Bought myself an Ampere Altra system

GHz/MHz wars and ultra-deep pipelines lead to long pipeline stalls and low efficiency. Clock != (IPS) cycle efficiency. What matters is measured performance. There's little point in buying the most expensive option, it's usually throwing money away.

Sometimes though, option are limited but there are also traditional and alternative channel vendors besides secondary markets. For example, a vendor or the mfgr might be willing to sample a part.

When comparing the same design, it absolutely makes sense, because it usually scales. Of course in this case, there's the SRAM which is shared, memory bandwidth, etc. As a rule of thumb: run your own benchmark!