You wouldn't have this question if you understand what NUMA is; or at least you would have a different question such as how to ensure RAM locality. You should also read how ESXi schedules VMs in NUMA systems.
https://docs.vmware.com/en/VMware-vSphere/7.0/com.vmware.vsphere.resmgmt.doc/GUID-BD4A462D-5CDC-4483...
As an extreme example. Suppose you have a system with 2 CPU package each with 10 cores. You currently have only one VM running assigned with 4 vCPUs. That VM runs a web server that caches recently used/frequently used static files like HTML/javascript files, image file like GIFs, JPEGs, PNGs in RAM to avoid reading from disk. So the VM starts all vCPUs in package 0 and all the cached files are in the RAM connected to this CPU. You don't want ESXi to be scheduling this VM to package 1 as it would have a penalty in accessing the RAM on package 0. So in this scenario, even if package 1 is fully idle, you want it that way as you don't want your single VM have a penalty to access RAM on package 0.
