Interleaving memory across NUMA regions has been an option in SWIFT for a few years now and it it probably time to make it the default. On AMD machines like COSMA8 it exploits the available memory bandwidth and it is recommended on that machine. On machines where using an MPI rank per NUMA region is best, then this option has no effect. It has also become true that MPI libraries that use UCX also benefit from this option as well.
This toggles the
--interleave option replacing it with
--nointerleave just for completeness.