when you do multiple run with “pre yes” as an argument, the memory increases dramatically. i attach the file, but the critical part is, in the melt example, with lj/cut/gpu as a pair style
run 1 pre yes post no
variable n loop 4000
label HERE
run 0 pre [yes|no] post [yes|no]
next n
jump SELF HERE
run 1 pre no post yes
when running this, the RAM goes up to 400 mb when the set is
pre yes post [yes|no]
while it keeps at ~70 mb when the set is
pre no post [yes|no]
when running with lj/cut without the gpu, the RAM is always around ~70 mb.
is this an expected issue?
in.gpu.y.y (538 Bytes)
when you do multiple run with "pre yes" as an argument, the memory
increases dramatically. i attach the file, but the critical part is, in the
melt example, with lj/cut/gpu as a pair style
run 1 pre yes post no
variable n loop 4000
label HERE
run 0 pre [yes|no] post [yes|no]
next n
jump SELF HERE
run 1 pre no post yes
when running this, the RAM goes up to 400 mb when the set is
pre yes post [yes|no]
while it keeps at ~70 mb when the set is
pre no post [yes|no]
when running with lj/cut without the gpu, the RAM is always around ~70 mb.
is this an expected issue?
not sure, could be just a bug in the memory accounting code. can you check
with some other tool whether the memory used is rising for real or just the
number that LAMMPS prints.
axel.
sorry, should have said that this is the value informed by the os, not by lammps.
axel, i don’t know if this is what you meant, but it’s the only memory info that lammps informs in one of the iterations in the loop variable:
sorry to bother again. i’ve been trying to find out what’s happening here, but i’ve hit a wall due to my lack of skills.
anyone had any breakthrough or ideas on where this leak (if so) is coming from?
Hi Pablo,
I cannot reproduce that behavior on my side with the 27May14 version.
Did you get the system memory usage by LAMMPS from another application like System Monitor (on Fedora) and Activity Monitor (on Mac)?
-Trung
the info i sent was the one informed by the os (through top) in debian. LAMMPS didn’t inform any particular problem.