i9 9900K overclocking - confused, what to do with AVX, LLC and OCCT stress test fail - Overclock.net - An Overclocking Community
Forum Jump: 

i9 9900K overclocking - confused, what to do with AVX, LLC and OCCT stress test fail

Reply
 
Thread Tools
post #1 of 17 (permalink) Old 05-01-2020, 07:55 AM - Thread Starter
Zireael
 
Wolfsblvt's Avatar
 
Join Date: Jun 2015
Location: Germany
Posts: 56
Rep: 1 (Unique: 1)
i9 9900K overclocking - confused, what to do with AVX, LLC and OCCT stress test fail

Heya.

I've recently finished my rig with a custom watercooling loop. I got the ASUS ROG Maximus XI Formula board and the i9 9900K CPU.
Watercooling temps are a non-issue. Got a MO-RA3, so I can cool quite a lot of heat.

Started with overclocking following guides like
by der8auer and
by Boosted Media.
I learned a lot of new things with those guides and googling a lot, even if it isn't my first OC.

My rig is mostly for gaming or photo editing.


Settings overview

I wanted to reach 5.1 GHz, and it seems like my chip may be able to do that. (5.2 GHz is impossible, crashes even in Cinebench at 1.45v)
I want to be as stable as possible though.

BCLK: 100
Core: 51
Cache: 48
Vcore: 1.36v (fixed)
DRAM voltage: 1.35v
VCCIO voltage: 1.15v
SA voltage: 1.10v
SVID: disabled
LLC: 7
AVX offset: 0

Here is an album of my exact BIOS settings, if this helps.

Vcore reads 1.305v during Intel Stress Test, 1.296v during Prime95 in CPU-Z.


Stresstest results

Cinebench r25 is the first test I do. Passes several runs without issues. (Avg 82°, Max 85°)
IntelBurnTest Maximum is fine as well. (Avg 68°C, Max 75°C)
Prime95 29.8 small FTT without AVX ran for 8 hours. High temps, but stable without errors. (Avg 86°C, Max 93°C)

I can't for the love of god run OCCT stress test. Neither the Linpack test nor the small data set test work. Both always crash right away. Medium works fine. (Avg 66°C, Max 69°C)
Can I just ignore those tests, because many sources seem to claim Prime95 should be enough, combined with the other stress tests I did?


Questions / Unclarity

Now I read a lot of differing opinions on stuff like LLC and AVX offset, and even more things. Which is really confusing. I know there a different ways to approach overclocking, and often different things all work, but I need something that works for me. Guess AVX offset could "fix" my OCCT test. Can run the small dataset with offset of 2. Getting a lot of temp errors though, because temps hitting 95°C regularly.

So is that high LLC really bad? With my voltage even the spikes of the vdroop shouldn't hit unsafe values, as far as I can guess. Tried lower LLC with 5, but I need to get a much higher Vcore in BIOS and it feels much less stable.
I can't run Prime95 with AVX. But I've read setting AVX offset higher is not that great because of fluctuating clock speeds in a lot of games then. Can I just ignore AVX then?

Haven't even started trying to squeeze a little bit more out of my RAM, and no GPU overclocking yet. Wanted to get the CPU stable first.

Any help how I can approach my OC and improve it is really appreciated.
Even if its telling my I should probably just forget it and go to 5.0 GHz with a lower voltage and much lower temps. Just thought with my cooler I could try it..


Last edited by Wolfsblvt; 05-01-2020 at 07:56 AM. Reason: Title
Wolfsblvt is offline  
Sponsored Links
Advertisement
 
post #2 of 17 (permalink) Old 05-01-2020, 08:49 AM
New to Overclock.net
 
reflex75's Avatar
 
Join Date: Aug 2016
Location: Paris, France
Posts: 132
Rep: 1 (Unique: 1)
Your LLC 7 is too high!
Highest recommended from my own experience on Maximus XI series MB is LLC 5, because of the transient response:
https://elmorlabs.com/index.php/2019...ne-visualized/

Decrease your LLC and increase your voltage to compensate.
You will be more stable and with lower load voltage, which is the most important (some vdroop is healthy).

But I'm afraid you will be out of spec from intel recommended max voltage for 8 cores CPU, which is: 1,52v - vdroop
vdroop = R * i
R = 1.6 mOhm (Intel 8 cores)
i = amps (watts/voltage)

For instance, if your CPU is at 1.30v voltage to run a load of 150w, then you're OK (because max is 1.34v = 1.52 - (1.6/1000 x (150/1.30))

Thanks to your custom watercooling, the limited factor is not temperature as usual with this CPU, but max safety voltage...

Last edited by reflex75; 05-01-2020 at 08:53 AM.
reflex75 is offline  
post #3 of 17 (permalink) Old 05-01-2020, 08:59 AM - Thread Starter
Zireael
 
Wolfsblvt's Avatar
 
Join Date: Jun 2015
Location: Germany
Posts: 56
Rep: 1 (Unique: 1)
Quote: Originally Posted by reflex75 View Post
Your LLC 7 is too high!
Highest recommended from my own experience on Maximus XI series MB is LLC 5
Isn't the default setting of this board already LLC 6? Not sure anymore, but I feel like I remember this.

As I said in my post, I tried LLC 5 already, but benchmarks were crashing even at 1.42v BIOS vcore.
What do you suggest for settings in this regard? Even higher vcore setting? I could try that again soon.

Wolfsblvt is offline  
Sponsored Links
Advertisement
 
post #4 of 17 (permalink) Old 05-01-2020, 10:27 AM
New to Overclock.net
 
Arni90's Avatar
 
Join Date: Apr 2011
Location: Norway
Posts: 1,491
Rep: 87 (Unique: 80)
Quote: Originally Posted by Wolfsblvt View Post
Cinebench r25 is the first test I do. Passes several runs without issues. (Avg 82°, Max 85°)
IntelBurnTest Maximum is fine as well. (Avg 68°C, Max 75°C)
Prime95 29.8 small FTT without AVX ran for 8 hours. High temps, but stable without errors. (Avg 86°C, Max 93°C)

I can't for the love of god run OCCT stress test. Neither the Linpack test nor the small data set test work. Both always crash right away. Medium works fine. (Avg 66°C, Max 69°C)
Can I just ignore those tests, because many sources seem to claim Prime95 should be enough, combined with the other stress tests I did?
You should test with AVX stability in mind, I'd recommend a lower clock and an offset of 0, as most games these days use AVX anyway.
In many cases, you will be more stable with a flat frequency at any rate.
Start out at 50x core and 46x for cache, it's generally a bad idea to overclock two things at the same time.

The reason you're instantly rebooting during AVX workloads is because you're not stable, I'm not sure if it's OCP, OTP, or some other inbuilt protection kicking in, but it's a sign that you're not stable.

Quote: Originally Posted by Wolfsblvt View Post
Now I read a lot of differing opinions on stuff like LLC and AVX offset, and even more things. Which is really confusing. I know there a different ways to approach overclocking, and often different things all work, but I need something that works for me. Guess AVX offset could "fix" my OCCT test. Can run the small dataset with offset of 2. Getting a lot of temp errors though, because temps hitting 95°C regularly.

Any help how I can approach my OC and improve it is really appreciated.
Even if its telling my I should probably just forget it and go to 5.0 GHz with a lower voltage and much lower temps. Just thought with my cooler I could try it..
A 9900k isn't really limited by cooling area, a 360mm radiator with 1000 RPM fans will handle the heat just fine, it's the heat density that starts becoming an issue around 1.30V load voltage. Higher flow rate, a better waterblock, liquid metal, and/or delidding are all tweaks which might help.

A static voltage or offset voltage won't change much temperature-wise, but too high LLC will be a stability killer. You will have better luck with higher VRM switching frequency.

Default LLC setting is 1

Main rig
(16 items)
CPU
9900k
Motherboard
Z390 AORUS MASTER
GPU
RTX 2080 Ti
RAM
4x 8GB B-die @4000 MHz 16-16-16-37-2T
Hard Drive
960 EVO
Hard Drive
860 EVO
Hard Drive
ADATA SX8200 PRO 1TB
Optical Drive
No
Power Supply
Fractal Design ION+ 860W
Cooling
MCP655
Cooling
Alphacool RTX 2080 Ti full cover block
Cooling
EKWB Supremacy EVO
Cooling
3x 180mm radiator
Case
SST-FT02
Operating System
Windows 10
Monitor
EX3501R
▲ hide details ▲


Arni90 is offline  
post #5 of 17 (permalink) Old 05-01-2020, 10:52 AM
Facepalm
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 7,346
Rep: 440 (Unique: 306)
Quote: Originally Posted by Wolfsblvt View Post
Isn't the default setting of this board already LLC 6? Not sure anymore, but I feel like I remember this.

As I said in my post, I tried LLC 5 already, but benchmarks were crashing even at 1.42v BIOS vcore.
What do you suggest for settings in this regard? Even higher vcore setting? I could try that again soon.
Please REMOVE the AVX offset and test again. Using an offset of any value destroys your stability in *NON AVX* or mixed loads due to a "issue" that must be improved in the future.
And LLC5+higher bios voltage gives you the lowest 'load' voltage floor overall. LLC6 would require about 10-15mv higher load voltage floor because of transients. But LLC7 is absolute terrible, avoid it!

[email protected] ghz, RX Vega 64, 32GB DDR4, Gigabyte Aorus Master, Seasonic Platinum 1000W, Corsair 760T
Alt: MSI GT73VR Throttlebook with 7820HK @ 4.7 ghz, GTX 1070 MXM TDP mod to 230W, 32 GB RAM

Quote: Originally Posted by sakete View Post
Well, I want you to know I have an academic degree in speculation.
Falkentyne is offline  
post #6 of 17 (permalink) Old 05-01-2020, 11:34 AM - Thread Starter
Zireael
 
Wolfsblvt's Avatar
 
Join Date: Jun 2015
Location: Germany
Posts: 56
Rep: 1 (Unique: 1)
Quote: Originally Posted by Falkentyne View Post
Please REMOVE the AVX offset and test again. Using an offset of any value destroys your stability in *NON AVX* or mixed loads due to a "issue" that must be improved in the future.
And LLC5+higher bios voltage gives you the lowest 'load' voltage floor overall. LLC6 would require about 10-15mv higher load voltage floor because of transients. But LLC7 is absolute terrible, avoid it!
I only did a small test this morning with AVX offset, everything else was with 0. Also my test with LLC5+higher voltage.
But as I said, it wasn't even stable in Cinebench at 1.42v in BIOS, from what I remember, so I considered it waste of time.

So you say LLC5, no AVX offset, and which BIOS voltage should I start testing at, if you have my load voltage values from above in mind with quite stable settings?

Wolfsblvt is offline  
post #7 of 17 (permalink) Old 05-01-2020, 01:11 PM
Facepalm
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 7,346
Rep: 440 (Unique: 306)
Quote: Originally Posted by Wolfsblvt View Post
I only did a small test this morning with AVX offset, everything else was with 0. Also my test with LLC5+higher voltage.
But as I said, it wasn't even stable in Cinebench at 1.42v in BIOS, from what I remember, so I considered it waste of time.

So you say LLC5, no AVX offset, and which BIOS voltage should I start testing at, if you have my load voltage values from above in mind with quite stable settings?
You need to determine your *accurate* minimum load voltage needed at LLC6 first. This is a Maximus XI and NOT a Strix, right? If it's a strix it gets hard because you don't have die-sense voltage readings, so what you see on "vcore" will be nothing like true readings.

If this were a Maximus XI, I could help you easily:

Die sense Vcore (same as VR VOUT on Gigabyte in HWinfo64) = Bios set voltage - (Loadline Calibration mOhms * CPU Current).
So if you were pulling 150 amps with a 1.350v BIOS set:

Maximus XI LLC5: 50% reduced loadline from 2.1 mOhms base=1.05 mOhms.
1350mv - ( 150 * 1.05 ) = 1.192v @ 150 amps

Maximus XI LLC6: 75% reduced loadline from 2.1 mOhms base= 2.1 * 25% (0.25)=0.525 mOhms:
1350mv - ( 150 * 0.525 ) = 1.271v @ 150 amps.

I don't know if that helps you any but HOPEFULLY that will give you an idea of how much you will need to increase BIOS voltage.

You can try up to 1.45v on LLC5 because LLC5 has a lot of vdroop.

Does your board have "Asus EC" in HWinfo64 which shows the current (Amps) reading?

[email protected] ghz, RX Vega 64, 32GB DDR4, Gigabyte Aorus Master, Seasonic Platinum 1000W, Corsair 760T
Alt: MSI GT73VR Throttlebook with 7820HK @ 4.7 ghz, GTX 1070 MXM TDP mod to 230W, 32 GB RAM

Quote: Originally Posted by sakete View Post
Well, I want you to know I have an academic degree in speculation.

Last edited by Falkentyne; 05-01-2020 at 01:19 PM.
Falkentyne is offline  
post #8 of 17 (permalink) Old 05-01-2020, 02:43 PM - Thread Starter
Zireael
 
Wolfsblvt's Avatar
 
Join Date: Jun 2015
Location: Germany
Posts: 56
Rep: 1 (Unique: 1)
Your comments are really helpful and thoughtful. Thank you so much for taking time to respond and explain everything!
Wanted to get this out first.

Quote: Originally Posted by Falkentyne View Post
You need to determine your *accurate* minimum load voltage needed at LLC6 first. This is a Maximus XI and NOT a Strix, right? If it's a strix it gets hard because you don't have die-sense voltage readings, so what you see on "vcore" will be nothing like true readings.
It is the ASUS ROG Maximus XI Formula, yes. The vcore reading nearly matches the value from CPU-Z. So I guess the voltage levels under load I have mentioned in the start post are the die-sense readings.

Quote: Originally Posted by Falkentyne View Post
Does your board have "Asus EC" in HWinfo64 which shows the current (Amps) reading?
It does. Te CPU Current in Prime95 without AVX goes up to 161 Ampere it seems. Cinebench and Realbench only go up to 152 Ampere.


So you've specified the loadline reduction for LLC5 and LLC6. I am currently running LLC7 though. Which percentage do I have to use for this?

If I follow your first formula I get a result of 0.367 mOhms. Does this make sense?
Vcore = Bios Voltage - LLC mOhms * Current
1.305v = 1.36 - x * 150A
=> x = 0,00036667

So if I need 1.305v under load to be stable, at LLC5 I should use
1.305v = x - 1.05 * 150A
=> 1.462v => 1.465v

Or if I want to calculate for the 160A then 1.473v => 1.475v.

But isn't 1.45v the absolute max that I should go for a stable overclock, like you said? Isn't heat an issue with that big of a voltage?
So that means I can't use LLC5, or I would have to go down in core clock like I feared?

Wolfsblvt is offline  
post #9 of 17 (permalink) Old 05-01-2020, 03:14 PM
Facepalm
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 7,346
Rep: 440 (Unique: 306)
Quote: Originally Posted by Wolfsblvt View Post
Your comments are really helpful and thoughtful. Thank you so much for taking time to respond and explain everything!
Wanted to get this out first.


It is the ASUS ROG Maximus XI Formula, yes. The vcore reading nearly matches the value from CPU-Z. So I guess the voltage levels under load I have mentioned in the start post are the die-sense readings.


It does. Te CPU Current in Prime95 without AVX goes up to 161 Ampere it seems. Cinebench and Realbench only go up to 152 Ampere.


So you've specified the loadline reduction for LLC5 and LLC6. I am currently running LLC7 though. Which percentage do I have to use for this?

If I follow your first formula I get a result of 0.367 mOhms. Does this make sense?
Vcore = Bios Voltage - LLC mOhms * Current
1.305v = 1.36 - x * 150A
=> x = 0,00036667

So if I need 1.305v under load to be stable, at LLC5 I should use
1.305v = x - 1.05 * 150A
=> 1.462v => 1.465v

Or if I want to calculate for the 160A then 1.473v => 1.475v.

But isn't 1.45v the absolute max that I should go for a stable overclock, like you said? Isn't heat an issue with that big of a voltage?
So that means I can't use LLC5, or I would have to go down in core clock like I feared?
LLC7 should not be used, period.
I suspect LLC7 is 85% reduced vdroop which makes it 0.15 * 2.1 mOhms = 0.315 mOhm.

All a higher LLC does to you is torpedo your transient response.
You can see very clearly here that the "Spike" below the load average increases a lot at LLC6 vs LLC5, compared to any lower level steps, and just gets worse from there. But that's because of the large step reduction in vdroop from LLC5 to LLC6 (50% reduced to 75% reduced vdroop). There really should be a step between LLC5 and LLC6. (1.05 mOhm and 0.525 mOhm is a large step).

https://elmorlabs.com/index.php/2019...ne-visualized/

Find what your STABLE load voltage is at LLC6.
Then find what bios voltage you need at LLC5 to get the SAME load voltage within 12mv.

Then reduce your bios voltage by about 10mv after, should keep the same stability because of better transients.

[email protected] ghz, RX Vega 64, 32GB DDR4, Gigabyte Aorus Master, Seasonic Platinum 1000W, Corsair 760T
Alt: MSI GT73VR Throttlebook with 7820HK @ 4.7 ghz, GTX 1070 MXM TDP mod to 230W, 32 GB RAM

Quote: Originally Posted by sakete View Post
Well, I want you to know I have an academic degree in speculation.

Last edited by Falkentyne; 05-01-2020 at 03:22 PM.
Falkentyne is offline  
post #10 of 17 (permalink) Old 05-02-2020, 05:39 AM - Thread Starter
Zireael
 
Wolfsblvt's Avatar
 
Join Date: Jun 2015
Location: Germany
Posts: 56
Rep: 1 (Unique: 1)
Woah. You are some kind of magician, right?

I was a bit annoyed after hours and hours of stress testing and thought the LLC5 can't work, I have tested this already, so I skipped your suggestion to first test stable voltage vor LLC6, and started with LLC5 right away.
Set it to 1.48v, just for a quick testing, because that was the number from testing. Was shocked that it worked. Cinebench passed. Prime95 didn't crash, was just giving temps up to 100c, so I stopped it fast. Reduced voltage by -0.2v, because I am impatient... Still worked, both tests, with lower temps. Got another -0.2 down and retested. Up until 1.40v, where I got a bsod in Prime95 after around 20 minutes. Increased voltage to 1.41v, another bsod after an half an hour or so.
Ran 1.42v over night in Prime95, must have been 11 hours now. No single crash, no thermal throttle, no WHEA logged.

Soooo... 1.42v vcore in BIOS seems to be the sweet spot. Which is still below the everywhere mentioned 1.45v cap you should not exceed, and only .06v higher than on LLC7.
The interesting part is - what you already said in your post - my load vcore is much lower. It goes down to 1.261v in Prime, while I still needed 1.296v with LLC7. That's a .035v drop, which seems amazing to me.
Temps are lower because of that. After 11 hours Prime95 my water temp leveled out at 29.4c, VRM at 52c and the CPU NEVER passed 90° in Prime95. Which is wasn't expecting at all... Average was 86c at CPU Max, Max was 89c.


Sadly, the Linpack test in OCCT still doesn't work. It takes some time to fill up all 64GB of RAM, then temps reach 100c, thermal throttling on Core2 kicks in (always Core2), and then it froze.
Small datasets in OCCT also does not work. Doesn't hit temp limit or anything, but I am getting the IRQL_NOT_LESS_OR_EQUAL bluescreen right after a few seconds.
Tried to upp the voltage to 1.44v but still same bluescreen. Is this still related to the CPU overclock or is RAM voltage an issue here now? Should I try to increase it first?

Sorry for rambling and asking so many questions.
I feel like I came quite a bit further, but I don't feel like my OC is more stable right now :-/

Wolfsblvt is offline  
Reply

Tags
9900k , Intel , llc , overclocking

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off