I9 9900K Maximus X Hero Adaptive Voltage

[neoforo87]

Hi there guys :).

Back in December I decided it was time to upgrade my old rig (I7 2600k) and build a new PC with the following configuration:

CPU: I9 9900k (P0 Revision) with Noctua NH-D15S
Mobo: Maximus X Hero
RAM: Corsair CMT32GX4M4K3600C16
GPU: EVGA RTX 2080TI XC Ultra
SSD: Samsung 860 Pro
HDD: Western Digital Black Caviar 1TB
ODD: Asus DVD Recorder
PSU: Seasonic Prime Ultra Titanium 850W
Case: Phanteks Enthoo Pro (no window side panel)
OS: Windows 7 Pro

I'm very happy with the system everything works fine. But because i know that cpu's usually get to much voltage with auto settings in the Bios I have been trying to find the

perfect voltage for the cpu.

After reading multiple guides I stumbled upon Raja Gill's guide for overclocking the I7 7700K https://edgeup.asus.com/2017/kaby-lake-overclocking-guide/, where he encourages
users to use adaptive voltage.

After reading it carefully I decided to just adjust the CPU voltage using the adaptive method but weird things started happening.

Here are the settings i have changed in the bios:

-X.M.P.on and MCE off (3600 CL16)
-SVID Behaviour Best-Case Scenario
-Cpu Load line Calibration LVL 5
-CPU Current Capability 140%
-Long Duration Packgae Power Limit 4095 (Max value that can be entered)
-Package Power Time Window 127 (Max value that can be entered)
-Short Duration Package Power Limit 4095 (Max value that can be entered)
-IA AC Load Line 0.01
-IA DC Load Line 0.01
-Cpu Core/Cache Current Limit Max 255.75
-Cpu Adaptive Voltage 1.250V
-Cpu VCCIO Voltage 1.15V
-Cpu System Agent Voltage 1.1V

Just to make it clear at this point i was not even trying to overclock the CPU. My only goal was to set the perfect voltage and unlock the cpu power capability since it was

downclocking in Prime95 and Aida 64 from 4.7GHz to 4.3GHz.

After saving the above mentioned Bios settings and booting up the OS, I started testing the stability with Prime 95(26.6 no AVX) and Aida 64.

When going on full load the Cpu Voltage fluctuates between 1.184V-1.200V (reported by CPUZ and HWinfo).

After thinking that i was fine I decided to test some games and fired up Battlefield 5. During the week of testing I had 1-2 times per week Battlefield 5 crashes to destop

with no error. First i thought that it might be the game since it's still kind of buggy but since this kept happening couple of times a week, I started investigating and found

out that BF 5 uses AVX instructions.
I found this out by setting an AVX offset in the bios.

So after encountering this issues i decided to do some more testing and came to the asumed conclusion, that the Cpu does not have enough voltage during load when using avx at

1.25V voltage and started increasing it up to 1.35V still in adaptive mode.
This made me even more confused when I fired up prime95(26.6) and saw that the voltage being reported through Cpuz and Hwinfo during load was still 1.84V-1.20 although minutes

ago I increased it in the Bios to 1.35V while still using LLC LVL 5.

So i went to the Bios once more and decided to increase the LLC to Lvl 6 while adjusting the voltage back to 1.25V adaptive voltage.
Once I was in windows i fired up the usual suspects again (Prime95 and Aida64) and this time around the voltage reported was 1.248V-1.280V.

Slowly but surely we're coming now to the last test and conclusion.

The last test that I ran was to see what happens when I go for 5GHz with a -3 Avx offset with the same Bios settings except for the LLC that i kept at LVL 6.
So when i fired up prime at 5GHz the voltage shot up to 1.35V-1.36V although it was set to adaptive mode at 1.25V.

So after writing this wall of text I have some questions regarding the above mentioned issues.

1.Is it normal for the voltage to drop from 1.250V or 1.350V to 1.184V-1.200V under load while using LLC lvl 5 with adaptive voltage?

2.Why does the voltage automatically jump from 1.250V adaptive voltage to 1.350V-1.360V when going from stock (4.7GHz all cores) to 5GHz while using LLC lvl 6?


Thank you in advance for your cooperation.

[Falkentyne]

All this is explained here.
https://rog.asus.com/forum/showthread.php?106375-MCE-explanations-and-others

The VID changes when the multiplier goes up, thus the change in voltage. And adaptive voltage cannot set a vcore lower than the CPU VID unless I'm wrong about that. I don't have these motherboards though.

[Carlyle2020]

1.Is it normal for the voltage to drop from 1.250V or 1.350V to 1.184V-1.200V under load while using LLC lvl 5 with adaptive voltage?

The numbers presented do not add up. I will ignore them and refer to wikipedias entry of "false dichotomy"
Check out "Vdroop". In your case it is somewhat appropriatetly managed by LLC 5 already. It is normal under heavy load.

2.Why does the voltage automatically jump from 1.250V adaptive voltage to 1.350V-1.360V when going from stock (4.7GHz all cores) to 5GHz while using LLC lvl 6?

For the sake of making sense of your numbers i will assume you ran an AVX load, like BF5 uses.
Also: AVX is used by windows in the background a little in milisecond bursts. Many of those little loads per minute in my case.

AVX loads charge more vcore than normal operations. More than you set anyway if you do use no AVX offset.
Use AVX-Offset : -2
Now you use around the same vcore for AVX and non-AVX loads with your hardware combo.

LLC5 is a good and longlasting choice.
turn XMP off since you already have set everything manually.
turn MCE on or set it on auto. Since why not?

Set a friggin multiplier, why won't you?

ps: Reading out HWinfo for example shows VID. It only shows what the chip thinks it would need from a pre-backed voltage table on the chip.
It is not a value of current running in your machine. Watch vcore for that.

Rgds
Carlyle

ps The system can run lower vcore than shown in VID