How does auto Vcore work when overclocking?

Astroflash · May 21, 2020

How does auto core voltage work when overclocking? If I understand correctly, when set to auto vcore at stock settings, the the mobo feeds the CPU the correct voltage according to a voltage table. However, I assume when you overclock, the mobo doesn't know the voltage values as it's out of spec, so how does it know what to feed the CPU?

Jurrunio · May 21, 2020

it just applies the same voltage as the max stock clocks for that CPU, but realistically it will drop further due to vdroop

that's of course, if the board vendor does not set custom behaviour. Auto voltage can sometimes still try to raise voltage without following the VID table ending with something unsafe.

Falkentyne · May 21, 2020

2 hours ago, Astroflash said:

How does auto core voltage work when overclocking? If I understand correctly, when set to auto vcore at stock settings, the the mobo feeds the CPU the correct voltage according to a voltage table. However, I assume when you overclock, the mobo doesn't know the voltage values as it's out of spec, so how does it know what to feed the CPU?

For Intel:

The voltage table (vCPU) read from when AC Loadline and DC Loadline are 0.01 mOhms, scales via multiplier up to the highest turbo boost frequency.

And it works because AC Loadline (which is set to an Intel spec value, not 0.01 mOhms) makes the CPU request an input voltage from the VRM, to counter the effects of vdroop via input request, which is determined by current (amps), via a resistance multiplication formula, and DC Loadline is a guess of the output voltage (vout)'s voltage droop, which is used for power measurements and power limit monitoring, but does not actually affect the voltage itself. Both of these values calculations are shown as CPU VID in the end.

That formula is VID=vCPU + (ACLL * I) - (DCLL * I)

where vCPU is the preseet VID when ACLL and DCLL are equal to 0.01 mOhms (basically, 0)

vCPU is affected by temps, where higher temps will make vCPU rise, usually at the highest turbo multiplier, it is +1.5mv every +1C, or -1.5mv every -1C

At lower turbo multipliers, the mv scaling is less, and usually between 8-10 multiplier points down, there is no scaling at all when temps change.

This scaling is officially known as "Thermal Velocity Boost voltage optimizations" and has been around since at least Skylake.

ACLL is AC Loadline in milliohms and DCLL is DC Loadline in milliohms.

It looks like they are redundant and cancel each other out, but you need to study electrical engineering, as the AC side and DC side occur at different points on the circuit.

The AC side is the request. The DC side is used to control voltage vdroop.

CPU Vcore however uses only the AC Loadline and the offset, not the DC Loadline.

Vcore = vCPU + (AC Loadline * I) - (Loadline Calibration * I) + vOffset.

Where loadilne calibration (desktops) is a percentage reduction of the intel default VRM loadline slope (which happens to be the default values for both AC loadline and DC Loadline, by the way), and on laptops, loadline calibration cannot be changed and is always the default VRM loadline value of max vdroop for that platform.

vOffset is offset voltage.

Now for AMD? I have absolutely no idea.