Help stabilize 9900KS overclock - water cooled!

[Alexauwa]

Hey guys,

I finalized my first custom water cooled pc and are very happy with my results. Still, I have some questions in terms of dialing in the right bios settings.

I run the following setup:

// PC
i9 9900KS @5.2 GHz
Gigabyte Z390 Aorus Master
64 RAM 3200 CL14
GTX 1080
1000 Watt PSU

// COOLING
2 x 360 radiator
EK Pump
EK CPU cooler
Heatkiller GPU cooler

I run the current bios settings:
CPU clock = 5.2 Ghz
Memory set to XMP
Loadline calibration = turbo (level 6)
Enhanced multi core performance = disabled
Uncore ratio = 47
C states = all disabled
CPU vcore current protection = extreme
Pmw phase control = extrem perf
Power limits are not limited
Rest voltages are set to auto

In terms of vcore and maybe related settings I have questions. I started off with a vcore of 1.290. I played around 1.290 and 1.310.
I tried Cinebench R20 and R15 a couple of runs. AIDA 64 as well and Prime95. All seem to work even with 1.290 vcore at the beginning. The longer the pc was running the more I run into crashes. My temps are absolutely fine. In between 80-85 on prime95.

So my question: is it normal that the vcore of 1.290 is not enough over time? Does it need a higher vcore over time? Do I have to keep raising the vcore? What additional bios setting would help stabilize the overclock apart from the vcore?

Memory is not overclocked and GPU got 200 MHz and 100 clock overclock though.

Thanks a lot for your help,
Alex

[Arne_Saknussemm]

From what I've seen of these CPUs...real stability at 5.2GHz takes more than 1.3v on average.

[Alexauwa]

Arne Saknussemm wrote:
From what I've seen of these CPUs...real stability at 5.2GHz takes more than 1.3v on average.

Thanks mate. On which tool do I get a exact read out of my current vcore under load?

[Arne_Saknussemm]

That's all core stability so you may run less for individual core OCs.

I've used CPUz and HWiNFO...for software readings...for what they are worth. RealBench is a good whole system stress test. If you pass an hour on that you're good to go for pretty much anything.

I guess you have avx offset? Or using older p95 version?

[Alexauwa]

Arne Saknussemm wrote:
That's all core stability so you may run less for individual core OCs.

I've used CPUz and HWiNFO...for software readings...for what they are worth. RealBench is a good whole system stress test. If you pass an hour on that you're good to go for pretty much anything.

Thank you.

So stability main relies on cvore or are there another voltage parameter that help stabilize?

So far Aida64 worked fine. Realbench seems to be harder somehow and gives me crashes.

I use hwinfo it was showing me cvore of 1.35 sometimes although I had 1.30 fixed. How is that possible and is it true? Does the loadline calibration does this offset?

[Arne_Saknussemm]

These CPUs are very Vcore dependent yes.

They also have their own algorithms for running depending on conditions.

Check this thread for an intro to complexity
https://rog.asus.com/forum/showthread.php?106375-MCE-explanations-and-others

[Alexauwa]

Arne Saknussemm wrote:
These CPUs are very Vcore dependent yes.[/url]

Here are screenshots of my vcore. It sits around 1.34 with almost 0 cpu usage. Also shows maxium vcore peaks of 1.42. Why is that? And can I trust these values?
Remember my vcore in bios is only set to 1.30. So what vcore shall I set? That is confusing me 🙂

When running cinebench R15 for example HWiNFO is showing me a vcore of only 1.302 ...cinebench R20 it shows me 1.35, that is confusing too .

[Arne_Saknussemm]

Are you setting manual voltage...or adaptive? And what power plan settings...
Any avx offset....latest Realbench and prime run avx and will pull a lot to run at high ghz.

[Alexauwa]

Arne Saknussemm wrote:
Are you setting manual voltage...or adaptive? And what power plan settings...
Any avx offset....latest Realbench and prime run avx and will pull a lot to run at high ghz.

I run 1.30 vcore fix (manual) at the moment. Power plan (you mean under windows?) is set to high power on all settings. Avx = 0.
Realbench hits the 1.42 on VID. But VID it not the vcore right?

To get this right... VID is the vcore with LLC kicked in. Giving me a total vcore load of for example 1.42 in real bench. But when not using benchmark and stress test in real bench instead, I get immediate crash. In order to have vcore and VID match, I set the LLC corret. In my case this is 6 (turbo on gigabyte). But having a vcore in bios set to 1.3 and a VID of 1.42 is big jump right? So I shall set my Vcore even higher to maybe 1.35?

[Falkentyne]

Alexauwa wrote:
I run 1.30 vcore fix (manual) at the moment. Power plan (you mean under windows?) is set to high power on all settings. Avx = 0.
Realbench hits the 1.42 on VID. But VID it not the vcore right?

To get this right... VID is the vcore with LLC kicked in. Giving me a total vcore load of for example 1.42 in real bench. But when not using benchmark and stress test in real bench instead, I get immediate crash. In order to have vcore and VID match, I set the LLC corret. In my case this is 6 (turbo on gigabyte). But having a vcore in bios set to 1.3 and a VID of 1.42 is big jump right? So I shall set my Vcore even higher to maybe 1.35?

Your Vcore in that screenshot is 1.24v-1.293v.
Use VR VOUT on Gigabyte boards to monitor. And this is an Asus forum. Did you really expect Gigabyte users to be here? 🙂

Asus maximus XI boards vcore sensor in HWInfo64 is like our VR VOUT (although 12mv resolution rather than 2mv).

VID has absolutely nothing to do with loadline calibration unless the mOhms value on your Vcore LLC is equal to the "DC Loadline" mOhms value, but VID is based on a scaling calibration from x8 to x50, for using Auto and DVID modes, is highly dependent on AC Loadline for how it affects VR VOUT, but will not correspond to VR VOUT on auto or DVID unless you fully understand the relationship between Thermal Velocity Boost Voltage optimizations, AC Loadline and DC Loadline, and any DVID Offsets in use do not affect VID whatsoever but will affect VR VOUT.

VID will *never* match VR VOUT on any fixed voltage because fixed voltage programs the VRM manually.