_ Credits: I would like to thank my colleague Vladimir Petrov from the VCF BU for providing me with additional information that allowed me to write this summary _

The vSphere 8 Update 3 release notes contain lots of changes, but one of them caught my attention:

Cluster-wide option to retain virtual NUMA topology: vSphere 8.0 Update 3 adds a setting to retain preconfigured vNUMA topology even if the VM moves, allowing for better NUMA topology tuning for VMs across all the hosts in the cluster. This is the advanced vCenter Server setting VPXD_PersistVnuma to keep the virtual NUMA topology on a cluster level under Configure > Advanced Settings in the vSphere Client.

Thanks to the information Vladimir gave me, I can now make sense of the enhancement.
I did not made the connection between the previous enhancements in vSphere 8 and this new feature.

What I had in mind is that the vNUMA of a virtual machine is fixed any way. Like the setting numa.autosize.once in the documentation tells you:

When you create a virtual machine template with these settings, the settings are guaranteed to remain the same every time you subsequently power on the virtual machine. The virtual NUMA topology will be reevaluated if the configured number of virtual CPUs on the virtual machine is modified.

In short, the first power-on of a virtual machine creates the vNUMA topology and (by default) it will retain the topology unless you change the config or modify the setting(s). A blog by Frank Denneman goes into the process of adjusting your VM when your CPU hardware layout changes and how to address that.

TIL, this was true up until vSphere 8 and virtual hardware version 20, now the virtual NUMA topology of the VM is always re-evaluated when the VM is powered on.

Whoa! That’s pretty slick.

But, getting back to the topic at hand:
Now, the feature VPXD_PersistVnuma makes sense as it allows the admin to revert to the previous standard.
This is more relevant in scenario where the Ops team is acting as a provider (think IaaS) and simply does not want a surprised customer trying to deal with a new NUMA layout on the next reboot.

Of course, the vNUMA auto adjust will not override manual configurations as described here. Also in the feature is a logging entry about the recommended NUMA size from the ESXi perspective to help you adjust the setting later on.