Baffled with my Core i9-9900K overclock (Asus Auto OC > Manual VCore) - Page 2 - Overclock.net - An Overclocking Community
Forum Jump: 

Baffled with my Core i9-9900K overclock (Asus Auto OC > Manual VCore)

Reply
 
Thread Tools
post #11 of 32 (permalink) Old 12-30-2019, 10:19 PM
New to Overclock.net
 
Join Date: Dec 2019
Posts: 36
Rep: 3 (Unique: 1)
Quote: Originally Posted by Falkentyne View Post
Don't use LLC7 or 8 ever. It makes your transients worse and worsens your minimum voltage floor (making it higher) for high load vcore stability. LLC5 is the best balance. LLC5 allows a lower load vcore than LLC6 because the transient dips are smaller. Just raise the bios voltage a bit to compensate for the increased vdroop.

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

BTW on these boards the spikes arent the issue contrary to what some are telling you except at very high bios voltages like >1.45v. The spikes happen at load release (low current) and won't become an issue unless they exceed 1.5v on an oscilloscope capture. Its the dips that are the problem.
My issue with not using high(er) LLC is that your idle voltage ends up being high. On my Z390 Dark, I'm currently running 5.1/4.8 with 1.335v in BIOS with +25% vdroop. That gives me 1.295v under load. My idle voltage is high as hell now, though. Higher LLC and I can bring the idle closer to my load voltage.

I tried messing with adaptive voltage today (using static/override right now) but I don't understand how it works. I was ending up with my idle voltage too low and then my load voltage would be through the roof. Got frustrated and went back to static/override.
acoustic is offline  
Sponsored Links
Advertisement
 
post #12 of 32 (permalink) Old 12-30-2019, 11:09 PM
New to Overclock.net
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 6,807
Rep: 418 (Unique: 293)
Quote: Originally Posted by acoustic View Post
My issue with not using high(er) LLC is that your idle voltage ends up being high. On my Z390 Dark, I'm currently running 5.1/4.8 with 1.335v in BIOS with +25% vdroop. That gives me 1.295v under load. My idle voltage is high as hell now, though. Higher LLC and I can bring the idle closer to my load voltage.

I tried messing with adaptive voltage today (using static/override right now) but I don't understand how it works. I was ending up with my idle voltage too low and then my load voltage would be through the roof. Got frustrated and went back to static/override.
Did you look at the link I gave you?
Bringing your idle voltage closer to your load voltage causes your transient voltage dips to increase.
You do NOT want your idle voltage and your load voltage so close.

Look at the chart again please.

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

Look at LLC7 and LLC8. Do you know what part of the line is your idle voltage? it's the far left and far right.
The load voltage is the line in the middle.

Now do you see that DIP below the load voltage?
*THAT* is what causes your computer to crash randomly. Sensors WILL NOT SHOW THIS. It will only show the averaged lines that you see. Those transient dips and spikes only last for several (up to 40'ish) microseconds. But that's enough (the dips) to crash something.

I'm fully aware of "what you want" but this is a case of you having to learn that you can't have what you want. You have to accept reality and use what works best. Check the maximus XI 9900ks thread in this same section. You will see that SEVERAL people tested LLC5 and LLC6 and they all found that LLC5 gives them a *LOWER* true load voltage on "RMS" Vcore (in HWinfo64) than LLC6. They even posted numbers.

But what you seem to want is LLC8. But LLC8 will cause your voltage to repeatedly drop 70+mv below the vcore shown at *light* load. It's worse at a heavy load (like 100 amps). You're talking about a 100mv drop basically.
Elmor tested a 100 amp prime95 run with LLC8 vs LLC1.

https://www.overclock.net/forum/6-in...l#post28022572

You can see that LLC8 had the voltage rise 100mv above the idle voltage in spikes and the voltage dip 100mv below the idle voltage. That means if you used LLC8 and set 1.25v in your BIOS, your voltage would repeatedly drop down to 1.15v (!) during medium loads over and over but the vcore sensor, which only shows RMS, would stay at 1.25v the entire time..

With LLC1, the vdroop is massive of course, but the transient "drop" below the load vcore shown would only be about 10-20mv.

LLC5 is the best balance.
I honestly don't understand why you're fighting this information.

[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
Falkentyne is offline  
post #13 of 32 (permalink) Old 12-30-2019, 11:40 PM
New to Overclock.net
 
munternet's Avatar
 
Join Date: Jul 2014
Posts: 360
Rep: 13 (Unique: 7)
^^That's what I was trying to say.
I went and did some testing for myself and had some interesting results.
My coolant temp is 35c which is hot for here so these are extremes and each time I loaded the BIOS I ran CB R15 "5 times" in a row to really get those transients jumping
You can see the idle temps are THE SAME even though the voltages are different.
And though the LLC5 run showed lower load voltage it was actually more stable. I managed to get WHEA errors on the LLC6 run with higher load voltage :O
Attached Thumbnails
Click image for larger version

Name:	compare.jpg
Views:	11
Size:	464.9 KB
ID:	315064  


Intel Memory Overclocking Quick Reference https://www.overclock.net/forum/27784556-post7836.html
CPU
9900ks
Motherboard
Asus Maximus XI Gene
GPU
EVGA 1080 Ti Superclock 2 Black
RAM
F4-4400C19D-16GTZSW
Hard Drive
Samsung M.2 1TB 970 EVO Plus
Hard Drive
Samsung M.2 1TB 970 EVO
Power Supply
Corsair HX1200i
Cooling
EK 140D5 Pump, waterblocked CPU, Ram, GPU with 1 x 360mm and 1 x 420mm Radiators and Vardar Furious Fans
Case
ThermalTake V51
Operating System
Windows 10 Pro
Monitor
Asus PG349Q 34.1 inches 3440 x 1440 pixels @ 120 Hz
Keyboard
Corsair K95 RGB Platinum
Mouse
Asus Spatha Wireless Gaming
Mouse
Ligitech G602 wireless
Mousepad
1000x500x3
Audio
Steelseries Siberia800 wireless Headset
CPU
8700k @ 5.2ghZ
Motherboard
Maximus X Hero
GPU
EVGA 1080 Ti Superclock 2 Black
RAM
32GB 1xF4-4400C19D-16GTZSW & 1xF4-4400C19D-16GTZKK
Hard Drive
Samsung M.2 1TB 970 EVO Plus
Hard Drive
Samsung M.2 1TB 970 EVO
Power Supply
Corsair HX1200i
Cooling
EK 140D5 Pump, waterblocked CPU, Ram, GPU with 1 x 360mm and 1 x 420mm Radiators and Vardar Furious Fans
Case
ThermalTake V51
Operating System
Windows 10 Lite
Monitor
Asus PG349Q 34.1 inches 3440 x 1440 pixels @ 120 Hz
Keyboard
Corsair K95 RGB Platinum
Mouse
Asus Spatha Wireless Gaming
Mouse
Ligitech G602 wireless
Mousepad
1000mm x 500mm
Audio
Steelseries Siberia800 wireless Headset
CPU
i7-4770k
Motherboard
Asus Maximus VII Ranger
GPU
Gigabyte NVIDIA GeForce GTX980
RAM
G.SKILL Trident X 16GB 2 x 8GB 2400 Memory F3-2400C10D-16GTX
Hard Drive
Seagate Barracuda 2TB
Optical Drive
LG DVD Rom
Power Supply
PSU Coolermaster V1000W
Cooling
Custom Loop
Case
Case Thermaltake V51 Full Window
Operating System
Windows 7
Monitor
Shimian 2650x1440
Keyboard
Thermaltake Poseidon Z
Mouse
Logitech G602
▲ hide details ▲
munternet is offline  
Sponsored Links
Advertisement
 
post #14 of 32 (permalink) Old 12-30-2019, 11:55 PM
New to Overclock.net
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 6,807
Rep: 418 (Unique: 293)
Quote: Originally Posted by munternet View Post
^^That's what I was trying to say.
I went and did some testing for myself and had some interesting results.
My coolant temp is 35c which is hot for here so these are extremes and each time I loaded the BIOS I ran CB R15 "5 times" in a row to really get those transients jumping
You can see the idle temps are THE SAME even though the voltages are different.
And though the LLC5 run showed lower load voltage it was actually more stable. I managed to get WHEA errors on the LLC6 run with higher load voltage :O
Yes. 1.234v (LLC5 at max load) was more stable than 1.252v (LLC6 at max load), which is because the transient DIPS with LLC6 were *larger* than the RMS load difference between 1.234v and 1.252v (18mv difference).

This can be seen very obviously, without numbers (the numbers show a larger dip from RMS to minimum because the average voltage seems to "rise" over the measurement, which throws off the results--look at the graph itself).

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

You can see that the huge dip below the middle voltage line is MUCH larger at LLC6 than LLC5. In fact it's larger (substantially) than the vdroop 'decrease' (the amount the middle line gets closer to the idle voltage) going from LLC5 to LLC6.

Also people keep getting afraid of 'high' idle voltages. When using a "healthy" vdroop, a high idle voltage isn't a problem. It only becomes a problem if it's above 1.45v. If you can keep the idle voltage at 1.40v or below when using a fixed voltage and LLC5 (or lower LLC), then there's no problems. Only worry about a high idle voltage if you need to set a bios voltage higher than 1.40v.

Vdroop is people's friend. Not their enemy.

[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
Falkentyne is offline  
post #15 of 32 (permalink) Old 12-31-2019, 05:16 AM - Thread Starter
TotallydubbedHD on YT
 
Totally Dubbed's Avatar
 
Join Date: Oct 2012
Location: Somewhere in England - Sway
Posts: 11,392
Rep: 627 (Unique: 350)
Quote: Originally Posted by munternet View Post
By setting 7/8 llc aren't you creating large undetectable transients as the load comes on and off thus entering blue screen territory?
Maybe setting a higher idle voltage with llc 6, or even 5 will create smaller transients and not drop into blue screen territory when load comes on. The higher idle voltage won't be creating too much heat without the load/current draw?
Correct me if I'm wrong, I'm trying to work this out for my mate also
See this is news to me, thanks!

Quote: Originally Posted by Falkentyne View Post
Don't use LLC7 or 8 ever. It makes your transients worse and worsens your minimum voltage floor (making it higher) for high load vcore stability. LLC5 is the best balance. LLC5 allows a lower load vcore than LLC6 because the transient dips are smaller. Just raise the bios voltage a bit to compensate for the increased vdroop.

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

BTW on these boards the spikes arent the issue contrary to what some are telling you except at very high bios voltages like >1.45v. The spikes happen at load release (low current) and won't become an issue unless they exceed 1.5v on an oscilloscope capture. Its the dips that are the problem.
Now this is the type of info I've been looking for.
Eye opener for me; I've been testing all night really, trying to find the balance with my VCore at LLC 7 - now it's a fresh day and have restarted testing.
Before reading your comment I went back to default bios then redid my BIOS based on this video guide:
What do you make of his settings vs my manual ones I have on this thread? (minus LLC of course)

I took his settings went for LLC7 and 1.34V; temps hit the floor and I think the CPU throttled down to 4.6GHz (down from 5GHz) - at which point this method wasn't viable.
Now, I'm running 1.37V with LLC5 (as per your suggestion), with the YouTuber's BIOS settings (XMP etc)
Right now my PC is showing a rather large vDroop though - it's down to 1.279V (-100mV); is that fine?

EDIT: that crashed; now raised to 1.39V. VCore under a certain P95 test dropped VCore down to 1.219V :O

What do you make of that/suggestions ?

Thanks for your insightful knowledge, really helpful; never new about the transient dips.


Totally Dubbed is offline  
post #16 of 32 (permalink) Old 12-31-2019, 05:39 AM
New to Overclock.net
 
Join Date: Dec 2019
Posts: 36
Rep: 3 (Unique: 1)
You didn't give me a link. I'm not the OP.

Interesting read though. I'm curious if the DARK doesn't suffer those transient dips as bad as the Gene. I know the Maximus boards are pretty awful for Z390, but I was under the impression the 3 that were built properly were the Gene (which elmor uses in the test), the Apex, and the Extreme. Maybe that's as good as it gets.

My issue is figuring out which setting on the DARK relates to LLC5. I think that will put me at +50 or maybe +75% vdroop.
acoustic is offline  
post #17 of 32 (permalink) Old 12-31-2019, 07:36 AM
New to Overclock.net
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 6,807
Rep: 418 (Unique: 293)
Quote: Originally Posted by acoustic View Post
You didn't give me a link. I'm not the OP.

Interesting read though. I'm curious if the DARK doesn't suffer those transient dips as bad as the Gene. I know the Maximus boards are pretty awful for Z390, but I was under the impression the 3 that were built properly were the Gene (which elmor uses in the test), the Apex, and the Extreme. Maybe that's as good as it gets.

My issue is figuring out which setting on the DARK relates to LLC5. I think that will put me at +50 or maybe +75% vdroop.
Hmm. Guess I shouldn't post while sleepy. Quoted the wrong person
LLC5 equivalent on the Dark is at a bit less than 50% reduced vdroop. I would put it at close to 40% reduced vdroop.
The gap between LLC5 and LLC6 on the Gene/Apex seems to be greater than on the Master. Seems like LLC5 on the Gene is 1.0 mOhm while on the Gigabyte it's 0.8 mOhm. That would put it closer to LLC: Medium on the Gigabyte board (which is 1.0 mOhm).

LLC6 on the Gene and Master is 0.4 mOhm, which is the same as 75% reduced vdrop on the Dark/Turbo on the Gigabyte.

50% reduced vdroop on the Dark is equal to LLC: High on the Aorus Master (0.8 mOhm; Intel spec is 1.6 mOhm)
75% reduced vdroop on the Dark is equal to LLC Turbo on the Master and LLC6 on the Asus. (0.4 mOhm; Intel spec is 1.6 mOhm)

[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
Falkentyne is offline  
post #18 of 32 (permalink) Old 12-31-2019, 08:08 AM - Thread Starter
TotallydubbedHD on YT
 
Totally Dubbed's Avatar
 
Join Date: Oct 2012
Location: Somewhere in England - Sway
Posts: 11,392
Rep: 627 (Unique: 350)
Quote: Originally Posted by Falkentyne View Post
Hmm. Guess I shouldn't post while sleepy. Quoted the wrong person
LLC5 equivalent on the Dark is at a bit less than 50% reduced vdroop. I would put it at close to 40% reduced vdroop.
The gap between LLC5 and LLC6 on the Gene/Apex seems to be greater than on the Master. Seems like LLC5 on the Gene is 1.0 mOhm while on the Gigabyte it's 0.8 mOhm. That would put it closer to LLC: Medium on the Gigabyte board (which is 1.0 mOhm).

LLC6 on the Gene and Master is 0.4 mOhm, which is the same as 75% reduced vdrop on the Dark/Turbo on the Gigabyte.

50% reduced vdroop on the Dark is equal to LLC: High on the Aorus Master (0.8 mOhm; Intel spec is 1.6 mOhm)
75% reduced vdroop on the Dark is equal to LLC Turbo on the Master and LLC6 on the Asus. (0.4 mOhm; Intel spec is 1.6 mOhm)
Update from me: 5GHz at 1.4V failed with LLC set to 5.
I'm a little lost right now - not really sure where to go from here. Could you chime in ?

Also, could you just let me know - what's actually the purpose of OC'ing a 9900K?
Unless you want to manually force 5Ghz all the time, most would like the SpeedStep function, whereby the CPU doesn't run at 5GHz all the time - in this case, stock setting basically do this - don't they?
3.6GHz base, 5GHz Turbo, albeit on Core 1-2 only while the rest run at 4.7-4.8GHz, which is practically fine unless you're trying to prove something - am I missing something here?

As unlike my 3770k - which ran at around 3.2 stock with turbo at 4.2, my 4.5ghz OC actually WAS an OC.


Totally Dubbed is offline  
post #19 of 32 (permalink) Old 12-31-2019, 08:11 AM
New to Overclock.net
 
Join Date: Dec 2019
Posts: 36
Rep: 3 (Unique: 1)
Quote: Originally Posted by Falkentyne View Post
Hmm. Guess I shouldn't post while sleepy. Quoted the wrong person
LLC5 equivalent on the Dark is at a bit less than 50% reduced vdroop. I would put it at close to 40% reduced vdroop.
The gap between LLC5 and LLC6 on the Gene/Apex seems to be greater than on the Master. Seems like LLC5 on the Gene is 1.0 mOhm while on the Gigabyte it's 0.8 mOhm. That would put it closer to LLC: Medium on the Gigabyte board (which is 1.0 mOhm).

LLC6 on the Gene and Master is 0.4 mOhm, which is the same as 75% reduced vdrop on the Dark/Turbo on the Gigabyte.

50% reduced vdroop on the Dark is equal to LLC: High on the Aorus Master (0.8 mOhm; Intel spec is 1.6 mOhm)
75% reduced vdroop on the Dark is equal to LLC Turbo on the Master and LLC6 on the Asus. (0.4 mOhm; Intel spec is 1.6 mOhm)
No biggie! I appreciate the info. Thanks for sharing the conversions from the DARK to the ASUS. I haven't ran anything other than an ASUS board in a long time, so it's a learning process with this EVGA board

Currently, I'm running +25% (more vdroop). So by running PLUS 25% (more vdroop), I'm actually around LLC2, it would seem. I will have to try -25% (reduced vdroop), I think that will put me where I want to be.

Quote: Originally Posted by Totally Dubbed View Post
Update from me: 5GHz at 1.4V failed with LLC set to 5.
I'm a little lost right now - not really sure where to go from here. Could you chime in ?

Also, could you just let me know - what's actually the purpose of OC'ing a 9900K?
Unless you want to manually force 5Ghz all the time, most would like the SpeedStep function, whereby the CPU doesn't run at 5GHz all the time - in this case, stock setting basically do this - don't they?
3.6GHz base, 5GHz Turbo, albeit on Core 1-2 only while the rest run at 4.7-4.8GHz, which is practically fine unless you're trying to prove something - am I missing something here?

As unlike my 3770k - which ran at around 3.2 stock with turbo at 4.2, my 4.5ghz OC actually WAS an OC.
How is it failing? Is it hard-locking, BSOD, or shutting down the test? Not all 9900K are going to run 5Ghz all-core, but at 1.4V pretty much 99% of them will, afaik. Your temps must be out of control at that voltage, or are you running a custom loop? I would disable all the C-states and run it at 100% until you find a stable setting. I'm not going to lie to you, the Hero is a pretty awful motherboard. I had one before I replaced it with this EVGA Dark, and it was a PITA. After watching the videos of the VRMs from Buildzoid and how ASUS sacrificed quality to make more money with almost every board in the Maximus line, I will be avoiding Asus for a while. The power delivery on the Hero is ****, and I had a lot of issues overclocking memory that I just don't have with this Dark.

I would also move to ROG Realbench for your testing. P95 is just so finnicky now. I had similar weirdness with voltage dropping tremendously running P95.

What are your VCCIO and VCCSA voltages set to? I would set the RAM to straight XMP if you haven't already and work from there. There is definitely an advantage to running 5Ghz all-core. The numbers don't lie, and you can see the difference in games as well.

Last edited by acoustic; 12-31-2019 at 08:25 AM.
acoustic is offline  
post #20 of 32 (permalink) Old 12-31-2019, 08:28 AM - Thread Starter
TotallydubbedHD on YT
 
Totally Dubbed's Avatar
 
Join Date: Oct 2012
Location: Somewhere in England - Sway
Posts: 11,392
Rep: 627 (Unique: 350)
Quote: Originally Posted by acoustic View Post
No biggie! I appreciate the info. Thanks for sharing the conversions from the DARK to the ASUS. I haven't ran anything other than an ASUS board in a long time, so it's a learning process with this EVGA board

Currently, I'm running +25% (more vdroop). So by running PLUS 25% (more vdroop), I'm actually around LLC2, it would seem. I will have to try -25% (reduced vdroop), I think that will put me where I want to be.



How is it failing? Is it hard-locking, BSOD, or shutting down the test? Not all 9900K are going to run 5Ghz all-core, but at 1.4V pretty much 99% of them will, afaik. Your temps must be out of control at that voltage, or are you running a custom loop? I would disable all the C-states and run it at 100% until you find a stable setting. I'm not going to lie to you, the Hero is a pretty awful motherboard. I had one before I replaced it with this EVGA Dark, and it was a PITA. After watching the videos of the VRMs from Buildzoid and how ASUS sacrificed quality to make more money with almost every board in the Maximus line, I will be avoiding Asus for a while. The power delivery on the Hero is ****. I had a lot of issues overclocking memory that I just don't have with this Dark.

What are your VCCIO and VCCSA voltages set to? I would set the RAM to straight XMP if you haven't already and work from there.

There is definitely an advantage to running 5Ghz all-core. The numbers don't lie, and you can see the difference in games as well.
I mean either something is going wrong or I've got the worst i9-9900K to date.
So to answer your questions:
-It either BSOD's or just restarts and boots to Windows on its own. I've had P95 randomly crash after an hour of benching too, which I read is also a sign of instability
-I'm running on air via a Noctua D15 - served me well and idle temps sit at around 25-30c; all of that element seems fine
-I don't want to disable C-states as in the long run I want the CPU to be throttling; I want to ultimately be on an offset configuration; regardless, I've had issues with C-States also disabled, so I've ruled that out as a possibility too...again, do say if I'm missing something here.
-It is set to XMP, the rest is on Auto, as per most guides I've read/watched

I've had an idea - could it be BIOS related? I'm on the latest BIOS, and a quick google seems to indicate some have had issues with it...I've now rolled back and going to run benches on STOCK (Asus multi disabled, intel forced) + XMP only.
If this now fails, there's a hardware problem...

As for the OC, but we're speaking 0.2-0.3Ghz right? On 6 out of the 8 cores; at which point this will only impact a very minor number of games - I'm just trying to get my head around it all.
I remember I had the same sort of thoughts with my 3770k: could have it run at 4.8GHz at very high temps, or run at 4.5GHz at much lower/acceptable temps. The difference in synthetic benches were minimal and were near non-existent in BF3/BF4 benches I did back in the day.


Totally Dubbed is offline  
Reply

Tags
auto asus oc core i9 9900k , core i9-9900k overclock , overclocking core i9 , problems with overclocking core i9 9900k

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