If you just start and stop VMs, you can easily run those VMs on a 5950x. Things change drastically when you start running you applications on the VMs, depends on how heavy those applications are. Even if half of these VMs start running heavy apps, your system will crash or you will see lot of errors and VMs will crash. If you are setting up this lab for your own learning, you start with a 5950x, unless you need PCIe 5 and DDR5 for sure (application dependency).
I run a 4 VM cluster on Ubuntu powered by 5800X (8 core 16 thread) and when I do not run anything in VMs, CPU usage remains under 10% and most of the cores are unused. But then, when I install applications (eg: openshift, kafka cluster, SQL database, backend modules, frontend modules etc), the CPU utilization starts going up but it still remains under 20%. Things will be vastly different if I turn this into a prod server or a SIT/PT/UAT server where multiple users hit the application. The 5800x wont be able to cope up with the load as more users hit the application/modules.
I would not go for two medium power desktops as you will end up paying more in the end. Also, you have to look at latency of home LAN when you combine two PCs for your cluster. Latency between two VMs residing on same PC is far far far less than those residing on different PCs on a home network. You will also need to setup a proper home network with switch to reduce latency and errors.
Go with top of the line CPU like 5950x or 12900 if you cannot invest in EPYC or Xeon. Start building the cluster and you will know whether you will need more power or not. If you still need more power, add another PC and form cluster.
What advantage does a server platform give over desktop platform?
You can seat more than 1 CPU in server motherboards and due to this, if you need more CPU power, you just buy another CPU and seat it in second CPU socket. As you pay more, you get motherboards that can take more CPUs. The ECC memory used in servers does error correction and this is critical in production like environments. All components are designed to last long. I have seen servers run for years and years without being turned off. Their reliability is at a different level. Also, usually, there are more PCIe lanes in servers when compared to desktops.
