cancel
Showing results for 
Search instead for 
Did you mean: 

i7-9700K - Getting stable 5.0GHz with adaptive mode

joppiano
Level 7
Hey guys.

I have the last 4 days worked on getting my i7-9700K stable on 5.0GHz with manual voltages, and i finally managed to make it happen.
My end goal was always to end up with adaptive voltages, and this is what I am working on now, so I hope some of you can help me archieve this.

First of all, this is my system:
CPU: Intel Core i7 9700K
CPU Cooler: NZXT Kraken X72
MOBO: Asus Rog Strix Z390-F GAMING
GPU: Asus Rog Strix RTX2070 O8G GAMING 8GB GDDR6
RAM: Corsair Vengeance RGB 16GB: 2x8GB DDR4 3000MHz
SSD: Samsung 970 EVO Plus MZ-V7S500BW 500GB (NVMe) - M.2 Card
HDD: Toshiba P300 3.5´´ 2TB - 7200rpm 64MB SATA-600
PSU: Corsair RM750x
CASE: NZXT H series H700i Window Black

Keep in mind, this is my very first time trying to overclock a CPU, so sorry in advance for the rookie questions and what not.

Stable 5.0GHz settings (w/ manual voltages)
Multiplier: 50
XMP: Enabled
AVX: Auto
MCE: Auto
SVID Support: Disabled
Load Line Calibration: Level 5
Manual Voltage: 1.385v (1.350v under load)
VCCIO: 1.250v
VCCSA: 1.150v
Max temp: 90c
Stresstest: Intel Burn Test, Realbench 8hrs, Prime95 v26.6 2hrs, OCCT Linpack nonAVX 1hr.

5.0GHz settings (w/ adaptive voltages)
Multiplier: 50
XMP: Enabled
AVX: Auto
MCE: Auto
SVID Support: Auto
SVID Behavior: Best Case Scenario
Load Line Calibration: Level 5
Adaptive Voltage: 1.370v (1.359v under load)
Voltage offset: Auto
VCCIO: 1.250v
VCCSA: 1.150v
Max temp: 85c
Stresstest: Intel Burn Test, Realbench 1hr, Prime95 v26.6 1hr. (I have to do some longer and more thorough stresstest)

Questions:
I am not sure nor certain if I have picked the right settings for adaptive, or is there other settings I need to dial on to make this adaptive stable?
Also, i tryid with adaptive voltage 1.360v but I failed pretty much instantly in Intel Burn Test, but passed with 1.370v.

What I have noted is, with manual voltage I have a load vcore at 1.350v, where with adaptive voltage I have a load vcore at 1.359v.
Is it perhaps possible to get that load vcore even lower or am I at a wall? Perhaps change voltage offset or such.

Thank you for reading through this.
Any critisism, inputs or feedback is very much appreciated.

Best regards
32,301 Views
22 REPLIES 22

ninezerofive
Level 10
LLC 5 is kind of bad in my experience, I was in a similar situation as you with it. 1.38v.

Try increasing LLC to 6 and lowering the voltage to ~1.33, with just this adjustment I was able not only keeping my 9700k 5g stable but even increase it to 5.2

Both of these settings seems kind of high as well :

VCCIO: 1.250v
VCCSA: 1.150v

There should be room for improvement, on 5.2 I think I am running 1.1 and 1.05 or the other way around (4000 C17) , and lowering both of these voltages I also dropped about 6 degrees during stress test.

Also for the whole BIOS setup I would advise to follow der8auer guide:

Remove the power limits as he advises, set the LLC to 6, see what kind of voltage can you cool but wih Kraken you should be good to hit more than 5g if the CPU is decent quality. Then try to see what kind of Core Cache you can do, leave as he does at the beginning 4300, when you are done adjusting the voltage and mhz you can do try to increase the Core Cache, mine can do 4900 but it depends on the chip. And lastly try to bring down the VCCIO and System Agent, I think you should be able to do both at 1.1v

For stress testing I would recommend the following steps :

Let's assume you start from beginning :

Adjust the bios, set llc to 6, sync all cores to 5G and set voltage to 1.3

stress test with prime 26.6 custom preset 8x8, run fft in place, for 1h ! monitor the temps, if they stay below 90C good.

If the system crashes increase the voltage by 0.5 to 1.305

Repeat the previous step until you find voltage you can run stable.

Repeat the prime custom test to confirm all dialed in values, cpu, core cache, vccio and system agent.

Only in the end run the long stress tests more than an hour, chances are if the system can stay stable for an hour on 8x8 prime then it can withhold anything you throw at it.

ninezerofive wrote:
For stress testing I would recommend the following steps :

Let's assume you start from beginning :

Adjust the bios, set llc to 6, sync all cores to 5G and set voltage to 1.3

stress test with prime 26.6 custom preset 8x8, run fft in place, for 1h ! monitor the temps, if they stay below 90C good.

If the system crashes increase the voltage by 0.5 to 1.305

Repeat the previous step until you find voltage you can run stable.

Repeat the prime custom test to confirm all dialed in values, cpu, core cache, vccio and system agent.

Only in the end run the long stress tests more than an hour, chances are if the system can stay stable for an hour on 8x8 prime then it can withhold anything you throw at it.


Please read my last reply regarding llc 6 etc.
I get vboost and not vdroops. What is your thoughts on this?

Also, when i ran prime95 v26.6, i used custom 1344-1344 fft. What is the difference from that to 8-8 fft ?

Another thing about stresstest.
With 1.350v manual, i passed 1344-1344 fft for 15min, but failed Intel Burn Test almost instantly.
What is your thought on IBT?

joppiano wrote:
Please read my last reply regarding llc 6 etc.
I get vboost and not vdroops. What is your thoughts on this?

Also, when i ran prime95 v26.6, i used custom 1344-1344 fft. What is the difference from that to 8-8 fft ?

Another thing about stresstest.
With 1.350v manual, i passed 1344-1344 fft for 15min, but failed Intel Burn Test almost instantly.
What is your thought on IBT?


With 8x8 you will know sooner if you have a problem with voltage or with max temperature.

I tried IBT in the beginning but gave up on it soon afterwards, it's a bit meh in my opinion. I went through most of the software and would recommend duo of Prime and Realbench to test cpu/cache/vccio/SA and memtest86 for ddr4 but to each their own.

First of all, let me get you up to speed.
I used the above settings (w/adaptive voltages), and i passed Realbench for 8 hours throughout the night. So happy about that.

ninezerofive wrote:
LLC 5 is kind of bad in my experience, I was in a similar situation as you with it. 1.38v.

Try increasing LLC to 6 and lowering the voltage to ~1.33, with just this adjustment I was able not only keeping my 9700k 5g stable but even increase it to 5.2


Well I did a lot of testing with LLC (on manual that is), and having LLC 6 gave me vboost. And i've read, heard and gotten told that I should work with vdroops instead, since vboost can be dangerous for the CPU coming off from load to idle or something.

LLC 5 worked perfectly for me, since it gives me a slight vdroop to work within.
Running with LLC 6 and 1.330v i wont pass any stresstests 😞

For instance, i've tryid adaptive 1.315v, LLC 6, and vcore was at 1.385v in windows. This is probably something to do with my VID being way higher than my set adaptive 1.315v. I've heard something about you can't force your voltages to be lower than what VID asks for, or something in that direction.

ninezerofive wrote:
Both of these settings seems kind of high as well :

VCCIO: 1.250v
VCCSA: 1.150v

There should be room for improvement, on 5.2 I think I am running 1.1 and 1.05 or the other way around (4000 C17) , and lowering both of these voltages I also dropped about 6 degrees during stress test.


This is interesting as well.
At ASUS' Default:
VCCIO: 1.312v - 1.344v
VCCSA: 1.216v

When I was on manual overclock, i tryid lowering these.
VCCIO: 1.250v
VCCSA 1.150v

Is what i landed on, where I was able to pass Intel Burn Test. I gotta add to that, that i was adjusting both of them at the same time.
So I can probably tweak it a bit more, by adjusting 1 by 1, and I will do that.

joppiano wrote:


For instance, i've tryid adaptive 1.315v, LLC 6, and vcore was at 1.385v in windows. This is probably something to do with my VID being way higher than my set adaptive 1.315v. I've heard something about you can't force your voltages to be lower than what VID asks for, or something in that direction.



Where are you getting that ? Use HWinfo64 and make sure you are checking vcore :

https://rog.asus.com/forum/showthread.php?110424-Apex-XI-overvolts-on-Manual-mode

ninezerofive wrote:
Where are you getting that ? Use HWinfo64 and make sure you are checking vcore :

https://rog.asus.com/forum/showthread.php?110424-Apex-XI-overvolts-on-Manual-mode


From this article.
https://rog.asus.com/forum/showthread.php?109183-Having-trouble-switching-to-Adaptive-voltage-mode-o...

ninezerofive wrote:
With 8x8 you will know sooner if you have a problem with voltage or with max temperature.

I tried IBT in the beginning but gave up on it soon afterwards, it's a bit meh in my opinion. I went through most of the software and would recommend duo of Prime and Realbench to test cpu/cache/vccio/SA and memtest86 for ddr4 but to each their own.

Ok, so in regard to prime95 v26.6, if I want to use 8x8, i just type in 8, and 8 in the other field, and tick "run fft in place". Just like I did with 1344 - 1344, yes?

Okay. I've heard good and bad about IBT. However the guy that assisted me to get my manual voltage stable, told me to use IBT as a quick reference, to see if my system could potentially be stable long term.

joppiano wrote:
From this article.
https://rog.asus.com/forum/showthread.php?109183-Having-trouble-switching-to-Adaptive-voltage-mode-o...


Ok, so in regard to prime95 v26.6, if I want to use 8x8, i just type in 8, and 8 in the other field, and tick "run fft in place". Just like I did with 1344 - 1344, yes?

Okay. I've heard good and bad about IBT. However the guy that assisted me to get my manual voltage stable, told me to use IBT as a quick reference, to see if my system could potentially be stable long term.


Yea bro, 8 in one box, 8 in other and check run in place. I prefer this over the IBT as it will give you the worst case scenario pretty much straight away, the temp wise and will show you the lack of vcore if you are experiencing it, but mostly it's time saving, if it's fine for an hour it will "generally" endure any longer test. (Who has time to run realbench for 8h to confirm the stability everytime you adjust slider by 0.5)