High Voltage on Auto Clock

I think I have cracked my problem though!

P-states and Offset voltage is the key.

CPB, C-states etc all enabled as default.

P-0:

Custom 3400MHz @ 1.300v

P-1:

Custom 3000MHz @ 1.275v (same as default)

P-2:

Custom 2200MHz @ 0.900v (default) or maybe higher

In the OC Tweaker tab:

The second instance of CPU Vcore Voltage set to "offset" and start dropping it.

I've set it to -0.05v that's 50mV lower.

CPU Load-Line Calibration Level 3 (allows some vDroop)

Idle @ 2.2GHz = ~0.819v (from ~0.869v)

All cores @ 3.5GHz = ~1.160v (from ~1.2v)

Single core @ 3.8GHz = ~1.369 (from ~1.495)

This is looking promising at doing what I want.

Raising the P2 idle state a little will keep things ticking over nicely there whilst I reduce the offset further.  I may have to actually raise the P0 state back up as well, to ensure the ~1.2v is delivered for all cores, as it's quite a taxing scenario.

Not fully tested for stability, and not tested with a multimeter, so I could be being lied to, but this all looks rather good.

This seems an odd way to circumvent things though, pushing up vCore on P-states whilst globally offsetting voltage down.  In this case, are we telling the motherboard to offset what requests come from the CPU, or we just telling the motherboard to offset what it ever delivers?

I seem to have carried on my own thinking rather than perhaps helping the original poster again.  Sorry.

[edit: typeo on vCore number!!!]

When I think about it, it does make sense that the CPU could request 1.55v as the VID value read for each core through HWinfo64 goes up to 1.55v.  Coupling in some vdroop from the motherboard a spike to just over 1.5v could be seen when running single or dual core tasks and thermals are good - the max possible frequency on those two cores will be achieved and so the chip will request that voltage.

However, when I asked about fixing the voltage to 1.30v, they also said that this was safe to do.  They add that I may experience instability due to the CPU perhaps requiring extra voltage [for the fastest frequency single/dual core loads] and the fixed voltage actually starving the chip. This seems like the issue that there are some chips that are just crud from the production line, and they will actually require about 1.5v to hit 3.9GHz, so all chips are programmed to request that.  Probably most chips won't need anywhere near that voltage for single/dual core operation at those frequencies.

Recommendation is to reset to defaults and use the processor [with it's ~0.9-1.5v range].

@yodivanbaped:

You should be able to let the CPU do everything auto but fix the voltage.

On the "OC Tweaker" page in BIOS, scroll down to the "Voltage Configuration" section (underneath the RAM timings).

Put "Voltage Mode" to Standard/Normal (not Auto, and not OC mode).

Put  "CPU vCore Voltage" to "Fixed Mode"

Put "Fixed Voltage" to 1.35000

Put "CPU Load-Line Calibration" to Level 3, 4 or 5.

Do not change the frequency of the CPU or change anything else from Auto.  This will make the CPU do everything as normal, except that you are forcing the motherboard to always provide 1.35v.  It should be stable.

The reason for the load-line calibration settings is that the greater the number the LESS it compensates for vDroop (under load), but also the less it overshoots when vDroop goes away (under no load).  Level 1 and level 2 can actually put the voltage higher than the value you set, and of course overshoots can be more extreme.  Equally Level 5 might not be aggressive enough, and if your chip needs the voltage, it may get a bit less than what value you entered.

You shouldn't need to set the clock frequency - because that is not in the voltage section.  Let the chip do whatever it does automatically.

Now, P-states are an interesting topic in themselves, and can provide the efficiency of a stock processor on auto settings when idling, but also a rather monstrous overclock when the system demands it.

However, what I am trying to do is not to overclock using P-states, but maintain a stable voltage under each load condition that is accessible (idle and all-core load) whilst using the global offset of voltage to provide a negative offset of the stock auto voltage.

I know my wording is probably hard to read.

So far, I am starting to see what I want to with my offset method and haven't needed to raise the idle or -all-core voltage just yet to compensate for the global offset.  It'd be nicer if they allowed use to enter a max ceiling for the motherboard instead of just fixed mode, but I am achieving a ceiling based on offsetting from what auto uses, and what fixed mode worked at.

...The most accurate CPU Voltage reading will be from SVI2 TFN which is reporting the voltage the VRM controller is seeing. Additionally some will freak out because of seeing CPU Voltage up to 1.55V in BIOS and OS at default settings. This is the default AMD behavior when CPB is Enabled. If up to two cores are loaded, the CPU will boost to a higher frequency with higher voltage (on 1800X up to 4.1 GHz 1.55V). If more cores are loaded, the frequency and voltage will drop. It's not dangerous and as stated, AMD spec.

So it's fine!  Which is kinda what the AMD rep said in the replay to my questions.

It seems strange that there was mention of 1.35 and 1.40 and 1.45v as maximum, do not go over for 24/7 operation.  Admittedly these were for overclocking, so it could well be that the CPU itself handles these voltages and actually can take in quite some volts to it's power distribution, which in turns sends out more sensible levels internally.  When overclocking is enabled, the power controller is less responsible and the user has to be mindful of not pushing volts into the thing.

As stated in a previous reply, running on 1.30v fixed is also fine and safe according to the AMD rep's reply.  Just that it's no longer within normal limits and may be unstable.

Looks like my chip might be a real low power consumer!