Once, this board/CPU combo would do 4.3 at 1.33V. That degraded quickly to more like 4.1 at 1.36V, at which point I started doing manual clock control and not letting it run over a volt unless necessary, which seems to have halted its degradation well. Those numbers are with all the power management and miscellaneous settings at stock/efficient/practical values. For this, I disabled one core, disabled all the power-saving features and safeties, boosted lots of miscellaneous voltages by 0.05V, threw most of the switches ASRock said might help with overclocking, and clocked the uncore and RAM at 3.2 and 1600 without turning down their voltages. Testing was 3-5 minutes of prime95 in-place large FFTs, because I didn't want to spend all day on this.
freq Vcore Vin status
4.4 1.40 1.90 pass
4.5 1.40 1.90 pass
4.6 1.40 1.90 pass
4.7 1.40 1.90 epic fail, had to use CMOS clear button
4.7 1.45 1.95 failed at Gnome initialization
4.7 1.50 2.00 pass
4.8 1.57 2.07 failed at early boot
4.8 1.60 2.10 failed at Gnome initialization
4.8 1.63 2.13 failed at early boot
4.8 1.66 2.16 failed at Gnome initialization
4.8 1.70 2.20 failed at Gnome initialization
I called it at that point because I don't want a literal halt-and-catch-fire, and that RAM is still useful and in the potential blast radius of such high-voltage shenanigans. I also wasn't stopping to check thermals, but I disabled that limiter and extrapolation says that if I got prime95 to run at 1.75+V it might well break 100C.
I was testing with a stock Antergos image, hence the Gnome. The real kicker here is that on 1C1T at 4.7 GHz with all the power management disabled, Gnome *finally* runs passably fast.