Supporting architecture for KOKKOS package

LAMMPS LAMMPS Installation
1

Hello all,

I am planning to buy a small server with RTX nvidia graphic card for MD simulations. I want to use KOKKOS package to do large-scale simulations. From the manual it seems that the newest intel cpu architecture supported by KOKKOS is Sky Lake (SKX) so the cpu should the intel cpu of 6~10th generations. Does the KOKKOS package in stable or developing versions of LAMMPS support intel cpu of 11th or 12th generation? If not, when would the newest architecture (e.g., Alder Lake) be supported?

2

The CPU architecture in KOKKOS is merely a hint for the compiler optimization settings and how to enable vectorization.
The Nvidia GPU is a much bigger concern, as KOKKOS only supports double precision and thus using a consumer grade GPU is not going to be a good idea.

1 Like
3

Thanks for your reply and kind advice. So that means it’s OK if I use SKX as the architecture ID when installing KOKKOS package even if cpu is the newest. For GPU part, I will consider both the performance and the price of GPU. Thanks a lot!