it seems to run fine, because the final structure
>is the same as the one acquired with the serial version.
It's possible that reducing the cut value might affect things. Another idea
is to set ntmin=3 in sander (which gives a much better minimizer than the
default value of ntmin=1).
It's probably worth someone's time to try to track down the original problem
(with 32 nodes). But given that you have a workaround, that
minimization is usually(?) a short process that doesn't necessarily need
large numbers of nodes, that the problem *might* be specific to your
hardware, and that it might be tough to track down, I'm guessing that this
is may get fairly low priority. You can send me your input files if you
wish.
...thanks....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jun 28 2025 - 08:00:02 PDT
