9900k 5GHz z390 Aorus Ultra: Looking for guidance on last hurdle - Overclock.net - An Overclocking Community
Forum Jump: 

9900k 5GHz z390 Aorus Ultra: Looking for guidance on last hurdle

 
Thread Tools
post #1 of 3 (permalink) Old 01-14-2020, 09:14 PM - Thread Starter
New to Overclock.net
 
Kurrgen's Avatar
 
Join Date: Oct 2015
Location: Illinois
Posts: 67
Rep: 6 (Unique: 5)
9900k 5GHz z390 Aorus Ultra: Looking for guidance on last hurdle

*TL;DR Warning: This is a long post and I wanted to provide as much info as possible, showing exactly what steps I took and what the results have been along the way. I think one has to enjoy small details and an analytical style for it, so I understand that not everyone has the time and/or desire to read it through. Additionally, maybe another new overclocker might find some information from my process that can help them. If you do not want to read it, you can skip down to Enter OCCT 5.5.1 Linpack 2019, which is where I am at now.

Summary: Currently at 5GHz 1.335v BIOS, High LLC, 45 Uncore, 1.20v VCCIO/SA, 0 AVX offset. Stable at everything I have thrown at it, except OCCT v5.5.1 Linpack 2019. There is not much info out there about people using Linpack for stability testing. It seems that most do not use it and I can understand why. Essentially, for what I do (gaming, browsing, light video and photo editing) I do not need to overclock. I am doing it to learn and do now, and when I upgrade my GPU next fall, then it will come into play more. And, of course, the tweaking and perfecting has become a bit addicting...just like tinkering with the build itself.

Background: I have been building my own PC's for six years or so. Additionally, I have done GPU overclocking for four years, and am comfortable with it. However, CPU overclocking is brand new to me. Sure, I have read about it and done some research on the basics over the years, but this time I am going all in. Right now, I am a little over two weeks into it, with the vast majority of that time being spent reading, researching, re-reading, and learning. I did not simply want to change voltages and settings, I wanted to know what they did, why I was changing them, and what effects occurred. My current knowledge level is that I am comfortable with using manual vcore set in bios along with vcore load line calibration and what occurs using them; ie: vdroop, vrout vs. vcore, etc. A big thanks to @Falkentyne , @elmor , and many more people here and on other sites. Without all of you and your experience, I would never have known what I do now.

Information:

Current rig and full parts list can be found here.

Current BIOS settings Gallery.

Current 5.0GHz HWinfo64 readings from yesterday showing 15hr spread. No gaming nor stress testing was done. Ambient room temp at time of pic was 77F (25C), fans kept at 60%, pump at 50% all day and night. *Note--room is on the upper floor and certain people just need to crank up the furnace, so can get quite warm. Normally, I keep the window cracked a little, but just let it be to give me an idea of what temps will be in summer when it is 95F (35C) outside.

Stock 4.7GHz MCE HWinfo64 readings for comparison. It seems that some Gigabyte boards come with MCE on for the 4.7GHz all boost. Only changes I made were to vccio.sa to lower them, and disabling onboard sound/graphics/VT-d. I keep this as a saved profile for reference now.

9900k is neither lapped nor delidded. I did a simple spin test by flipping it onto the IHS on a piece of glass and see if spun like a top. It did not spin one bit, and upon closer inspection with a straight edge (razor blade), I confirmed that thing is flat. TiM is MasterGel Maker Nano. Prior build was Hydronaut, and before that I used Pk-3.

Testing used to reach stability:

Spoiler!

*Testing Note -- When I first started attempting to overclock, I did not think about the vccio and vccsa that XMP applied. It set them to 1.254 and 1.296 respectively. Consequently, at first I did not touch them. Reading some of the guides, they mention them, but not just how important they can be, or they effects they can have. Consequently, as I tested and got errors, I would go in and bump the vcore. What I found was that my system would get more and more unstable (windows would close themselves and other odd behavior) as I increased vcore. The opposite of what I expected to happen. At 1.295 vcore, I stopped and dug more into it. When I lowered vccio/sa to 1.150v, I was more stable at a lower vcore and able to start again from 1.265v BIOS.

The Initial Result: *Note-- All below info is using High LLC (vcore loadline calibration bios) AVX offset=0, 1.200 vccio/sa.

At 1.330v BIOS 45 Uncore, I seemed to have a highly stable overclock (woohoo!), so I started to work on upping the Uncore. I bumped it to 46 and ran a few quick stress tests. Then, I loaded OCCT 5.5.1, begged it to be kind, and ran the OCCT HT enabled AVX2 Large Data Set ). Ran it for a couple hours (pics one, two, and three), with no occt errors or whea errors. It was 2am at this point, so stopped for the night. Sorry for the camera pics; I was beat and there was no way I was going near that keyboard and accidentally ending the test.

The next day I ran another: Blender Benchmark Full , 3600sec loop of Cinebench r20, and two hours of Realbench 2.56. All passed fine again with no errors.

Enter OCCT 5.5.1 Linpack 2019:

Throughout my learning process, I saw bits and pieces about Linpack testing with overclocks, but compared to everything else, there was not a whole lot. In any case, I had gotten this far and decided to give it a shot. Not knowing exactly how hard it would hit my system, I just used the recommended default settings and started it up. About 15 minutes in, no linpack errors, but HWInfo said "I laugh at your stability...here, have a Cache hierarchy error." Fuuuu--dge!!! I was so mad that I closed HWinfo64 and OCCT. A few more choice swear words later, I collected myself and went back to BIOS.

Bumped vcore to 1.335v with same 46 uncore. Rebooted and ran a 3600sec loop of Cinebench r20 and then a couple hours of Prime95 29.8 small fft no-avx (Temps and voltages ). No errors, so off to bed and dreams (nightmares?) of the Prime95 minions hammering away.

The next day it was time for more Linpack 2019. This time I got 35min in with no WHEA errors, but one error popped up in Linpack itself. Went back to BIOS and dropped the uncore to 45.

*Note--For testing here on out, I kept my window cracked as I normally do, which maintains an ambient of 75F +/-1F (24c). About 10 seconds or so after starting the tests, I would reset HWInfo64 to remove the idle values.

Rebooted and ran a full Blender benchmark to make sure no errors there. Loaded up OCCT 5.5.1 Linpack 2019 and away we go. Here is a shot of it at the 44min mark. Hooray...doing great. At 52min in there were still no linpack errors, but HWinfo popped a WHEA error Cache L0. From what I learned, I think that is related to uncore, vccio/vccsa.

Tried 1.34v BIOS, 45 Uncore, Kept 1.2 VCCSA and dropped VCCIO to 1.190. At 18min into Linpack, got a WHEA correctable error.

Next tried 1.34v BIOS, 45 Uncore, and put vccio back to 1.20. Result was that it popped 1 Linpack error at 35min mark, but zero WHEA errors. This is a shot of the 80A current pull and here is where it can pull 180A.

Odd...I was more stable at 1.335v. So went back down to 1.335v and ran, yet another 7200sec loop Cinebench r20, which went off without any errors again.

So, this is where I am at now. Prior to this point, I have been able to search on my own and find the answers to any questions that came up. But now, I am stuck, have had no luck finding the answers online, and am not sure what adjustments I should make. It seems that I may need to bump vccsa, but to what? 1.210? 1.250? And also, do I change uncore and raise it now? If so, do I leave vcore at 1.335 or push it up to 1.34 again? I feel I am close, yet not sure which to adjust and by what amount.

I know there are other Linpack programs out there, and I know I might be at the overkill stability test point, yet I am enjoying the ride and learning a ton. I am not looking to be NASA certified stable or anything; all I want to do is get an hour or so on this Linpack and I will be happy. But right now, it is like that one unchecked box on a list that gnaws at you. Not to mention, I happen to like my system very much and do not really want to kill it by constantly running high amps into it.

Thank you all in advance for your time and guidance.

*One final note-- If @Falkentyne reads this, could you please clarify that High on Gigabyte is 0.8 mohm and Turbo is 0.4 mohm. While I am pretty sure it was a mistype, in this post you say High is 0.4, yet here, and pretty much everywhere else, you say 0.8. Given that I am pulling 180A, I just want to double check and be sure about my voltages at that kind of current. Thanks!*

Kurrgen, Ravenwolf, Ravenbane....all depends on where you are
CPU
9900k
Motherboard
Gigabyte z390 Aorus Ultra
GPU
EVGA GeForce GTX 980 Ti 6 GB Superclocked+ ACX 2.0+
RAM
32GB G.Skill Ripjaw V Series
Hard Drive
Samsung 970 EVO Plus
Hard Drive
Samsung 850 EVO
Hard Drive
Samsung 850 Pro
Power Supply
EVGA G3
Cooling
Custom Water cooling loop
Case
Phanteks Enthoo 719 (Luxe 2)
Operating System
Windows 10 Home
Monitor
Dell AW3418DW 34.1" 3440x1440 120 Hz Monitor
Keyboard
Logitech 810 Orion Spectrum
Mouse
Logitech G502 Proteus
▲ hide details ▲

Last edited by Kurrgen; 01-14-2020 at 09:25 PM. Reason: Some pics were not direct linking properly.
Kurrgen is offline  
Sponsored Links
Advertisement
 
post #2 of 3 (permalink) Old 01-14-2020, 10:54 PM
New to Overclock.net
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 6,818
Rep: 418 (Unique: 293)
CPU L0 errors have to do with hyperthreading, the virtualized register store, and the duplicated instructions/threads.
The CPU L3 cache and the IMC are important for hyperthreading. That's why VCCIO (maybe SA also, I don't know) affects this so much. A CPU L0 error occurs when a hyperthreaded core (either physical or logical thread) generates a corrected error. You won't usually see this if Hyperthreading is disabled--instead an uncorrected error is usually the typical normal Clock Watchdog Timeout / WHEA uncorrectable Error everyone is used to traditionally.

Turbo Loadline Calibration (LLC) is 0.4 mOhms and High LLC is 0.8 mOhms.
You're probably confusing Loadline Calibration (Load Line Calibration=LLC) with something called AC/DC Load Line, which ALSO has mOhm values.

The difference is that AC Load Line works on the CPU power supply coming from (Or rather, TO, if I understand AC voltages vs DC voltages) the VRM from the +12v,
while DC load line is SUPPOSED To be Load Line Calibration, but it is NOT load line calibration, since motherboard vendors have their own setting for LLC. Instead, DC Loadline is for power measurements only, and affect the CPU VID that is reported to windows (CPU VID influences CPU Package Power (a MSR), which is equal to VID * Amps, in watts). DC Loadline (in mOhms) affects VID dropping, the EXACT Same way that Loadline calibration affects "VR VOUT" dropping (Amps * resistance, where amps is your Current IOUT in HWinfo64, and resistance is the mOhms value). The "original" VID is then subtracted by this value:

Like this:
Amps * resistance:
if DC loadline=1.6 mOhms, then at 100 amps of Current IOUT, VID would be REDUCED by:
Starting VID - ( 100 * 1.6 ) = Starting VID - 160mv = VID reported to operating system.

But you need the "original" starting VID first for this to make any sense. This is the VID that is sent to the VRM as a target voltage (On adaptive/offset modes only; on FIXED vcore, the VRM is directly set to a fixed voltage, before vdroop).

To get the original VID, set DC Loadline to 1 (0.01 mOhms), which completely removes the VID dropping, since 0 * amps is 0. Ohm's law and algebra.

AC Load Line is very difficult to explain, as it influences the CPU power supply in a strange way, where the higher the current load, the higher the AC Loadline's influence. Also the HIGHER the AC Loadline mOhms value, the higher the "Base VID" will be boosted. This is sent to the VRM as a source voltage (before vdroop is applied). This value is usually limited to 1.520v (the so-called "max VID" you may have heard about).
This max VID can be exceeded by using VRM command 33h (Listed in the Intel specification sheets as "Offset capability", do NOT confuse this with offset VOLTAGE, aka DVID--they are NOT the same!!), but please don't enable SVID offset to do that. It disables all voltage control (freezing the last known settings as the current setting) and seems to be bugged if set below 5 ghz.

Asus boards have "Serial VID offset mode" enabled by default.

Don't bother trying to get Linpack 2019 / LinX 0.9.6 / Linpack Extreme 1.1.1 / 1.1.2 with matching residuals stable on a non golden 9900k @ 5 ghz. It's harder to get matching residuals on LinX 0.9.6 (35000 sample size) / LinEX 1.1.2 than it is to get Prime95 FMA3 small FFT stable (29.8 build 6). And the heat output is absurd. I would only test this at 4.7 ghz at close to stock voltages.
LinX 0.9.6 (i use this because it's easier to use, who cares if it's in Korean) with 35000 sample size is much harder to get matching residuals in than it is to pass Prime95 small FFT FMA3. A mismatched residual (residuals different on each loop--they should be the same on each loop) means the system did a math calculation error. If LinX actually crashes, it means the calculation was so severe, it could not even continue.

When using linpack extreme, use the command /residualcheck. At least you had to on 1.1.1. I haven't tested 1.1.2.

[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 #3 of 3 (permalink) Old 01-15-2020, 12:27 AM - Thread Starter
New to Overclock.net
 
Kurrgen's Avatar
 
Join Date: Oct 2015
Location: Illinois
Posts: 67
Rep: 6 (Unique: 5)
Quote: Originally Posted by Falkentyne View Post

Turbo Loadline Calibration (LLC) is 0.4 mOhms and High LLC is 0.8 mOhms.
You're probably confusing Loadline Calibration (Load Line Calibration=LLC) with something called AC/DC Load Line, which ALSO has mOhm values.
I was referring to the load line calibration, which I believe in my BIOS reads as CPU Vcore Loadline Calibration, and not the one above it, which is CPU Internal AC/DC Load Line. As indicated in the pics, CPU Vcore is what is set to high, so I was just trying to clarify that is what you meant had 0.8 mOhms. I think I worded it wrong. What I was referring to is all of those posts you made showing the formula to help people estimate safe voltages at different LLC. Clarifying that at High LLC, I would use, for example, 1335 - (170A * 0.8) to compare to 1520-(170A*1.6)...in order to not exceed inputting too high a number into BIOS.

Thank you for going through the explanation of it again though. I read pretty much every post you made, all over the internet, regarding LLC before I even started trying to overclock. That is how I knew that I wanted to use High instead of Turbo. I knew using High that my vcore I entered in BIOS would be higher and there would be more vdroop, but that was not a bad thing. Learning from you, Elmor, and others, when I entered 1.335 in BIOS, I expected the Idle values 1.299 VR OUT/ 1.308 Vcore because of vdroop and it was good. I also knew that 1.260v Load was normal and what was supposed to happen.

Quote: Originally Posted by Falkentyne View Post

To get the original VID, set DC Loadline to 1 (0.01 mOhms), which completely removes the VID dropping, since 0 * amps is 0. Ohm's law and algebra.
Oh yeah, I read about this many times over. What I did not understand is where in the BIOS you meant. Was it here? And if so, can I change that now with the current overclock bios settings I am running? I understand they are just the voltages from tables that the cores are requesting at certain loads and not what they are getting, yet still interesting stuff.

Most of what you have considerately retyped to me, I have combed through and read many times over; found it informative and interesting. Like I said...I wanted to know why I was doing things and what they were. So, I apologize for not making that clear when I asked about Turbo vs. High. And no, not using SVID offsets, auto voltages. I read about that too from you, but first still understanding all the nuances of manual set voltage.

Quote: Originally Posted by Falkentyne View Post
Don't bother trying to get Linpack 2019 / LinX 0.9.6 / Linpack Extreme 1.1.1 / 1.1.2 with matching residuals stable on a non golden 9900k @ 5 ghz. It's harder to get matching residuals on LinX 0.9.6 (35000 sample size) / LinEX 1.1.2 than it is to get Prime95 FMA3 small FFT stable (29.8 build 6). And the heat output is absurd. I would only test this at 4.7 ghz at close to stock voltages.
I am not looking for extreme overclocks, nor do I know what constitutes a golden sample at full load. Simply learning and trying new things. Been pretty pleased with my chip. I will take a 9900k that is stable at 1.194v VR OUT/ 1.232 vcore reading while pulling 150 amps running Prime95 29.8 small fft (no avx). From what I have read, there are people who have 9900k's that need ti run over 1.30v at full load.

Regarding Linpack 2019..not in me to give up on it yet. I have the cooling capability and purposely designed my build around knowing the 9900k ran hot. Yeah, I am a bit stubborn, so will probably give it another go. Already got 52min into it with pulling 170A at 1.180 VR OUT...no crashes or temp issues, just the WHEA correctable error, and I just want 8min more

I don't plan on the other Linpacks that are out there, just the one I have now in OCCT 5.5.1. And that too, I read all about your experiences on LTT...PS: the fans ramping up and down annoyed me too. The first time I tried running it, I stopped it 10min in and set a fan curve just for it.

Thank You again. I learned a lot from you the past couple of weeks.

Kurrgen, Ravenwolf, Ravenbane....all depends on where you are
CPU
9900k
Motherboard
Gigabyte z390 Aorus Ultra
GPU
EVGA GeForce GTX 980 Ti 6 GB Superclocked+ ACX 2.0+
RAM
32GB G.Skill Ripjaw V Series
Hard Drive
Samsung 970 EVO Plus
Hard Drive
Samsung 850 EVO
Hard Drive
Samsung 850 Pro
Power Supply
EVGA G3
Cooling
Custom Water cooling loop
Case
Phanteks Enthoo 719 (Luxe 2)
Operating System
Windows 10 Home
Monitor
Dell AW3418DW 34.1" 3440x1440 120 Hz Monitor
Keyboard
Logitech 810 Orion Spectrum
Mouse
Logitech G502 Proteus
▲ hide details ▲

Last edited by Kurrgen; 01-15-2020 at 12:32 AM. Reason: Clarification
Kurrgen is offline  
Reply

Tags
9900k , overclock help , z390

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