[SOLVED] 9900K OCed to 5Ghz - Smoking Hot

Regev

Reputable
Jul 3, 2020
248
13
4,585
Hey guys, I tried to OC my CPU for the first time. My PC:

9900K + Noctua D15 + Gigabyte Aorus Z390 I Mini ITX, inside a Silverstone LD03. No GPU. PSU is a Platinum+ 750W Silverstone.

I set the ratio to 50, ring ratio to 47, voltage to 1.265V, but the system fails (blue screens) when I run Intel BurnTest (30 Loops, Standard). Ramped up the voltage to 1.3V, still fails. Set it to 1.4V, still fails. HOWEVER, when I set the voltage on Auto, it does pass the test. The problem? All cores' Max Temp recorded is 95 to 100.
On stock speeds they reached a maximum of 80 on that test.

What am I doing wrong, is the chip simply bad lottery (I see folks hitting 5Ghz on 1.265 no probs), and is there anything I can do?

Thanks <3
 
Solution
However, lowest voltage I could get it to run stable on (for 15 mins of Prime + 1 hour of RealBench) is 1.230V. Runs for 4.7K all cores - CPU package tops at 75c (11c less!) .
This is close to where I have my 9900K. It sounds like you need better cooling, to be able to pump the voltage higher while keeping temps lower, to be stable at 4800MHz+.
Is there a reason you want to go higher? Are you experiencing a performance deficit somewhere that you think a couple/few hundred MHz could help with. I have run my CPU at 5+GHz - some extra points in synthetic benchmarks (and A LOT of extra heat) is all your gonna see.


HOWEVER,
Today I ran another test on the 4.8K settings - failed this time! And failed on the 15mins Prime...

Regev

Reputable
Jul 3, 2020
248
13
4,585
Big difference ! but whats the difference between the "Delid" and the "Direct Die" comparisons he did? I tried to watch it again but I dont get it - in both variations he cools the CPU die directly without the Intel metal layer on - what am I missing?

Also - isn't the Noctua as good as a 240 AIO cooler? I bought it because I was told it can OC the 9900K easily
 

USAFRet

Titan
Moderator
What am I doing wrong,
Not being methodical in the overclock procedure.
You can't just jam in high numbers to start with.

You may have a great chip, or it may be merely average.

Increase in steps until it fails.
 

jasonf2

Distinguished
Big difference ! but whats the difference between the "Delid" and the "Direct Die" comparisons he did? I tried to watch it again but I dont get it - in both variations he cools the CPU die directly without the Intel metal layer on - what am I missing?

Also - isn't the Noctua as good as a 240 AIO cooler? I bought it because I was told it can OC the 9900K easily
The Delid puts the stock heat spreader back on with better thermal interface material. The direct die mounts the cooler directly to the chip. Just a word of wisdom. The Utube video makes this look like no big deal. His disclaimers at the end are not only very real, but if you mess up, especially with the utility knife, you just threw away 400 bucks and potentially trashed your mother board too. Beyond the deliding tool there are also a couple of accessory brackets that are a must. From about the point that Intel froze at the 14nm process node they have continued to tweak every last bit of OC headroom from their chips. So while you can go to some crazy extremes to get the OC it isn't as simple as just upping the size of a cooler on this chip like it was in the past. Overclocking is clocking above design spec and is not only not guaranteed but can do damage. With all of the binning being done the chances of winning the "lottery" are not nearly as good as they were in the past.
 
  • Like
Reactions: Regev

Zerk2012

Titan
Ambassador
Hey guys, I tried to OC my CPU for the first time. My PC:

9900K + Noctua D15 + Gigabyte Aorus Z390 I Mini ITX, inside a Silverstone LD03. No GPU. PSU is a Platinum+ 750W Silverstone.

I set the ratio to 50, ring ratio to 47, voltage to 1.265V, but the system fails (blue screens) when I run Intel BurnTest (30 Loops, Standard). Ramped up the voltage to 1.3V, still fails. Set it to 1.4V, still fails. HOWEVER, when I set the voltage on Auto, it does pass the test. The problem? All cores' Max Temp recorded is 95 to 100.
On stock speeds they reached a maximum of 80 on that test.

What am I doing wrong, is the chip simply bad lottery (I see folks hitting 5Ghz on 1.265 no probs), and is there anything I can do?

Thanks <3
5.0 is not a magic number so why must it be 5.0? Your trying to pump up the voltage for a just average binned chip is what it sounds like.

From this only 30% of the chips tested could do 5.0

9900K4.80GHz4.60GHz1.275V100%
9900K4.90GHz4.70GHz1.287VTop 91%
9900K5.00GHz4.80GHz1.300VTop 30%
9900K5.10GHz4.90GHz1.312VTop 5%
 

Phaaze88

Titan
Ambassador
@lvt
You do know that the 9900K is soldered, right? If the OP isn't careful, they could end up doing this:
It's not impossible, but the difficulty level is higher, to the point of being a bad idea for more casual users.



Silverstone LD03: Where's the airflow in this model?
I see that it uses a chimney effect, and the D15 is most likely oriented vertically, but there's a single bottom intake and single top exhaust.

Aside from the unknown methods being used to run this cpu, the airflow looks questionable.
 

Regev

Reputable
Jul 3, 2020
248
13
4,585
It does use a chimeny effect. I set the Noctua fans in such a way that all the fans in the case (one bottom, one top, two Noctuas on the CPU cooler) are oriented to pull from bottom and push to top.
 

Regev

Reputable
Jul 3, 2020
248
13
4,585
5.0 is not a magic number so why must it be 5.0? Your trying to pump up the voltage for a just average binned chip is what it sounds like.

From this only 30% of the chips tested could do 5.0

9900K4.80GHz4.60GHz1.275V100%
9900K4.90GHz4.70GHz1.287VTop 91%
9900K5.00GHz4.80GHz1.300VTop 30%
9900K5.10GHz4.90GHz1.312VTop 5%

Is it 30% to reach 5Ghz without failing regardless of voltage? Because if I set the voltage to Auto, it does manage to reach 5Ghz, just that it reaches 90-100c at peak.
 

Regev

Reputable
Jul 3, 2020
248
13
4,585
How do you have it set up?
If it is set to quiet mode it will let the CPU reach throttle temps, or whatever it is set to, before starting to ramp up.
Have you tried putting it to max rpm all the time?

It ramps up its speeds to 100% once the temps are 60c+, I believe. I never use max RPM at all times because it's too noisy... or should I?
 

Regev

Reputable
Jul 3, 2020
248
13
4,585
Not being methodical in the overclock procedure.
You can't just jam in high numbers to start with.

You may have a great chip, or it may be merely average.

Increase in steps until it fails.

Would you use PRIME95, RealBench or Intel BurnTest?

I'm using Intel BurnTest because I figured when Intel themselves are telling you their chip is stable, it's your best bet. Also I was told the BurnTest results in stress higher than the other softwares (due to Linpack), therefore providing better evidence for stability?
 

Phaaze88

Titan
Ambassador
A mix of:
Prime95 - Small FFT, AVX/ AVX2/ AVX-512 all off <- tests cpu cooler strength
Cinebench R23 - the longer the better, or at least as long as you normally use the PC in one sitting? <- tests cpu Vcore stability
Realbench - though it's gone ~4 years with no updates, it's still useful. Stress test, longer is better, and half of your ram. <- tests cpu and gpu Vcore stability, and some memory.

Intel Burn Test: wasn't actually created by Intel, unless AgentGOD is an Intel engineer?
It uses Intel Linpack, but the loads this puts on users' cpus is a bit beyond the norm, even more so than Prime 95 with the above suggested settings.
Now that I take a look at your motherboard, you may have been overwhelming the VRMs under IBT, but 'passing' on auto voltage because the cpu is thermal throttling before the VRM has a chance to 'tap out'.
 
First of all, how well you can OC will be determined by the quality of your chip.
Those who have good chips will tell you, but those with dogs will be silent.

Overclocking requires you to increase voltage to support higher multipliers.
This higher voltage heats up the chip.
At some point(100c.) the chip will slow down or shut off to protect itself from damage.
That is why a good cooling system helps to raise the max multiplier.
I think the twin tower NH-D15 is an excellent pick for your case. It has the cooling potential of a 240 aio.
The key is to be able to feed it with sufficient cooling air to let it do it's job.
The Silverstone LD03 is a very good ITX case. But, the main airflow comes from a single 120mm intake fan on the bottom.
To get better cooling, at the expense of noise, install a higher cfm fan as intake.
Whatever fresh air comes in the bottom will exit out the top taking component heat with it.

On the details of getting maximum OC, I can't ell you.
But, I might suggest that for gaming, you would do better to let the turmo mechanism prevail.
That will increase the multiplier for a couple of cores when the need and conditions are right.
These processors are so strong that overclocking brings negligible performance to gaming.
A good OC might be worth it if your main use was multithreaded batch apps where all cores come into play.
 
  • Like
Reactions: Regev
@lvt
You do know that the 9900K is soldered, right? If the OP isn't careful, they could end up doing this:
It's not impossible, but the difficulty level is higher, to the point of being a bad idea for more casual users.

Glued, not soldered.

Intel has done this for more than a decade since the age of Core 2. Sometime the metal cover just falls off itself.

It's easier to remove the cover while the CPU is warm than when it's cold.
 

Regev

Reputable
Jul 3, 2020
248
13
4,585
First of all, how well you can OC will be determined by the quality of your chip.
Those who have good chips will tell you, but those with dogs will be silent.

Overclocking requires you to increase voltage to support higher multipliers.
This higher voltage heats up the chip.
At some point(100c.) the chip will slow down or shut off to protect itself from damage.
That is why a good cooling system helps to raise the max multiplier.
I think the twin tower NH-D15 is an excellent pick for your case. It has the cooling potential of a 240 aio.
The key is to be able to feed it with sufficient cooling air to let it do it's job.
The Silverstone LD03 is a very good ITX case. But, the main airflow comes from a single 120mm intake fan on the bottom.
To get better cooling, at the expense of noise, install a higher cfm fan as intake.
Whatever fresh air comes in the bottom will exit out the top taking component heat with it.

On the details of getting maximum OC, I can't ell you.
But, I might suggest that for gaming, you would do better to let the turmo mechanism prevail.
That will increase the multiplier for a couple of cores when the need and conditions are right.
These processors are so strong that overclocking brings negligible performance to gaming.
A good OC might be worth it if your main use was multithreaded batch apps where all cores come into play.

You mean just let the CPU be as is (stock settings)?

This is my work computer - I don't game at all. Heck, I use the built-in Intel UHD 630 GPU, there's no discrete card. I do heavy browser-based work, and some heavy text file work, all simultaneously. I also have Darktable and some photo-editing apps installed (I use Linux Debian Stable, I attached a secondary hard drive and installed Windows on it just so I can stress test with all those apps you mentioned). I doubt OC will benefit much, but since its a K CPU, and I got the cooling, and a good motherboard, I figured heck why not? For the fun of it. If it fails I'll just buy another one - I got 50% off Intel CPUs ^^

What exactly happens when I set the multiplier to 50 at 1.3V? Does it run on 5Ghz and 1.3V at all times, or does it simply change the turbo boost profile to reach 5Ghz when it's needed? I didn't find any setting to set it as fixed, or adaptive, etc.
 

Regev

Reputable
Jul 3, 2020
248
13
4,585
A mix of:
Prime95 - Small FFT, AVX/ AVX2/ AVX-512 all off <- tests cpu cooler strength
Cinebench R23 - the longer the better, or at least as long as you normally use the PC in one sitting? <- tests cpu Vcore stability
Realbench - though it's gone ~4 years with no updates, it's still useful. Stress test, longer is better, and half of your ram. <- tests cpu and gpu Vcore stability, and some memory.

Intel Burn Test: wasn't actually created by Intel, unless AgentGOD is an Intel engineer?
It uses Intel Linpack, but the loads this puts on users' cpus is a bit beyond the norm, even more so than Prime 95 with the above suggested settings.
Now that I take a look at your motherboard, you may have been overwhelming the VRMs under IBT, but 'passing' on auto voltage because the cpu is thermal throttling before the VRM has a chance to 'tap out'.

If Intel BurnTest uses the CPU beyond the norm, does simply using it alone enough instead of spending the time to run all the other "lesser" tests?
 

Phaaze88

Titan
Ambassador
If Intel BurnTest uses the CPU beyond the norm, does simply using it alone enough instead of spending the time to run all the other "lesser" tests?
No, because there is no 'all in one' test, if that makes sense.
It's possible to pass IBT and still crash in other applications;
Can pass Cinebench R23(ran overnight), pass Realbench(ran overnight)... but if I run the following together:
-Cinebench R23
-A movie
-Use the browser
-Have Steam on in the background
= make Watchdog BSOD pop up within minutes XD
 
Last edited:

Regev

Reputable
Jul 3, 2020
248
13
4,585
As long as you have speedstep and/or turbo boosts settings enabled it will run at lower clocks whenever possible.
If you have TDP pl1/pl2/tau setup it will also clock down your CPU after a while.

Also how long would it take you to set these settings boot into windows and just look up the clocks?!

Cause the clocks are weird. It keeps rotating around 5K (4998, 5001, etc), even when the computer is idle. I didn't set it on fixed or anything, is why i'm confused.