[SOLVED] 9700k stable @ 1.278v unstable @ 1.341v (under load value)

[chugginmilk]

Hi everyone,

Can somebody please answer this for me. Basically, I have a 5.0ghz O.C on my 9700k @ 1.345v with LLC set to 5; everything here is stable. While running prime95 small FFT (avx disabled) voltage drops to 1.278, which is understable as my LLC is pretty much at medium setting. So my thinking is that I could set my vcore to 1.3v and bump up LLC to 7 (generally only go to 6 but bumped it to 7 for the sake of this test) and the test would still run stable. Well I was wrong, during the same small FFT test vcore went to 1.341v (which is way higher than the 1.278v) but BSOD after less than 5 minutes. Does anyone know why this is and why my system becomes unstable with "more" vcore under load and a higher LLC setting? Anyone know why this is, or what changes I can make? Should I just keep it at LLC5 and 1.345v core? I'm thinking that a lower vcore setting is better with a high, not extreme, LLC than high vcore with low LLC, what are your thoughts? Thanks in advance to anyone that can help!

STABLE

UNSTABLE

 

[CompuTronix]

chugginmilk,

Overclocking is always limited by two factors; voltage and temperature. No two processors are identical; each is unique in voltage tolerance, thermal behavior and overclocking potential. Many processors have a "sweet spot" where they achieve their optimum overclock at ideal Vcore and Core temperatures, which for 14 nanometer processors, is typically less than 1.4 volts and less than 80°C.

When a particular processor is pushed beyond its optimum settings, some processor will become unstable, regardless of how much you increase Vcore or tweak the voltages for the various on-die devices such as Ring (Uncore), System Agent (SA), Integrated Memory Controller (IMC), Input/Output (I/O), Phase Lock Loop (PLL) and Load Line Calibration (LLC).

LLC is designed to compensate for "Vdroop" when heavy loads cause Vcore to significantly decrease or "sag". When LLC is set correctly, Vcore should remain relatively constant (within 25 millivolts or less) when transitioning from light to heavy loads, rather than "drooping", which can cause BSOD crashes. If LLC is set too high, then heavy loads can cause Vcore to surge or "spike" well above its BIOS setting, which can be damaging to the processor. So the objective is to set LLC to maintain Vcore at constant values during light and heavy loads, favoring a slight droop in preference to any amount of spike.

For example, If you set Vore to 1.300, then you need to find an LLC setting that doesn't sag or "droop" more than 25 millivolts, which would be 1.275. Keep in mind that even though most voltage values in BIOS use settings in increments of 0.005, the actual values step in increments of 0.008. This means that a decrease in Vcore of 25 millivolts will actually result in 24 millivolts, which is 3 increments of 0.008. If your motherboard has 12 phase VRMs with tight regulation, then about the best voltage regulation you can expect is within 2 increments or 16 millivolts.

In contrast, 4th generation Haswell and Devil's Canyon processors were excellent in this respect, as they featured Fully Integrated Voltage Regulators (FIVR). That made Vdroop virtually non-existent, so what-you-set is what-you-get.

CT

 

[ZeCasinhaOficial]

I agree with CompuTronix, I took a long time to find the correct configuration for my 9700k more I got, look at the picture with the results ...

https://www.clubedohardware.com.br/...8/hw.png.6c003b47dc81e438fe222ddde48c28d6.png

 

[tennis2]

What sampling rate do you need to set in order to properly capture vdroop? 1Hz, 2Hz, 10Hz?

 

[CompuTronix]

The default values in any of the popular monitoring utilities work fine without being modified. Vdroop is easy to visualize, especially when using utilities that have graphs. To avoid introducing unwanted Vcore fluctuations, it's necessary to use a steady-state 100% workload, such as P95 v29.8 Small FFT's (all AVX selections disabled). The difference between the Vcore value set in BIOS and the load Vcore value observed in a Windows utility (such as CPU-Z) will be obvious.

CT

 

[tennis2]

So the vdroop is a constant value? I always thought it was a short term "inaccuracy" that normalized back to the intended voltage after a short period of time of constant load.

 

[CompuTronix]

Yes, Vdroop is a constant "condition" which depends upon the LLC setting, and no, it doesn't normalize if LLC is set too low. You might be thinking of the initial moment of when load is first applied, which can result in an "overshoot" spike when LLC is set too high. To capture that very brief event an oscilloscope is needed.

CT