Hi guys, I just read this article.
Intel's comments confirmed my worst fears; Intel does NOT know what AC Loadline is doing to the processor's voltage request!
AC Loadline is supposed to be "predicted current" * mohm value. This was the system used for Z170-Z390.
Intel saying that AC Loadline and DC Loadline must match and that AC LL must match the Voltage regulator loadline (Loadline calibration) means they are using the old Z390 formula.
They are trying to target the native VID of the processor.
The problem is, this math falls apart quickly.
Because AC Loadline is not predicted current anymore. It is a fixed value * mohm value, and this fixed value depends on the SKU type and the number of cores and even the CPU ratio of the P cores!
On an i9, this value is 307, which is equal to the ICCMAX of the processor (307 amps).
This is where the wheels fall off the wagon.
Lets say the 5.6 ghz V/F point of a 14900K is 1.349v.
Or 1349mv.
Let's say AC loadline at the board level, if power delivery and design is "good enough" is set to 0.01 mohms.
DC loadline must match AC loadline, so DCLL is 0.01 mohms. VRM Loadline must also match, so Loadline calibration will be 0.01 mohms, or LLC8 (Asus), Ultra extreme (Gigabyte), Mode 1 (MSI) etc.
CPU's operating voltage at 50 amps of load will be:
1349mv + (ACLL mohms * 307) - (DCLL mohms * IOUT) + offset voltage
1349mv + (0.01 * 307) - (0.01 * 50) +0 = 1349mv = 1.349v.
CPU's operating voltage at 307 amps of load will be:
1349mv + (0.01 * 307) - (0.01 * 307) +0 = 1.349v.
LLC8 is the only situation where the V/F point is maintained at all times. But you have no vdroop whatsoever, transients will be terrible.
Now let's do the worst case scenario of a 1.1 mohm AC Loadline and a 1.1 DC Loadline and a 1.1 mohm VRM LL (Loadline calibration)/
1349mv + (1.1 * 307) - (1.1 * 50) + 0 =
1349mv + (337) - (55) =
1687mv - 55
1637mv = 1.630v !!!!!
The ACLL VID request will be 1.687v before vdroop.
At a 50 amp load, your VCORE will be 1.632v after vdroop !!!!!!!!
Note: this is based on the CPU being 100C. In reality, TVB voltage optimizations will be enabled so the lower the temp, the lower the voltage will be.
ANYONE can test this manually by setting your CPU to the max turbo ratio and setting AC and DC Loadline to 1.1 mohms and LLC to the same matching value.
1.1 mohms loadline calibration = LLC3 (Asus), Mode 7 (MSI), LLC: Standard (Gigabyte), etc: Note: LLC standard on gigabyte depends on the processor core configuration of enabled P/E cores.
At a 307 amp load:
1349mv + (1.1 * 307) - ( 1.1 * 307):
1349mv + (337) - (337) = 1.349v.
You are back at 1.349v only at exactly 307 amps.
Note that this is not exact as TVB Voltage optimizations will be enabled so voltage could be 0 to 150mv lower depending on the temp of the CPU.
The entire problem with this is that AC Loadline is not predicting anything. It's going to be a fixed value * the mohm value, no matter what. But DC loadline is predicted vdroop.
So the higher the value of ACLL, the operating voltage ends up scaling improperly, because ACLL is always going to "predict" 307 amps rather than the actual current you are truly using.
Why is AC Loadline using a base value of 307 on an i9, at the max turbo ratio, rather than using the SVID IOUT value?
. Ob dem tatsächlich so ist wird man dann sehen.
).
