cmelab / polybinder

Initialization of thermoplastic polymer systems to simulate thermal welding
GNU General Public License v3.0
2 stars 5 forks source link

CUDA memory error when running large-ish systems #4

Closed chrisjonesBSU closed 4 years ago

chrisjonesBSU commented 4 years ago

I haven't done any bench marking to see at what system size the error occurs, but I am getting it with relatively small systems around 20,000 atoms.

I wonder if its related to using the hoomd_simulation.py method available within mbuild.

This is the current process:

Generate an mbuild compound of the system Using foyer to type this compound, we get a parmed structure Initialize a hoomd simulation from the parmed structure.

chrisjonesBSU commented 4 years ago

Reading this: https://groups.google.com/g/hoomd-users/c/nVSCQtcTCzc?pli=1

Some possible places to start looking:

So far, I've been using r_cut = 1.2

Also, the systems are initialized with a large box (calculated volume edge length * 5) to allow for PACKMOL to pack.

Looking at the simulations that did run, I think something is wrong/off with the calculation of box edge lengths as a function of density and system mass _calculate_L() in simulate.py. The final systems are very non-dense..

chrisjonesBSU commented 4 years ago

After testing a couple things out on a new branch, I think the memory error was related to having such a large initial box size. Now I'm dealing with particle-out-of-box errors, but I'll close this issue for now.