I've decided to finally take the time and set up my per core undervolt. I've attached the results and performance gains compared to PBO+200, Scalar 1x, -20mv all core undervolt.
I've ended up as follows (strongest to weakest core – using HWInfo Core Performance Order):
-18, -22, -25, -25, -35, -35, -40, -40
Tested with core cycler for 5+ hours, multiple iterations of y cruncher stress tester, Aida64Extreme and OCCT.
I kept undervolting every core individually until I achieved the highest performance and then tested the stability as mentioned. Clearly the -20mv on my strongest core (3) as limiting performance and perhaps long term stability (never had an issue) but I did see performance degradation when going above -18 and the best performance at -18.
Oddly enough, Cinebench 2024 and 2026 performance remained very similar (few points more with per core, but not worth mentioning). Overall, I think it was worth it, gave me a slight 3DMark boost, and clearly more performance in R23. I wouldn't say -20mv all core is unstable at all, but there was clearly there was some headroom left for performance. Further, with setting the weakest cores to more than -30, the CPU runs significantly cooler at PBO +200, another benefit.