Just a quick comment about TDPs and the like, since I did a lot of tinkering with undervolting way back in the skylake era (oh wait, we're still in the skylake era!
).
If all else is equal in a processor, one with lower TDP
may perform less well under intense load (I'm talking HPC, prolonged, load -- maybe longer than short consumer benchmarks),
if the load causes it to exceed the rated TDP longer term. The benchmarks you ran may not have sufficiently stressed out the CPU for any differences to become visible, if it indeed is just a TDP-capped version of the exact same CPU as its consumer counterpart.
But here's where things get interesting. I prefaced my earlier statement with the qualifier "if all else is equal." This is often not the case, as CPUs can be configured with different
voltages per clock. Let's consider the corrollary with 2 chips of identical TDP, but lower stock voltages per clock, all else equal. In this case, the CPU with the lower voltages would be expected to outperform the one with higher voltages on workloads that push them to their max TDP, because power dissipated is a function of voltage applied! So what if the processor had
both lower voltage and lower TDP? Then it gets hard to say without having the exact numbers. In theory, there is a voltage "break-even point" where if you lower the voltage
just enough to compensate for the decreased TDP, it would perform exactly identical to a higher-voltage, higher-TDP CPU of the same clocks under all loads (this could be calculated, but I am too lazy and/or ignorant of the specs), and also in theory, you may have found this lower-voltage, lower-TDP CPU. In theory, though.... just theoretically.... you may have found a processor
under that tipping point where it can actually
outperform the higher TDP CPU in certain workloads, specifically those that would push the higher TDP processor to throttle. How's that for mind-bending?
In reality I suspect you have a CPU that is either just TDP capped (and all else truly is equal) but you haven't really pushed it to do the high-AVX in-cache linear algebra operations needed to induce throttling, leading to the appearance that they are equal (but really only equal in the workloads tested by these benchmarks); or you have a CPU that is indeed somewhat lower voltage, helping to mitigate the difference in TDP.
I am absolutely stunned you went to China for this CPU by the way. Very impressed.
One final comment -- I really love my HP Z8 workstation's cooling, as well as my Colfax Intl Knights Landing watercooled tower. Absolutely fantastic beasts that caused me to believe, quite wrongly and to my bitter misfortune, that I could trust a third-party company to build me a dual Epyc Rome 7742 tower that had similar high-performance cooling properties. Unfortunately my machine was kiddie-grade, absolutely horrifically built nightmare where every component would take turns overheating under moderate HPC load. So I do appreciate an industrial-grade cooling system built to-spec!
