Specs for server running LIGGGHTS

Submitted by jpdorsch on Wed, 12/20/2017 - 15:35

Hi everyone. I'm attempting to simulate some granular problems (1~10M number of particles) , mostly silo storage kind-of.

So, I have some questions about the server (or compute node) that I should use for that amount of particles.

Which specs should I use in my compute node?

* Memory (seems not so important, so I think 64 GB could be OK.)
* Processors (Intel Gold 10 cores dual socket?)
* Storage (10 TB?)
* GPU?

Thanks in advance, any advice would be really helpful!

Juan.

richti83's picture

richti83 | Fri, 02/02/2018 - 12:30

Hi Juan,
from my point of view it depends on the problem size. Normally 1M particles is not a big problem for a good Workstation. We own some DELL T7910 each with:
- DUAL E5-2687W v4 @ 3.00GHz (=24 physical cores)
- 64GB RAM (ParaView needs some more for Postprozess than LIGGGHTS itself)
- 256GB SSD for operating system
- 2x1200GB SAS 10K drive for simulation data in raid0
- backup on Synology DS416 NAS with 12TB
For really big problems (like 10M particles) I would prefer a cluster (our university owns 100 nodes each 16cores, but most of the time there is no capacity for my problems left)
Keep in mind that communication costs become higher when more cores are involved. A rule of thumb is that each core should own 10.000 particles (but of course this depends on geometry, moving meshes, particle size distribution and so on)

Best,
Christian.

I'm not an associate of DCS GmbH and not a core developer of LIGGGHTS®
ResearchGate | Contact