so what's the deal with LLC?

[soccerballtux]

why do we have manual settings for it, why isn't there a feedback network that handles it for us?

It feels like a step back from what I used to have on my GA-MA790X-UD4P, which was a rock solid CPU voltage that wouldn't budge.

I don't see why we need "Low" "Medium" "Extreme" settings when a feedback network could just react accordingly as the load requires.

 

[Idontcare]

I never understood this as well.

The mobo know the target/desired voltage (be it at idle or at partial load or at full load) based on either the VID or the user-specified Vcc, and it knows what the actual voltage is, so why isn't there a feedback loop involved to simply drive the actual voltage to the target voltage?

I'm guessing its just a cost (hardware) and validation (people) issue. You don't want to implement algos that drive the voltage feedback to aggressively or you'll risk creating destabilizing swings. But there are solutions to that, IP to be leveraged from other industries. So we are back to it just being a cost issue?

 

[soccerballtux]

I never understood this as well.

The mobo know the target/desired voltage (be it at idle or at partial load or at full load) based on either the VID or the user-specified Vcc, and it knows what the actual voltage is, so why isn't there a feedback loop involved to simply drive the actual voltage to the target voltage?

I'm guessing its just a cost (hardware) and validation (people) issue. You don't want to implement algos that drive the voltage feedback to aggressively or you'll risk creating destabilizing swings. But there are solutions to that, IP to be leveraged from other industries. So we are back to it just being a cost issue?

how did they manage it before? My GA-MA790X-UD4P never budged more than maybe 5mV with CPU at 4ghz/no CnQ enabled, and starting/stopping Prime95 over and over

given the impedance of the supply it should be possible to tune the active feedback network to compensate on the same time-constant that the supply is working on.

I have this gut feeling that somebody in engineering decided they could decouple it and expose it to the user for the purpose of marketing; everyone else followed suit; then we wrote a bunch of articles about *what* LLC does and *how* to tune it, but never really answered the question about 'why' we needed to be managing it in the first place...

 

[MrTeal]

I never understood this as well.

The mobo know the target/desired voltage (be it at idle or at partial load or at full load) based on either the VID or the user-specified Vcc, and it knows what the actual voltage is, so why isn't there a feedback loop involved to simply drive the actual voltage to the target voltage?

I'm guessing its just a cost (hardware) and validation (people) issue. You don't want to implement algos that drive the voltage feedback to aggressively or you'll risk creating destabilizing swings. But there are solutions to that, IP to be leveraged from other industries. So we are back to it just being a cost issue?

The problem isn't so much voltage drop during steady state due to resistance, it's overshoot due to a step change in current.
Vos = (ΔIout² x L) / (2 x Vout x C)
You can't compensate for voltage swings faster than your VRM switching frequency, so Intel specifies that you need to drop voltage at higher currents. Since the overshoot voltage rises with the square of output current, you can end up with quite large spikes at higher currents that can exceed the maximum VID spec of the processor. My understanding of the LLC feature on most motherboards is that they essentially override Intel's Voffset and Vdroop specifications.