On Fri, Jun 09, 2023, 刘野 via AMBER wrote:
>1) What is the maximum limitation of the number of atoms that tleap can
>build? We tried to build a simulation system of 6 million atoms previously
>and got the following error message:
>Building topology.
>!FATAL ERROR----------------------------------------
>!FATAL: In file [/home/AMBER22/amber22_src/AmberTools/src/leap/src/leap/varArray.c], line 371
>!FATAL: Message: Realloc: Cannot allocate memory!
>!ABORTING.
I don't know if this represents a hard limit in tleap, or if something you
are doing requires extra amounts of memory. I've built systems with 3.5
million atoms without seeing this. You could experiment with smaller
systems, to see when/if the error goes away.
It might be good to (a) keep your system as simple as possible, just as a
test; (b) post the tleap input file you are using, just in case something
pops up as being odd.
>2) What is the maximum box length that tleap can build?
I don't think there is any maximum.
>3) What is the maximum limitation of the number of atoms that pmemd.cuda
>can simulate using A100 GPU, and what is the corresponding calculation
>speed?
I hope some one with experience on big systems may chime in here. Amber has
certainly been rarely tested with 20 million atoms: it might work, but it
outside the design scope.
>We also noticed that there is a STMV simulation system of 1.06 million
>atoms on the AMBER official website, and we want to simulate a system of 20
>million atoms. Do you have any suggestions?
You can look at size vs performance numbers on the GPU benchmark page, and
do some extrapolation. Time is very roughtly linearly proportional to
number of atoms, but memory constraints are much more likely to be limiting.
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Jun 20 2023 - 15:30:02 PDT
