GPU in LIGGGHTS

Tarun De's picture
Submitted by Tarun De on Sat, 09/12/2020 - 19:33

Hey all,
I just wanted to know that whether LIGGGHTS-Public is compatible with GPU or not? Is there any process to install LIGGGHTS using GPU? Does the premium version support GPU? Any information will be too useful for me.
Thanks in advance.

Best regards,
Tarun
Research Scholar
IIT KGP, India

mschramm | Sun, 09/13/2020 - 03:51

LIGGGHTS-PUBLIC is not built with GPU support. I also don't believe their new product has GPU support also due to it being built on top of LIGGGHTS.

LAMMPS (what the "L" in LIGGGHTS stands for) does have support for GPUs but I do not know anything else about it.

How many spheres are you trying to simulate?

Tarun De's picture

Tarun De | Sun, 09/13/2020 - 10:52

Hey Matt Schramm,
thanks for your quick reply. I wanted to simulated a system with at least 7500000 particles. But it is not possible to simulate in usual CPU. That's why I was interested in GPU if exists for LIGGGHTS.

paul | Tue, 09/15/2020 - 08:26

It is possible - just use more CPUs. LIGGGHTS is MPI-parallel, meaning you can use many cluster nodes to run it.

mschramm | Mon, 09/14/2020 - 16:17

Hello,
From the link provided, I saw tags for omp but not gpu for the granular pair forces. This makes me believe that gpu support does not exist for granular contact forces. They may still exist but you would need to check in the gpu package, in the src folder, if the pair wise interaction is there or not.

nasato | Tue, 09/15/2020 - 17:19

Hi,

I haven't build it myself for GPU (I should though, it looks interesting), so I may be wrong. But the link I mentioned above states "These accelerated styles are part of the GPU, USER-INTEL, KOKKOS, USER-OMP and OPT packages, respectively"

Additionally, from their website (https://lammps.sandia.gov/) it is provided the following in the version history "(10/18) Kokkos support (GPU) for granular interactions." As far as I understood, you can enable GPU by building KOKKOS package https://lammps.sandia.gov/doc/Build_extras.html#gpu

Bests, Daniel Nasato