Coming to this thread from a Veeam user perspective looking to migrate from ESXi to Proxmox VE. Veeam seems to only understand specifying vCPU in its setup of the PVE Worker which results in multiple sockets of a single core. As I am reading it above there is nothing to be gained by exceeding the number of physical sockets and some benefit in matching what is physically available as it enables NUMA to operate correctly.