It may be useful to combine all PMEM in a system into a single volume for apps which need one very large memory-mapped address range or managing more than one volume is too cumbersome.
This cannot be done across CPU sockets by ipmctl (https://docs.pmem.io/ipmctl-user-guide/provisioning/concepts, "Regions cannot be created across CPU sockets."). It can be done by LVM, which is how PMEM-CSI could support this. The downside is that region affinity gets lost.