Yuriman
Diamond Member
Over the past few days I've been playing with LLC on my ASRock Z77 Extreme4, and I'd like to share my experiences in my own thread.
__________
The idea behind LLC is to reduce vdroop experienced under high load. In theory, this allows you to run a lower offset or additional turbo voltage, which in turn reduces your partial load and/or idle voltages with a fixed "lowest stable" full load voltage. My testing has shown that the implementation of LLC on my particular board is actually counter-productive.
I decided to do my testing at 4.5ghz which draws around 120w under a Linpack load. My configuration is as follows:
offset voltage mode, +.005v
C3/C6/C package states disabled
My settings were otherwise completely stock (or on auto), aside from LLC which was my independent variable, and additional turbo voltage, which is my dependent variable.
One big limitation in my testing is that I have to rely on CPU-Z for voltage measurements.
__________
With LLC set to "auto" or "level 5" which I gather is the "off" setting or at least the lowest LLC, I need a +voltage of approximately +101mv which I have tested to be my minimum stable voltage with 48 hours of Prime small FFT, 48 hours of Prime blend, and 24 hours of IntelBurnTest. At LLC5 and with +101mv, CPU-Z reports that my load voltage can go as low as 1.248v while running IBT, and 1.256v under Prime small FFT. Using Prime blend on 1 thread, my un-drooped voltage is as high as 1.304v.
With that as a baseline, here are my voltages running the various levels of LLC with Prime small FFT with an unchanging +101mv:
LLC lvl5 (least LLC):
1 core Prime blend: 1.304v
4 core Prime sFFT: 1.256v
LLC lvl4:
1c 1.312v
4c 1.280v
LLC lvl3:
1c 1.328v
4c 1.328v
LLC lvl2:
1c 1.336v
4c 1.352v
I didn't test LLC at level 1 because at lvl2 it was already overcompensating for vdroop. This showed me that LLC settings higher than 3 will overcompensate for droop.
__________
Now here's where it gets interesting.
I next tried to lower my voltage with LLC at level 3 to the minimum required for stability. My expectation was that I would be able to get down to, or at least close to 1.248v as reported by CPU-Z, which was my stable 4-core load voltage at LLC5. I found that my lowest stable voltage was at +63mv, which resulted in a 1-core load voltage of 1.296mv and a 4 cores at 1.304mv! Anything lower than that would result in Prime crashing or a BSOD.
Next, I tried the same thing with LLC at lvl4, with similar results. My lowest stable vcore was at +82mv, which resulted in a 1-core voltage of 1.296v and 4 cores at 1.264v.
__________
What I gather is, LLC is actually counter-productive on my board. Enabling it at higher levels only serves to raise my minimum stable load voltage without doing anything to help my idle and low-load voltages.
IDC has had a positive experience with LLC on his Asus Maximus IV Extreme-Z though, so I would love some additional data. Would anyone else be willing to experiment with their minimum stable voltages with various levels of LLC?
__________
The idea behind LLC is to reduce vdroop experienced under high load. In theory, this allows you to run a lower offset or additional turbo voltage, which in turn reduces your partial load and/or idle voltages with a fixed "lowest stable" full load voltage. My testing has shown that the implementation of LLC on my particular board is actually counter-productive.
I decided to do my testing at 4.5ghz which draws around 120w under a Linpack load. My configuration is as follows:
offset voltage mode, +.005v
C3/C6/C package states disabled
My settings were otherwise completely stock (or on auto), aside from LLC which was my independent variable, and additional turbo voltage, which is my dependent variable.
One big limitation in my testing is that I have to rely on CPU-Z for voltage measurements.
__________
With LLC set to "auto" or "level 5" which I gather is the "off" setting or at least the lowest LLC, I need a +voltage of approximately +101mv which I have tested to be my minimum stable voltage with 48 hours of Prime small FFT, 48 hours of Prime blend, and 24 hours of IntelBurnTest. At LLC5 and with +101mv, CPU-Z reports that my load voltage can go as low as 1.248v while running IBT, and 1.256v under Prime small FFT. Using Prime blend on 1 thread, my un-drooped voltage is as high as 1.304v.
With that as a baseline, here are my voltages running the various levels of LLC with Prime small FFT with an unchanging +101mv:
LLC lvl5 (least LLC):
1 core Prime blend: 1.304v
4 core Prime sFFT: 1.256v
LLC lvl4:
1c 1.312v
4c 1.280v
LLC lvl3:
1c 1.328v
4c 1.328v
LLC lvl2:
1c 1.336v
4c 1.352v
I didn't test LLC at level 1 because at lvl2 it was already overcompensating for vdroop. This showed me that LLC settings higher than 3 will overcompensate for droop.
__________
Now here's where it gets interesting.
I next tried to lower my voltage with LLC at level 3 to the minimum required for stability. My expectation was that I would be able to get down to, or at least close to 1.248v as reported by CPU-Z, which was my stable 4-core load voltage at LLC5. I found that my lowest stable voltage was at +63mv, which resulted in a 1-core load voltage of 1.296mv and a 4 cores at 1.304mv! Anything lower than that would result in Prime crashing or a BSOD.
Next, I tried the same thing with LLC at lvl4, with similar results. My lowest stable vcore was at +82mv, which resulted in a 1-core voltage of 1.296v and 4 cores at 1.264v.
__________
What I gather is, LLC is actually counter-productive on my board. Enabling it at higher levels only serves to raise my minimum stable load voltage without doing anything to help my idle and low-load voltages.
IDC has had a positive experience with LLC on his Asus Maximus IV Extreme-Z though, so I would love some additional data. Would anyone else be willing to experiment with their minimum stable voltages with various levels of LLC?
Last edited:
