Ever wish you hadn't started tinkering with things!!?
So, my 2600x now seems to have lower boosts and higher voltage spikes than when I first started. I've reset to defaults and started again, but it still seems worse than when I first had it.
Over the last week I've turned on precision boost, tried offset voltages, set fixed voltages etc. Now back to defaults and it just seems like voltages are way to high.
Is it worth doing a fixed 1.38/1.4 voltage and just let the rest take care of itself? I know 4.3/4.4 boosts are easy enough to achieve on some cores with a 1.4
Cannot find any info on this damn board to see how others are fairing (Prime Pro - I think a couple of folks on here have the same board)
Sorry about the wall of text below, but well you did ask ...
2600x, Prime X470-Pro here too, with mid-range Be-Quiet air cooler (Shadow-rock slim).
Mostly run with the 4011 BIOS, the current one.
Board Defaults/Auto gives 4250 (4241/bus 99.8 reported) peak boost, off-load boosted core voltage peaks at around 1.47-1.48. Don't recall seeing it ever hit 1.5. 1.45+ might seem high, but its only doing this very briefly under low-load/low-power/low temperature situations.
All-core load drops core voltage to ~1.33 or less (AMD specs probably require this).
Seemed happy at 4.2G fixed at the 1.3875V I used for the few days I ran it at fixed frequency, though I tend to run it on Auto/XFR.
4.26G too at same voltage, though not much stability testing done (all-core load still brings the core voltage down to ~1.33V, so isn't cooking itself). 4.3G and beyond not particularly stable.
PBO doesn't seem to do much if anything, other than probably throwing more voltage at the processor for little/no gain. I have both PBO settings under CBS disabled. Still boosts to 4.25G, all cores ~4-4.1G depending on load.
I tend to run both the core and SOC voltages with a negative offset (-80mV core and -60mV SOC at the moment), better for heat management of the system as a whole. This gives SVI2 TFN core voltage of 1.3-1.4V and reported VSOC of ~1.025/1.03V.
SOC voltage doesn't seem to have a huge impact on memory scalability/stability in my case, so lower is better for power-dissipation reasons. Not tried below 1.025 or above 1.1 however.
BIOS 4008, the one applied to the board from the factory did seem to allow for higher sustained all-core boosting under load.
In Cinebench runs, I was seeing 4.1G (4091 reported) on all cores during the run, dropping down occasionally to 4.075G, sometimes 4.05G as the cores heated up.
BIOS 4011 seemed to drop that Cinebench all-core loaded boost to 4.05/4.025G, dropping to 4G or slightly below sometimes as it heated up.
The 43.18SMU update provided by AMD to the board manufacturers is probably what is responsible for this, and likely done for long term stability/longevity of the processor.
Some blue screens in light/moderate/transient-loads, initially thought this was probably due to the negative offsets on the core/SOC, but more likely due to pushing some of the ram timings / IF a bit too hard. Hasn't done this recently after re-doing some of the timings.
Memory at 1.32V 3466 C16 15 15 15 Trfc 350 Geardown/1.5T (TG Xtreem 4000 B-die).
C14 at this frequency needs over 1.4 volts to be ~stable (I ran/HCI'd at 1.425V for a while, 1.4V failed MT86 after 90secs).
Probably OK at that, but prefer the higher-margin, significantly less stressed C16 option.
Aida DDR latency reporting doesn't seem consistent at C14 3466 either, sometimes returns 60.6nS, mostly ~62.7nS which is basically the same as C16 tightened.
Real-world usage/testing is probably a better guide as to the effects of memory timings on performance anyway.
For me, its a great processor/board setup. Dial in an negative voltage offset or two to help lessen heat and assure longevity, pick the optimal moderately stressed memory frequency in the 3200-3533 region your memory/chip can handle at reasonable voltage, and remember that the harder you push it the less margin you have to stability on multiple fronts.
Settings which may seem stable after a battery of stress tests, may still be skating right on the edge of stability with errors only occasionally showing up when the system is transiently loaded, which is true for any system.
