Yes you must run the CPU Benchmark whenever you change the CPU. But what is your question?

BOINC performs a benchmark every so often (was something like every 4-5 days), and I think at restart when the last benchmark was N days back or you update BOINC.

The benchmark values have a relative small impact on how credit is computed, but artificially hiking your machine benchmark is kindly met with the opposite effect. If you tell WCG it's able to do 10GFlops and the machine really does 5Gflops, the machine gets overassigned work. You ask for a 1 day buffer, for the benchmark values drive that, but the device is really only able to do the buffered work in 2.

So, " the claimed BOINC credit" is not necessarily what you get... the host is found out quickly by the rest of the devices in the pool.

You can actually set a cc_config.xml to never do a benchmark test again with <skip_cpu_benchmarks>1</skip_cpu_benchmarks>. There used to be a time when a benchmark would result in tasks being in progress got unloaded. Newer client don't do that anymore, I think to remember, but I've left the config setting in. No point in wasting good crunchtime.

I set all the extra work buffers to 0, when a task is at 99% it starts downloading a new one, I don't want all the backlog on my PC, I think the buffer is a bit an obsoleted feature when perhaps internet connectivity was more erratic in the past. This way I get the shortest turn around for the WUs because they are immediately being worked on.

I accept the low risk that if the connection goes down on my or the server side I cannot do work temporarily.