Hello everybody,
assume we have a liggghts simulation with only 1000-2000 particles (single spheres) and a timestep size in the order of X^(-6) seconds. We have to simulate a fixed time period of 8 seconds.
I read in this forum that parallelization will speed up simulations when there are approx. 10,000 particles per CPU. However, I do not think it will speed up the above simulation, where the timestep size is the limiting factor, will it?
Except from altering the material parameters, are there any tricks or hints on how the simulation could be made to finish faster?
Kind regards,
MiRa
NTT1508 | Thu, 04/21/2016 - 09:23
Not significant
If you take time to look at previous discussions in this forum, you will find that running in parallel does not work effectively for the case where the number of particles under 10,000. Many factors to accelerate the speed of simulation, e.g., reduce skin, reduce the size of domain, use reasonable stiffness for materials...