Originally Posted by nangu
I ran a lot of tests on my Aorus Master / 3900X setup, and the behaviour of my system is the same as yours. I tried Stock PB, PBO Auto, PBO Manual, PBO Maxed, undervolting 0.05 and 0.1 vCore, etc. Cinebench R15, R20 and CPU-z benchmarks all have variances which you can say are between margin of error. It's the same with game benchmarks. I tried F1 2016 which is a multithreaded one, and IL-2 which is heavily ST and core clock dependant. Max sustained core clock I can see on ST on this game is 4.3Ghz, bouncing between core 0 and core 2 constantly. I know my CPU can handle 4.4Ghz on the first CCX, so CCX OC gives me better results in both ST and MT scenarios, with better thermals as a bonus. I don't fully understand why PBO can't sustain at least 4.4Ghz on heavily ST workloads.
I recorded spikes in ST 25Mhz higher than stock with PBO enabled, but these spikes are not sustained to show an improvement.
At this moment, PBO does nothing to performance, and increase power consumption and heat a lot, so it's useless. I can run Stock with 0.05 to 0.075 vcore undervolt and performance it's ~1% lower, but with much better thermals and power consumption, and I can regain that 1% with memory OC eventually.
You can read the strictly technical post by the @The Stilt here.
I found it to be pretty informative. Essentially, the Zen2 CPUs are all voltage bound. The have an acceptable voltage that they allow for light and heavy threaded workloads. Most of the processors are already voltage bound at stock, so enabling PBO and altering the PPT, TDC, and EDC limits does nothing, as the CPU wasn't bound by those parameters. I have attached a few excerpts from that post that best illustrate the issue, but basically Matisse CPUs are voltage maxed even a stock.
Since the FIT voltage is met before stock PPT, TDC and EDC and/or temp limits, buying a motherboard with better VRMs does nothing, that wasn't the limiting factor anyway.
Adding better cooling does nothing, as the FIT voltage is designed to prevent the heat density on the tiny CCD from getting to high. The IHS simply cannot transfer heat away fast enough to make high voltages feasible without affecting silicon reliability.
Essentially, Matisse CPUs being bound by FIT will perform virtually identically for everyone, regardless of motherboard or cooling solution used. (Multicore may improve a bit with better cooling, but not single core boost).
That isn't really a bad thing, because AMD can argue that they want everyone to get the most out of these processors and they do. I think what muddies that water with this whole boost thing, is that AMD hasn't really published anything on FIT. They say that a CPU will boost to Fmax if TDC, EDC, PPT and temp limits haven't been met. So a user, who doesn't see that 4.6 GHz, can say, well I'll just buy a X570 with an awesome VRM, remove all limits with PBO and raise Fmax to 4.7 GHz and then I'll have my max boost no problem. Except, they don't hit it. In fact nothing seems to happen at all.
Post like this serve to muddy the waters further.
Hallock seems to say that a better VRM and cooling will get you better clocks. Except, with The Stilt's data and being FIT bound, that just isn't going to happen. I realize AMD wants to sell X570 mother boards with the chipsets they spent R&D money designing, but this floats really close to false advertising. Better to focus on the better audio DACs, PCIe 4.0, and next gen wifi to sell boards.