Ok, so, yesterday I really started testing and OCing my Ryzen 7 beyond that silly "let's find maximum bootable clockspeed with 1.2v). At home I have some more detailed data, but for now, I can share some little things off my experience:
At 1.2v and LLC 1, custom AIDA 64 (1344k, FFTs in place, 15 minutes per test) immediately crashed (
@Charlietwo). I started increasing core voltage by 0.2, and at 1.3v with LLC on Mode 1, the computer was rock solid on any workload (actual voltage on high load was 1.336 due to LLC).
HOWEVER, I still had some boot issues: usually, board failed to boot at least once every time I turned on the PC with the selected voltage, though after restarting, everything went fine and dandy. Fiddled around with memory, SoC, LLC voltages, nothing. Started messing with LLC, Mode 2, 3, 4, 5, 6... Didn't solve the boot problem.
However, testing on Windows after the PC booted, I noticed something very interesting: LLC Mode 1 gives the highest voltage increase on workload. MODE 1 gives the highest voltage bump (+0.36 at 1.3), with Mode 2 giving +0.24 (I believe) and subsequently less (0.14 Mode 3, 0.8 Mode 4), until LLC Mode 5 which LOWERED Voltage being delivered to the CPU. So, Mode 1, highest voltage, with decreasing voltage added and even negative voltage applied to your defined Vcore as you increase the LLC Modes. Very, very interesting behavior, something I wasn't expecting and never even saw referenced anywhere. This is with the latest stable 1.8 BIOS on the X370 Gaming Pro Carbon; not sure if bug or feature.
However, my boot issue wasn't solved yet. So I bumped nominal vcore again to 1.3125... Still failed on first boot. 1.325 nominal voltage was the value I needed for my Ryzen 7 to boot stably at 3.7 GHz. Immediate voltage required was 1.325 for boot up.
Now here's the thing... My CPU doesn't need nearly that much voltage on heavy workloads to be rock solid.
Fiddling around with the LLC Modes, I finalized with a 1.325 vcore for stable boots, and LLC Mode 6... which means that actual voltage that kicks in after LLC is applied (which doesn't happen during the boot sequence, but after initial 1.325v for the processor to boot fine) ends up oscillating at 1.304 and 1.312v, on full AIDA64 load (1344k, FFTs in place, 15 minutes per test). This allowed me to bring rock solid stability (for now...), full load temperatures down by around 7ºC compared to 1.3v + LLC mode 1 (1.336 actual.) So my processor is actually running at 1.304/1.312v, despite vcore at 1.325 for stable boots, due to LLC Mode 6.
Very interesting to me, and might help some of you achieving a good overclock with as little voltage as possible. I wish it was possible to define a boot voltage value and another voltage value for actual operation of the CPU, but if this option exists in BIOS, still haven't seen it. However, for the time being, I consider these LLC Modes with reduced voltage an actual feature, since it allowed me to solve my boot issues while keeping OS and load voltage as low as possible.
My voltages are stable at stock 1.35v for memory (@2933), and I locked SoC voltage to 0.95 with LLC 1 (actual 0.97 or some such, have to check.) I recommend locking SoC voltage at 1.0 or 0.95v (my system runs solid at 0.95 SoC voltage with LLC 1, as I mentioned), since the mobo delivers around 1.16v on Auto, which is just waayyyyy more than the SoC needs. I shaved some 5ºC on CPU temperature by lowering the SoC voltage from 1.1 to 0.95.
I'm not home, so I don't have any screenshots, but I'll update this post later with my notes and some screenshots.)
Hope this helps someone. Gotta say, I really loved fiddling around with these settings