9900k Direct Die Overclock Help Needed - Page 2 - Overclock.net - An Overclocking Community

Forum Jump: 

9900k Direct Die Overclock Help Needed

Reply
 
Thread Tools
post #11 of 34 (permalink) Old 08-15-2019, 05:53 PM - Thread Starter
New to Overclock.net
 
Join Date: May 2019
Posts: 24
Rep: 0
Only took 11 min of prime 95 small ffts full avx to get a peak of 101c while averaging in the 90s @ 5.1ghz Turbo Loadline and vcore at 1.335v manually. I happened to grab the screenshot of P95 right when voltage hit its highest peak of 1.38v on core but was taking 1.368 on core majority of the time. I pulled the plug on the test as soon as it hit 100c. Is it really possible that I lost THAT BADLY on the silicon lottery?!?!?! Like I feel like this should be able to do a whole lot better on direct die and liquid metal than this. There has to be something else I am missing?!?!?! What do I do? Leave this where it is? Take the uncore to 4.7 and try again? Or take it down to even 5.0ghz and try there?!?! Or am I taking this prime 95 testing way too seriously?!?! Everywhere says if it cannot pass full AVX on p95 for at least an hour its not really stable. What should I do?! I want to run this thing at the highest clocks I can for gaming and local recording while streaming. I know streaming and some games use AVX and I know P95 is an absolute ridiculous overkill for AVX instructions so how seriously should I take that test?! Because I can get it to pass MANNNNNNNYYY other benchmarks at much higher clocks with much better temps but they all almost instantly fail p95 small ffts full avx
Attached Thumbnails
Click image for larger version

Name:	prime51run.jpg
Views:	20
Size:	2.02 MB
ID:	289426  

Attached Images
File Type: bmp 190815204143.BMP (2.25 MB, 9 views)
File Type: bmp 190815204337.BMP (2.25 MB, 10 views)
poopsLIVE is offline  
Sponsored Links
Advertisement
 
post #12 of 34 (permalink) Old 08-15-2019, 06:01 PM - Thread Starter
New to Overclock.net
 
Join Date: May 2019
Posts: 24
Rep: 0
Quote: Originally Posted by Socks keep you warm View Post
Damn i have been out of the loop for years! That's so cool it used to be dangerous stuff running bare die setups as the CPU was only held in by the block nothing else lol. I would say you haven't got good contact with your block though. Have you checked your block? Might need lapping.
EDIT: The 9900K is also a very hot chip, i am literally going today to buy another 360 rad just to try and cool this thing down.
I have not checked the block no. I dont know exactly what I would lap. Doing the die would just make it harder to reach the cpu block if that were the case. Lapping the OC frame I dont think would be a good idea and I think the same goes for lapping the block, wouldnt that just make it harder to make contact? I was honestly thinking things couldve been too tight. I mean I can crank down a bit more on the block but I am not going to go overboard as it is already feeling resistance from the springs
poopsLIVE is offline  
post #13 of 34 (permalink) Old 08-15-2019, 06:32 PM
New to Overclock.net
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 5,987
Rep: 383 (Unique: 274)
Quote: Originally Posted by poopsLIVE View Post
Only took 11 min of prime 95 small ffts full avx to get a peak of 101c while averaging in the 90s @ 5.1ghz Turbo Loadline and vcore at 1.335v manually. I happened to grab the screenshot of P95 right when voltage hit its highest peak of 1.38v on core but was taking 1.368 on core majority of the time. I pulled the plug on the test as soon as it hit 100c. Is it really possible that I lost THAT BADLY on the silicon lottery?!?!?! Like I feel like this should be able to do a whole lot better on direct die and liquid metal than this. There has to be something else I am missing?!?!?! What do I do? Leave this where it is? Take the uncore to 4.7 and try again? Or take it down to even 5.0ghz and try there?!?! Or am I taking this prime 95 testing way too seriously?!?! Everywhere says if it cannot pass full AVX on p95 for at least an hour its not really stable. What should I do?! I want to run this thing at the highest clocks I can for gaming and local recording while streaming. I know streaming and some games use AVX and I know P95 is an absolute ridiculous overkill for AVX instructions so how seriously should I take that test?! Because I can get it to pass MANNNNNNNYYY other benchmarks at much higher clocks with much better temps but they all almost instantly fail p95 small ffts full avx
Ok Ok Ok.
First of all I'm going to shout loudly here.

DO NOT USE SMALL FFT AVX TO TEST WHETHER YOU WON THE SILICON LOTTERY OR NOT! You are exceeding the Intel maximum electrical specifications for that processor by SO MUCH that not only can barely ANY CPU pass prime95 at 5.1 ghz with that VR VOUT (see below), you are also going to slowly degrade that processor (or maybe more than slowly degrade it)--see below.
To test whether you won the silicon lottery, lay off the AVX small FFT/FMA3 worship. Use Realbench 2.56 two hour stress test, Cinebench R20, 3600 second custom stress test, and Prime95 29.8 build 5 small FFT preset with AVX and AVX2 *disabled*. This is all you need to do to avoid damaging your processor and driving yourself to the mental hospital in the process.
Also, are you aware that Silicon Lottery bins their golden processors with an AVX -2 offset?

Even THEY aren't crazy enough to test with small FFT AVX at 5.1 ghz. If they did, maybe 1% rather than 4-8% of their 5.1 ghz bin processors would pass, and they would need cooling solutions most people simply don't use.

You're reading the wrong vcore sensor.

CPU-Z is the Super I/O chip and is going to be wildly inaccurate. CPU-Z reads the ITE 8688E chip which tends to show higher than than the bios voltage you set. The ITE 8792E with LLC=Turbo shows a +/12mv voltage within bios voltage, regardless of your current, so even this is inaccurate.

You need to read VR VOUT. And your load voltage is 1.250v.

Also please reset your HWinfo fields (choose reset layout or reset to default layout or restore fields in the settings). You have two iGPU fields active (duplicated 0.004v VR VOUT which is the inactive iGPU).
And please don't run that prime95 test you are running!

You're pushing ***222 amps*** into that thing! That's WAY past the maximum Intel allowed specification of 193 amps! And at 193 amps, VR VOUT must not exceed 1.225v (tl;dr: the higher the amps, the lower the maximum safe VR VOUT allowed before your CPU risks faster than "normal wear and tear" degradation.

tl;dr:
if you can pass realbench 2.56 WITHOUT CPU CACHE L0 errors (shown in HWinfo64) at THESE SETTINGS:

x51 / x47
1.310v bios set (Fixed)
Loadline calibration: Turbo
Then you have anywhere between a very good to a golden processor (this depends on how low you can go on the bios set voltage without L0 errors).

[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

Last edited by Falkentyne; 08-15-2019 at 06:43 PM.
Falkentyne is offline  
Sponsored Links
Advertisement
 
post #14 of 34 (permalink) Old 08-16-2019, 12:08 AM - Thread Starter
New to Overclock.net
 
Join Date: May 2019
Posts: 24
Rep: 0
Quote: Originally Posted by Falkentyne View Post
Ok Ok Ok.
First of all I'm going to shout loudly here.

DO NOT USE SMALL FFT AVX TO TEST WHETHER YOU WON THE SILICON LOTTERY OR NOT! You are exceeding the Intel maximum electrical specifications for that processor by SO MUCH that not only can barely ANY CPU pass prime95 at 5.1 ghz with that VR VOUT (see below), you are also going to slowly degrade that processor (or maybe more than slowly degrade it)--see below.
To test whether you won the silicon lottery, lay off the AVX small FFT/FMA3 worship. Use Realbench 2.56 two hour stress test, Cinebench R20, 3600 second custom stress test, and Prime95 29.8 build 5 small FFT preset with AVX and AVX2 *disabled*. This is all you need to do to avoid damaging your processor and driving yourself to the mental hospital in the process.
Also, are you aware that Silicon Lottery bins their golden processors with an AVX -2 offset?

Even THEY aren't crazy enough to test with small FFT AVX at 5.1 ghz. If they did, maybe 1% rather than 4-8% of their 5.1 ghz bin processors would pass, and they would need cooling solutions most people simply don't use.

You're reading the wrong vcore sensor.

CPU-Z is the Super I/O chip and is going to be wildly inaccurate. CPU-Z reads the ITE 8688E chip which tends to show higher than than the bios voltage you set. The ITE 8792E with LLC=Turbo shows a +/12mv voltage within bios voltage, regardless of your current, so even this is inaccurate.

You need to read VR VOUT. And your load voltage is 1.250v.

Also please reset your HWinfo fields (choose reset layout or reset to default layout or restore fields in the settings). You have two iGPU fields active (duplicated 0.004v VR VOUT which is the inactive iGPU).
And please don't run that prime95 test you are running!

You're pushing ***222 amps*** into that thing! That's WAY past the maximum Intel allowed specification of 193 amps! And at 193 amps, VR VOUT must not exceed 1.225v (tl;dr: the higher the amps, the lower the maximum safe VR VOUT allowed before your CPU risks faster than "normal wear and tear" degradation.

tl;dr:
if you can pass realbench 2.56 WITHOUT CPU CACHE L0 errors (shown in HWinfo64) at THESE SETTINGS:

x51 / x47
1.310v bios set (Fixed)
Loadline calibration: Turbo
Then you have anywhere between a very good to a golden processor (this depends on how low you can go on the bios set voltage without L0 errors).
x51 / x47
1.285v bios set
loadline turbo
2 hour realbench 2.56 run 32gb memory 0 errors. About to give the cinebench test a go. Based on these results do you think there is more room for higher clocks? I took a SS a few minutes before the test ended and also after to show it passed. VR OUT idle sits at 1.275 but under load hits 1.23 max
Attached Thumbnails
Click image for larger version

Name:	realbench51run.png
Views:	21
Size:	6.44 MB
ID:	289458  

Click image for larger version

Name:	realbenchpass.png
Views:	18
Size:	6.39 MB
ID:	289460  


Last edited by poopsLIVE; 08-16-2019 at 12:15 AM.
poopsLIVE is offline  
post #15 of 34 (permalink) Old 08-16-2019, 12:57 AM
New to Overclock.net
 
robertr1's Avatar
 
Join Date: Sep 2013
Posts: 155
Rep: 6 (Unique: 5)
Quote: Originally Posted by poopsLIVE View Post
What would your suggestion on absolute max voltage considering the setup in question for everyday gaming/local recording while streaming?
Falk made a lot of good points so I won't repeat the same thing.

I'd go back to static vcore and set your llc to turbo. Pick a fixed vcore that you can cool. Let's say that 1.35v in bios with llc turbo. Great. Now see what the max NON avx frequency you can run at that and see what your temps are like. For that, you P95 non avx, real bench, cinebench etc.

It took me a while but I eventually got to the following. 1.325v in bios with LLC = High. 5.2ghz all core with HT disabled. HT in my primary use case hurts performance thus I have it disabled (VR stuff). If you want to run HT, the same applies above just have HT enabled the whole time.

Once you that stable, you can go back and mess around with offset. I'd also recommend connecting the second 8 pin for cpu power. Don't ask me why but it's made my system more stable at lower voltage. It shouldn't but it has.
robertr1 is offline  
post #16 of 34 (permalink) Old 08-16-2019, 03:36 AM
New to Overclock.net
 
zGunBLADEz's Avatar
 
Join Date: Apr 2012
Location: Chicago
Posts: 3,502
Rep: 99 (Unique: 72)
for avx try blender = classroom benchmark loop it using animation XD

i would put the ihs back btw

zGunBLADEz is offline  
post #17 of 34 (permalink) Old 08-16-2019, 07:14 PM
New to Overclock.net
 
Socks keep you warm's Avatar
 
Join Date: Feb 2011
Location: Melbourne, Australia
Posts: 3,104
Rep: 131 (Unique: 116)
Quote: Originally Posted by poopsLIVE View Post
I have not checked the block no. I dont know exactly what I would lap. Doing the die would just make it harder to reach the cpu block if that were the case. Lapping the OC frame I dont think would be a good idea and I think the same goes for lapping the block, wouldnt that just make it harder to make contact? I was honestly thinking things couldve been too tight. I mean I can crank down a bit more on the block but I am not going to go overboard as it is already feeling resistance from the springs
I am at a disbelief right now, i have just added another 420mm rad equalling 900mm of rad in my case and NOW at 1.35V 5Ghz i am breaking 78c, it used to konk out at 98c and throttle.
This is ridiculous the amount of cooling you need for this chip.
EDIT: Add to this comment and the VR VOUT issue, since you are not using IGPU set IGPU voltage to .5V or lowest setting.

I like to have fun.
https://hwbot.org/user/sockskeepyouwarm/
https://www.trueachievements.com/gamer/II+TH3PRO+II
16 Threads is Good
(10 items)
CPU
Intel Core I9 9900K 4.8Ghz
Motherboard
Gigabyte Aorus Elite
GPU
Aorus 2080Ti Xtreme Waterforce WB 11GB
RAM
Trident Z
Hard Drive
Samsung 250 SSD
Power Supply
Cooler Master Silent Pro Hybrid
Cooling
Watercooling Custom
Case
900D
Operating System
Windows 10 x64
Monitor
28" ACER 4K Monitor
▲ hide details ▲


Socks keep you warm is offline  
post #18 of 34 (permalink) Old 08-16-2019, 08:22 PM
New to Overclock.net
 
Join Date: Oct 2018
Posts: 28
Rep: 0
the 9900k has a thicker silicon die which also causes it to be hotter.der8auer went as far as lapping the die to reduce its thickness and got another 5c cooler by doing so compared to just delid and conductonaut
some of the graphs are for the 9600k aswell? but same gen and same die thickness.

also cinebench r20 is one of the hottest stress tests ive seen on my 9900k get hotter then any other test ive run but i prefer occt over prime95.
[email protected] avx 0 vcore 1.325v nh-d15 cooler





Last edited by metalspider; 08-16-2019 at 08:25 PM.
metalspider is offline  
post #19 of 34 (permalink) Old 08-16-2019, 08:29 PM - Thread Starter
New to Overclock.net
 
Join Date: May 2019
Posts: 24
Rep: 0
Quote: Originally Posted by Falkentyne View Post
Ok Ok Ok.
First of all I'm going to shout loudly here.

DO NOT USE SMALL FFT AVX TO TEST WHETHER YOU WON THE SILICON LOTTERY OR NOT! You are exceeding the Intel maximum electrical specifications for that processor by SO MUCH that not only can barely ANY CPU pass prime95 at 5.1 ghz with that VR VOUT (see below), you are also going to slowly degrade that processor (or maybe more than slowly degrade it)--see below.
To test whether you won the silicon lottery, lay off the AVX small FFT/FMA3 worship. Use Realbench 2.56 two hour stress test, Cinebench R20, 3600 second custom stress test, and Prime95 29.8 build 5 small FFT preset with AVX and AVX2 *disabled*. This is all you need to do to avoid damaging your processor and driving yourself to the mental hospital in the process.
Also, are you aware that Silicon Lottery bins their golden processors with an AVX -2 offset?

Even THEY aren't crazy enough to test with small FFT AVX at 5.1 ghz. If they did, maybe 1% rather than 4-8% of their 5.1 ghz bin processors would pass, and they would need cooling solutions most people simply don't use.

You're reading the wrong vcore sensor.

CPU-Z is the Super I/O chip and is going to be wildly inaccurate. CPU-Z reads the ITE 8688E chip which tends to show higher than than the bios voltage you set. The ITE 8792E with LLC=Turbo shows a +/12mv voltage within bios voltage, regardless of your current, so even this is inaccurate.

You need to read VR VOUT. And your load voltage is 1.250v.

Also please reset your HWinfo fields (choose reset layout or reset to default layout or restore fields in the settings). You have two iGPU fields active (duplicated 0.004v VR VOUT which is the inactive iGPU).
And please don't run that prime95 test you are running!

You're pushing ***222 amps*** into that thing! That's WAY past the maximum Intel allowed specification of 193 amps! And at 193 amps, VR VOUT must not exceed 1.225v (tl;dr: the higher the amps, the lower the maximum safe VR VOUT allowed before your CPU risks faster than "normal wear and tear" degradation.

tl;dr:
if you can pass realbench 2.56 WITHOUT CPU CACHE L0 errors (shown in HWinfo64) at THESE SETTINGS:

x51 / x47
1.310v bios set (Fixed)
Loadline calibration: Turbo
Then you have anywhere between a very good to a golden processor (this depends on how low you can go on the bios set voltage without L0 errors).
x52 / x49
1.41 bios set
Loadline Extreme
2 Hours Realbench W/ Zero Errors


I think that would pretty much be the thermal limit. The 51 settings I posted earlier passed all three tests. Was about to put this through Cinebench r20 3600 sec test and see how it goes. I think this is the limit of this chip unless you have some better way of lowering the voltage. Do you think I can get away with these as an everyday streaming while locally recording and gaming rig?
Attached Thumbnails
Click image for larger version

Name:	realbench52.png
Views:	8
Size:	6.44 MB
ID:	289660  

Click image for larger version

Name:	realbench52pass.png
Views:	6
Size:	6.37 MB
ID:	289662  

Attached Images
File Type: bmp 190816232334.BMP (2.25 MB, 3 views)
File Type: bmp 190816232356.BMP (2.25 MB, 3 views)
File Type: bmp 190816232403.BMP (2.25 MB, 3 views)
File Type: bmp 190816232414.BMP (2.25 MB, 3 views)
File Type: bmp 190816232429.BMP (2.25 MB, 3 views)
poopsLIVE is offline  
post #20 of 34 (permalink) Old 08-16-2019, 08:51 PM
New to Overclock.net
 
Falkentyne's Avatar
 
Join Date: Dec 2007
Location: Riverside
Posts: 5,987
Rep: 383 (Unique: 274)
Quote: Originally Posted by poopsLIVE View Post
x52 / x49
1.41 bios set
Loadline Extreme
2 Hours Realbench W/ Zero Errors


I think that would pretty much be the thermal limit. The 51 settings I posted earlier passed all three tests. Was about to put this through Cinebench r20 3600 sec test and see how it goes. I think this is the limit of this chip unless you have some better way of lowering the voltage. Do you think I can get away with these as an everyday streaming while locally recording and gaming rig?
If it were me, I would not stress test at those settings and I would also not exceed "Turbo" LLC at voltages that high.
I cannot assure the longevity of your CPU before it starts getting unstable eventually and then needs more vcore.

I calculated the maximum safe voltages (on-die sense voltage--VR VOUT) based on different current draw amounts; these are for people that want to be safe. Exceeding these limits could result in accelerated degradation.
All these values are extrapolated from the Intel specification documents (1.52v VCC maximum (CPU must be pulling ZERO AMPS (meaning--not subject to a clock) if VR VOUT is 1.520v!!), 193 amps maximum, with DEFAULT loadline (Intel does not allow for loadline calibration in any of its specifications). The default VRM Loadline will drop the 1520mv based on amps, down to a safe MAXIMUM level--meaning these are absolute limits (on ambient cooling).

Voltage given is at maximum load, measured on the CPU die.

193 Amps: 1.212v to 1.230v (8 core CPU + 1.6 mOhms VRM loadline, versus 6 core CPU + 2.1 mOhms VRM loadline). Value: 1520mv - ( 193 * 1.6 ) = 1520mv - 308.8 = 1211.2 mv
150 Amps: 1.280v (1520 - (150 * 1.6)).
100 Amps: 1.360v (1520 - (100 * 1.6))
50 Amps: 1.440v (1520 - (50 * 1.6))

When using loadline calibration, the mOhms of loadline is reduced (standard/Normal loadline calibration is 1.6 mOhms), so this totally changes the entire formula, but the voltage/amps target is identical, but the BIOS voltage must now be set lower.

Example: Loadline calibration=Turbo is 0.4 mOhms of loadline.
So if you had a 1.340v bios voltage set and were pulling 150 amps in Realbench, for example:
1340 mv - (0.4 * 150) = you would be pulling 1280mv load VR VOUT at 150 amps.
As you can see that matches up with the above table so you're okay.

But if your bios voltage were higher (like 1.360v) with LLC Turbo, then that's asking for trouble at 150 amps.

Extreme LLC is 0.2 mOhms of loadline, so you can see the load VR VOUT is shifted upwards.
So tl;dr: I would not feel safe running at that voltage and that loadline. I know you may be unstable with a lower LLC, but keep that in mind. You want your CPU to last awhile without needing more voltage.

I'm happy that you're stable but I would absolutely not keep the CPU at these settings. Using Extreme LLC

[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  
Reply

Tags
9900k , direct die , Overclock , temperature , water cooled

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