Originally Posted by vonPelz
I believe all three options will increase load voltage. But upping LLC and/or SVID behaviour will also affect non-turbo volts I think, and LLC will counter vdroop (for better or worse?).
So why use LLC instead of just upping the turbo voltage? Won't one have better control of the resulting load voltage with the latter option? I see everyone using LLC in their OCs, but I fail to see the immediate benefit...
As for adaptive being broken, changing SVID behaviour to best, IA AC & DC load to 0.0.1 and LLC to default worked for me (Z370-E). Otherwise the voltage was way higher than needed at load. But it sounds strange if the Maximus boards are providing less voltage than expected...
There's a difference between turbo and turbo. Your CPU can be clocked at 4.8GHz without actually doing
much and without consuming much power at all (try setting the power management in Windows to "high performance" and I think it'll stay clocked at max turbo all the time). Using the turbo voltage will raise that "idle but clocked up" voltage, but you don't really need that. When the CPU actually starts doing stuff and actually consumes more current, then the voltage will droop and counteracting that is the point of LLC. You don't need to counteract it completely, in fact you probably don't want that at all, but it's still helpful in your situation. What LLC won't
do though is increase voltages when the CPU isn't heavily loaded.
Now, granted, what LLC is really for is when you need a pretty high voltage for heavy loads but you don't want to set your idle-but-clocked-up voltage to something dangerously high in order to get your load voltage after vdroop high enough, but I still think it's a pretty handy tool to use in your case. Raising the turbo voltage would also work though. The SVID behavior seems like a crude tool to me, though.
I've only tried manual so far to get a baseline for this CPU at 4.7GHz with no AVX offset. Setting 1.3v with LLC 4 leads to a reported vcore of 1.232v when running OCCT's small dataset test. HWinfo says the CPU package draw is just shy of 150W and a glance at the watt meter in the wall socket seems to indicate that's in the right ballpark. Seems quite stable so I think I'll take a look at 4.8GHz tomorrow.