Slow calculation speed when using MEAM potential

Dear LAMMPS users,

I would like to ask about slow calculation speed of LAMMPS when using meam potential.
I am using different potential files (eam/fs and meam) to calculate melting temperatures of structures in the Ni-Ti system.
NPT was used to control the temperature of the system, and around 11000 atoms were constructed.

The calculation speeds when using different potential files were as following;
1. eam/fs potential : 181 steps per second
2. meam potential : 6 steps per second

Is it normal for simulations using meam potential has very slow calculation speed (30 times slower)?

Thanks in advance,

Dear LAMMPS users,

I would like to ask about slow calculation speed of LAMMPS when using meam
potential.
I am using different potential files (eam/fs and meam) to calculate
melting temperatures of structures in the Ni-Ti system.
NPT was used to control the temperature of the system, and around 11000
atoms were constructed.

The calculation speeds when using different potential files were as
following;
1. eam/fs potential : 181 steps per second
2. meam potential : 6 steps per second

Is it normal for simulations using meam potential has very slow
calculation speed (30 times slower)?

​what version of LAMMPS exactly are you using and which fortran compiler
was used to compile the meam library with which flags?

according to the benchmarks posted here
http://lammps.sandia.gov/bench.html#potentials there is a significant cost
to using MEAM over EAM, but not quite as much as you report. it should be
below a factor of 10, more like 7-8x.

axel.​

Dear Axel,

16Feb16 is the version of LAMMPS being used, and was compiled using Makefile.mpi with the mpicxx compiler (flags activated are -g and -O3).

For the MEAM library, gfortran with the flags [-O -fPIC -fno-second-underscore] is used. Do you believe compiling -O2 or -O3 rather than -O is necessary?

Thanks,

Dear Axel,

16Feb16 is the version of LAMMPS being used, and was compiled using
Makefile.mpi with the mpicxx compiler (flags activated are -g and -O3).

For the MEAM library, gfortran with the flags [-O -fPIC
-fno-second-underscore] is used. Do you believe compiling -O2 or -O3
rather than -O is necessary?

for gfortran you should upgrade to a newer version of LAMMPS (best the 14
May 2016 version). there is a bad interaction between the gfortran compiler
and the glibc math library, that causes a significant slowdown (~2-3x). we
have recently added code to the lib/meam code to remove this restriction.

the alternative approach would be to use the intel fortran compiler (which
doesn't use the gnu math library for time critical mathematical functions).

axel.

Dear Axel,

I downloaded the most recent version of LAMMPS, and compiled the another lmp_mpi executable file.

Unfortunately, the calculation speed is the same, 6 steps/sec for meam potential calculation while it is 180 steps/sec for eam/fs potential calculation.

A gfortran compiler was used in lib/meam. Is there any other solutions for this issue?

Thanks in advance,

Dear Axel,

I downloaded the most recent version of LAMMPS, and compiled the another
lmp_mpi executable file.

​please note, that e-mails from this mailing list are archived so that
people with similar problems in the future can search for those were solved
in the past. for that it would be much more helpful to state which version
exactly you downloaded. ​what is "latest" can change any day. thanks.

Unfortunately, the calculation speed is the same, 6 steps/sec for meam
potential calculation while it is 180 steps/sec for eam/fs potential
calculation.

A gfortran compiler was used in lib/meam. Is there any other solutions for
this issue?

​there definitely should be some speed difference between the two LAMMPS
versions​. quite significant on most machines.
but it also has already been pointed out, that MEAM is significantly slower
than EAM, so don't expect any miracles.

at this point, i would assume that you have made a mistake somewhere, but
with so limited information, it is impossible to say anything for certain.

i may be able to do a little test/comparison, if you can provide a complete
set of inputs (potential files, data files inputs etc.) and log files from
your runs, so that i can compare to. it would also be helpful to know what
kind of CPU/OS platform you are using.

axel.