General queries on Lammps capabilities and benchmark study!

Dear Lammps users,
As I am in the process of making some tutorial to the new lammps or MD user, I need some facts regarding lammps capabilities:

  1. How does lammps scale with the simulation domain? any bench marking article or report should help.
  2. Irrespective of the study type, what is the largest simulation volume being modeled by lammps?
  3. Where do I find reference to those (large) simulations, where they clearly state the data like no of atoms, length, time scale and processors being used?

I really appreciate the input from the expert users.

Thanks in advance.

Dear Lammps users,
As I am in the process of making some tutorial to the new lammps or MD user,
I need some facts regarding lammps capabilities:
1. How does lammps scale with the simulation domain? any bench marking
article or report should help.

LAMMPS scaling and performance is *very* dependent on the features
used. in principle LAMMPS has almost linear weak scaling for simple
models with only short range potentials (like a good MD code should
have). there are some benchmarks and discussions on the LAMMPS home
page.

2. Irrespective of the study type, what is the largest simulation volume
being modeled by lammps?

there is no answer to that. for some models, then limit is then
computer. since LAMMPS can do not only atomistic system, but also all
kinds of coarse grain, discrete element and lattice-like
representations it can cover an extremely wide range of length-scales.

i suggest you some more research on the LAMMPS homepage and look
especially also at presentations from the LAMMPS user's workshops,
where people report their findings.

3. Where do I find reference to those (large) simulations, where they
clearly state the data like no of atoms, length, time scale and processors
being used?

in the published literature. classifying simulations by their size
alone is generally a mistake. it is easy to extend systems, but it is
not so easy to extract useful information that cannot be obtained from
smaller systems. as of recent, LAMMPS support compilation with 64-bit
counters in many places, so the theoretically possible size of
systems, even if they are not simple atomic system is far beyond what
a future exascale supercomputer could handle within a reasonable
amount of time (i.e. until it becomes obsolete).

even though LAMMPS was conceived in the USA, the developers are
smarter than the usual "bigger is better" paradigm, even though LAMMPS
is deliberately aimed at supporting very large calculations.

I really appreciate the input from the expert users.

would you mind explaining, how your quest for the largest possible
LAMMPS simulation relates to a new(!) user tutorial? a new user is
generally struggling with understanding MD at all, and thus such a
user should never ever have access to a resource that allows running
as large a simulation that it reaches the limitations of LAMMPS. such
simulations require not just some but a *lot* of experience and care.

axel.