cr VS pcm

I much prefer the approach taken by cr.h which monitors a "plugin" binary instead and acts as a simple wrapper around platform specific dll/dylib/so loading. You then have a multitude of strategies if you need persist state such as serialization, placing code or data on the host side and using virtual methods/fptrs, holding third party lib references to dlls on the host side or using the CR_STATE mechanism for simple things.

Depends on your use case, but I would second the option of just hot reloading c++. I have been experimenting with that (using cr.h) and its great - as long as you make sure your compile and link times are fast enough.

Also read about Intels PCM.. and the OPCM but I cannot determine if it supports Big Sur.. I've ran into issues after the cmake processes so the binaries fail to run.

You can monitor anything you want: https://github.com/opcm/pcm up-to the cache hit rates, memory latency and the pipeline states.

There are tools to report memory controller utilization, they tend to depend on your CPU. This might be more advanced than what you are looking for, but recently I was using this https://github.com/opcm/pcm#downloading-pre-compiled-pcm-tools to monitor memory read/write and comparing it to the same values when running a memory benchmark to get an idea of the utilization. That tool only works on Intel CPUs, there is probably a similar tool for AMD cpus.

Can you check with e.g. https://github.com/opcm/pcm whether the PCM isn't unset when you shut down the VM? Alternatively livekd and https://docs.microsoft.com/en-us/windows-hardware/drivers/debugger/-pmc

Is anyone out there running PCM utility under VMWare? AWS EC2? https://github.com/opcm/pcm

Now, if you really want to be more precise about throttling, you can use a process counter: https://github.com/opcm/pcm. You can run these commands in a terminal and see how much power your CPU is using and that's another indicator of throttle. Some workloads will hit the power ceiling of your CPU without clocking all the way to the turbo frequency. A good example in my experience is prime95. Even with adequate cooling in a workstation tower and xeon processors, my cores tend to run below the all-core frequency.