9900K OC 5GHz but stress test only 4.6GHz?

shadow2761

Limp Gawd
Joined
Mar 11, 2016
Messages
455
Just built new rig:
Corsair 500D SE RGB
EVGA SuperNova P2 1000w
ASUS Strix ROG Z390-E
i9-9900K
Corsair 115i Platinum
16GB Gskill DDR4 @ 3600 C14
MSI RTX 2080 Ti SeaHawk X @ 2050MHz
Samsung EVO 970 PLus
Samsung 850 PRO

Trying to OC this 9900k to 5GHz, in CoreTemp and HW64 it shows 5Ghz some times but when I do any stress testing it goes down to 4.6GHz and stays there. How do I get it to stay at 5GHz when it is under load and not just randomly under no load, as it seems to be?

I have basically done this so far:
1) Enable XMPII for RAM
2) Set CPU Core Ratio to "Sync All Cores" and set the multiplier to 50
3) Disable ASUS Multi Core Enhancement
4) Disable CPU SVID
5) in "External Digi+ Power Control" set Loadline Calibration (LLC) to level 6, and CPU Current Capability to 140% (the max I could see?)
6) Set CPU Core/Cache Current limit to max
7) Change CPU Core voltage to Manual, 1.29v
8) Go to "Internal CPU Power Management", set both Short Duration and Long Duration Power Limits to max (4095)

So far max temps I have seen on CPU is 60-70°C

Am I missing something?
 
What bench marks are you using? You probably have avx offset auto. Which will lower your max OC boost under a avx load. AVX loads hit your CPU really hard compared to a normal load so they have the offset to try and keep power and temps under control. You can disable it but unless you have very good cooling you will run really hot.
 
What bench marks are you using? You probably have avx offset auto. Which will lower your max OC boost under a avx load. AVX loads hit your CPU really hard compared to a normal load so they have the offset to try and keep power and temps under control. You can disable it but unless you have very good cooling you will run really hot.

Like he said, it's the avx offset. Many games and applications use avx these days so it's not something you can ignore like you could a few years ago. For the 9900k@5ghz, you need a custom wc loop to sustain true 5 ghz with no avx offset. Otherwise it's best to let it drop down if you are on air or using AIO.
 
I agree and disagree with some of the above, though a large part of it depends on how good of a silicone lottery winner you are.
My personal experience with a H110i GTX 280mm AIO, I was able to get 5.0ghz with no AVX but it got to 85-90c when doing stress testing which is a bit uncomfortable ~ normal heavy usage capped at about 80c.
I currently have it set for 5.0ghz with 2 AVX with 1.26v, and it maxes around 80-85c stress testing, but normal usage caps at 70c when doing video rendering (gaming is under 65c) which is acceptable for daily usage.

I would expect the 115i to have slightly more cooling capacity, but I also feel I got a really good chip.
You could also consider delidding it and putting a copper top, Cecil has done alot of them and generally sees pretty significant temp drops (often double digits).
 
I agree and disagree with some of the above, though a large part of it depends on how good of a silicone lottery winner you are.
My personal experience with a H110i GTX 280mm AIO, I was able to get 5.0ghz with no AVX but it got to 85-90c when doing stress testing which is a bit uncomfortable ~ normal heavy usage capped at about 80c.
I currently have it set for 5.0ghz with 2 AVX with 1.26v, and it maxes around 80-85c stress testing, but normal usage caps at 70c when doing video rendering (gaming is under 65c) which is acceptable for daily usage.

I would expect the 115i to have slightly more cooling capacity, but I also feel I got a really good chip.
You could also consider delidding it and putting a copper top, Cecil has done alot of them and generally sees pretty significant temp drops (often double digits).

No way a 5 ghz avx on aio will hold up on p95 with AVX on. It would reach 100C very fast even at 1.26v. Now for light avx usage and daily gaming it should be okay but it's still never 100% solid on AIO. If you delid it might work but not without.
 
Hmm, before I bought the 9900k I thought 4.9-5GHz was not hard to obtain on AIO as alot of users have reported.

Anyhows, my temps are only getting 60--70°C max., is that enough to cause the throttling to 4.6 GHz?
In my BIOS settings, it states my cooler score is 184 pts.

How can I maintain it closer to 5GHz. I'd be happy with 4.9GHz for daily gaming.

I'll check AVX later tonight, can't remember what setting mine was on. what ever it is default I guess becsuse I don't think I changed it.
 
Last edited:
Hmm, before I bought the 9900k I thought 4.9-5GHz was not hard to obtain on AIO as alot of users have reported.

Anyhows, my temps are only getting 60--70°C max., is that enough to cause the throttling to 4.6 GHz?
In my BIOS settings, it states my cooler score is 184 pts.

How can I maintain it closer to 5GHz. I'd be happy with 4.9GHz for daily gaming.

I'll check AVX later tonight, can't remember what setting mine was on. what ever it is default I guess becsuse I don't think I changed it.
check that there isn't an offset set in bios. 60-70 is fine and wouldn't cause throttling.
 
Ok guys thnx for the input I now understand what AVX offset does, and yes you guys were correct I had an offset of 4. So I disabled it just to check and yep it maintained 5GHz on P95 small FFts (max power,heat) test for a few mins but I stopped it because it shot upto 95°C on some cores.

Is it possible I damaged the CPU at all by running that much heat for a few mins on 1.29v.

Does p95 actually represent the heat I would experience in real world gaming @ 4K? If not then I might roll with this setting for now?

I don't like the high heat reading. Strange because an AVX offset of -4 will only give me a max temp of 70°C, but 0 offset will shoot me above 90°C instantly.
 
I tried -1 and -2 and they also both get me close to 90°C @ 1.29v

Going to try 1.28v for now.
 
Last edited:
Like he said, it's the avx offset. Many games and applications use avx these days so it's not something you can ignore like you could a few years ago. For the 9900k@5ghz, you need a custom wc loop to sustain true 5 ghz with no avx offset. Otherwise it's best to let it drop down if you are on air or using AIO.

More then likely didn't cause any damaged. Prime really hits the CPU hard and unrealistic load. Try using cinebench r20. It will give you a more realistic work load. I get around 75c on cinebench r20 and during game play with gears 5 with a 7820x at 4.8Ghz with no offset. This is a custom water loop also.
 
Ok guys thnx for the input I now understand what AVX offset does, and yes you guys were correct I had an offset of 4. So I disabled it just to check and yep it maintained 5GHz on P95 small FFts (max power,heat) test for a few mins but I stopped it because it shot upto 95°C on some cores.

Is it possible I damaged the CPU at all by running that much heat for a few mins on 1.29v.

Does p95 actually represent the heat I would experience in real world gaming @ 4K? If not then I might roll with this setting for now?

I don't like the high heat reading. Strange because an AVX offset of -4 will only give me a max temp of 70°C, but 0 offset will shoot me above 90°C instantly.
no you didn't damage it. your chip can handle the voltage and will throttle if too hot.
 
Remember if you run with AVX offset to test with prim95 both with and without AVX.
 
More then likely didn't cause any damaged. Prime really hits the CPU hard and unrealistic load. Try using cinebench r20. It will give you a more realistic work load. I get around 75c on cinebench r20 and during game play with gears 5 with a 7820x at 4.8Ghz with no offset. This is a custom water loop also.

you stress test to ensure stability now and in the future now just to said it passed something
Cinebench will not verify calculation aka slow data rot is a possibility.
Besides it a very short test. not recommend for stress testing.

Prime95 is a real world application with realworld optimizations


The " if i don't pass just run a weaker stress test" mentally defies the idea behind stress testing.
If you go this router you might as well just not do it as you are going to do the stability gamble anyway.
 
Last edited:
Was gaming 4K PUBG for a few hrs, with 5GHz -3 offset , and 1.28v and got a max of 71°C. No crash yet.

I might try lowering to 1.27v soon lol.
 
Was gaming 4K PUBG for a few hrs, with 5GHz -3 offset , and 1.28v and got a max of 71°C. No crash yet.

I might try lowering to 1.27v soon lol.

If you're pretty exclusively gaming then P95 small FFT is way too extreme. Personally I don't use an AVX offset and use x264 loop and OCCT to test it. They are very good at finding a sweetspot.
 
you stress test to ensure stability now and in the future now just to said it passed something
Cinebench will not verify calculation aka slow data rot is a possibility.
Besides it a very short test. not recommend for stress testing.

Prime95 is a real world application with realworld optimizations


The " if i don't pass just run a weaker stress test" mentally defies the idea behind stress testing.
If you go this router you might as well just not do it as you are going to do the stability gamble anyway.
Oh I am not saying to not use prime95. It is good for stress testing just that for most people it will not be a realistic load. I didn't mean a cinebench 20 pass was good stability test by no means. I can run prime95 all day and it hits mid/upper 80s but under my day to day stuff I have never seen it break 75. Which is around where cinebench 20 does.
 
Oh I am not saying to not use prime95. It is good for stress testing just that for most people it will not be a realistic load. I didn't mean a cinebench 20 pass was good stability test by no means. I can run prime95 all day and it hits mid/upper 80s but under my day to day stuff I have never seen it break 75. Which is around where cinebench 20 does.

I believe I misunderstodd you. Thank you for the correction.
I agree Prime95 is for worstcase testing
I agree Cinebench could be better for "daily Thermal testing" and fan adjustment aka where you want to fit something for your more typical load
 
I believe I misunderstodd you. Thank you for the correction.
I agree Prime95 is for worstcase testing
I agree Cinebench could be better for "daily Thermal testing" and fan adjustment aka where you want to fit something for your more typical load

I interpreted it wrong too until I read his response.
 
So I ran Cinebench R20 twice and played another 4 hrs of PUBG tonight @ 5Ghz (-3 AVX) 1.28v.

Again, max temps 65°C.
Voltage drops to 1.27v under load according to HW64. I have LLC6 selected.
 
Put about 15hrs into PUBG @ 4K 120Hz with 5.0GHz (-1AVX) @ 1.27v so far.
Max temp 68-72°C

Can I consider that stable lol?
 
Put about 15hrs into PUBG @ 4K 120Hz with 5.0GHz (-1AVX) @ 1.27v so far.
Max temp 68-72°C

Can I consider that stable lol?

Nope. is you CPU load at 100% nope. so you haven proven it can do the same amount of stress as non overclocked.
but if you want to run with the risc of instability that is always your choice.


This was actually how i got into propper stress testing
my dual celeron 366 at 564 was running quake3 stable for hours. but every rar archive i made i could not decompress on other computer due to data corruption from the overclocked CPU when compression the data.

In you situation you load on the cpu is low and you have NO error detections going on
There is a different between stable and "not crashing" where he entire spectrum of data trust and corruption lies.
 
A real overclocker ALWAYS stress tests. More than likely, if you do nothing else with your PC, a random crash was likely caused by an unstable overclock. Personally I ensure my systems are RealBench stable as it tends to simulate a realistic full load for most uses. It still maxes RAM and CPU to 100% as well. Never had an issue if I make an OC stable using RealBench to test. But of course, everyone has their own personal preferences on what software tests best. I've stayed clear of Prime95 because of the insane amount of electrical current it can pull through a processor. A buddy of mine killed an i9 using Prime95 and his temps were ok. Granted, I did not witness this myself, but i helped him OC his replacement using RealBench and he has never had an issue with his OC to date.
 
A real overclocker ALWAYS stress tests. More than likely, if you do nothing else with your PC, a random crash was likely caused by an unstable overclock. Personally I ensure my systems are RealBench stable as it tends to simulate a realistic full load for most uses. It still maxes RAM and CPU to 100% as well. Never had an issue if I make an OC stable using RealBench to test. But of course, everyone has their own personal preferences on what software tests best. I've stayed clear of Prime95 because of the insane amount of electrical current it can pull through a processor. A buddy of mine killed an i9 using Prime95 and his temps were ok. Granted, I did not witness this myself, but i helped him OC his replacement using RealBench and he has never had an issue with his OC to date.

Real bench is pretty weak. it does have the beenfit of doing both CPU and GPU. but 25% of it comes form 7-zip that is pretty bad for cpu stability measuring
but again ppl choices.
but it not stable if you system runs differently than non overclocked


P.S.
The prime killing a CPU seems like BS prime95 does dot anything you cpu is not designed to do.
if you CPU died it was due to poor build quality or a defective CPU getting revealed as being defective
Prime cant PUSH power through you CPU. it has no control of that.
Prime95 just deliver optimal code from a realword application and the CPU subsystem decided how to handle that.
 
Last edited:
Code:
Core I7 2700K           4.4GHz +0.030v                  4.4GHz +0.20v                   4.5GHz +0.20v                   4.6GHz +0.20v                   4.7 +0.20v
                        Status  temp    Power           Status  temp    Power           Status  temp    Power           Status  temp    Power
Asus RealBench          OK      79-84   318w*           OK      80-85   318w*           OK      80-87   303-319/326*    C  0:22 76-83   xxx/325*        C 0:00  XX-XX   xxx/315*
Aida64 Extreme          OK      74-81   181w            OK      73-79   176-180w        OK 7:40 74-81   180-185/187     OK 4:50 76-82   182-186/190     C 0:15  73-82   xxx/193

Core Damage 0.8h                                        OK      76-82   180w            OK15:28 79-85   184/188         OK 9:50 79-85   186/190         C 1:16  79-85   xxx/192
Cpu Burn-in 1.01                                                                        OK 4:05 73-77   181/184         OK 8:15 81-77   185/187         C 0:06  73-78   xxx/190

Folding@Home            OK      73-78   182w            OK      73-79   179w            OK 9:30 73-79   180/182         C  2:49 74-81   xxx/186         C 0:00  31-31   xxx/185

HeavyLoad               OK      72-76   184w            OK      72-76   182w            OK 4:20 73-78   187/189         C  2:59 74-80   xxx/193         C 0:00  xx-xx   xxx/193
Handbrake 1.0.2         OK      71-76   165-177w        OK      72-76   96-180          OK 8:08 73-77   95-180/185      OK 6:30 73-80   160-182/186     C 0:29  72-76   xxx/186

IBT 2.54 Standard                                       OK      78-88   167-211w        OK 4:00 80-90   171-215/216     C  0:02 71-79   xxx/218         C 0:00  xx-xx   xxx/215
IBT 2.54 Maximum                                        OK      80-91   106-212w        C  1:19 83-94   xxx/220         C  0:02 72-82   xxx/217         C 0:01  63-68   xxx/206

LinPack V11.2.2.010     OK      78-90   191-210w        OK      78-89   162-207w        OK 4:15 80-91   161-211/213     c  0:05 78-88   xxx/218         C 0:00  xx-xx   xxx/210
LinPack V11.2.2.010 x2  OK      84-96   220w            OK      81-92   169-216w        C  0:55 82-94   xxx/223         c  0:01 73-76   xxx/223         C 0:00  xx-xx   xxx/182

OCCT 4.4.3 (Own/large)  OK      78-88   211w            OK      76-84   185-204w        OK 4:00 77-87   198-205/211     C  0:06 73-81   XXX/212         C 0:00  31-31   xxx/193
OCCT 4.4.3 (Linpack90%) OK      82-93   207-214                                         OK 4:05 74-83   191-196/199     C  0:48 76-85   xxx/204         C 0:02  37-53   xxx/199

Prime95 26.6 (small)                                                                    OK 4:05 77-85   195/198         C  0:12 76-85   xxx/201         C 0:00  xx-xx   xxx/195

PRime95 28.5 (8-8-30)   OK      84-93   217w                                            C 0:40  82-93   xxx/218         C  0:00 68-76   xxx/215         C 0:00  xx-xx   xxx/213
Prime95 28.5 (small)    OK      79-89   212w            OK      80-90   211w            C 1:52          xxx/219         C  0:00 63-72   xxx/214         C 0:00  xx-xx   xxx/215
Prime95 28.5 (Blend     OK      82-92   215w            OK      84-94   201w            C 2:26  83-93   xxx/226         C  0:03 67-71   XXX/196         C 0:00  36-36   xxx/194

Passmark BurnInTest 8.1 OK      67-72   104-180         OK      67-72   103-178w        OK      69-73   105-179w/183    OK 6:35 69-74   108-175/182     OK 4:25 70-74   109-183/187
XTU (CPU)               OK      73-79   176-190w                                        OK 7:25 73-79   173-186/188     C  2:17 73-81   xxx/210         C  0:04 67-73   xxx/191

Winrar                  OK      60-67   151w            OK      60-67   150w            OK      61-67   15-/155/177     OK11:00 62-69   155/156         C  1:20 61-67   xxx/165
7-zip                   OK      68-71   124-173         OK      68-72   126-171w        OK 9:20 70-74   126-175/178     OK 9:40 71-74   128-177/179     OK 6:40 71-76   134-179/181


* This uses OpenCL to put load on GPU as well.
  Not direct comparable against other candidates

Status key
OK = Did not find any errors
C  = Crash/BSOD
R  = Reboot/Reset
D  = Error detected
 
as we can see above realbench will run fine for more thna 7 hours in a system that will cras around the 1 hours mark in hard stress testing (prime95)
That is despited realbench pushing on the GPU as well while prime95 does not

Aida 64 is also oftne used but horribel as it will run for over 4 hous with no crashes in a systme that craes in less than a sec under hard load
This make realbench and aida 64 close to useless imho

This is just one of the multiple stress test tool test ive done and only prime95 and linpack based test seems to be "good"



Testing was based on a 4 hours testing interval. if no issues was detected it was considered a negativ results (CPU is OK)

Due to no being able to sit and monitor my computer for all times some test are running longer than 4 hours but however clearing the 4 hours marks with no detecting wold still be considered no detection even if a detecting would arrive later (like at a 6 hours mark)
However none of these cases where present during this testing.


I've done the same kind of testing for memory tester as well and prime95blend and memtest86 from passmark (booted in uefi mode) are the best options here.
 
Last edited:
guess you better make a new stress tester then, so you can pimp it every chance you get...
 
Nope. is you CPU load at 100% nope. so you haven proven it can do the same amount of stress as non overclocked.
but if you want to run with the risc of instability that is always your choice.

In you situation you load on the cpu is low and you have NO error detections going on
There is a different between stable and "not crashing" where he entire spectrum of data trust and corruption lies.

What do you mean NO error detections going on?

I had 2 crashes from PUBG last night, they seem to be app related crashes because it asked me to send an error report and I was able to immediatly launch the game again (no hard lockups or blue screens etc). Could that crash be due to an unstable OC?

A real overclocker ALWAYS stress tests. More than likely, if you do nothing else with your PC, a random crash was likely caused by an unstable overclock. Personally I ensure my systems are RealBench

I ran Cine R16 a few times, can that count for a good stress test.

I don't like using P95 because I hate seeing red numbers on my CoreTemp app lol
 
What do you mean NO error detections going on?

I had 2 crashes from PUBG last night, they seem to be app related crashes because it asked me to send an error report and I was able to immediatly launch the game again (no hard lockups or blue screens etc). Could that crash be due to an unstable OC?



I ran Cine R16 a few times, can that count for a good stress test.

I don't like using P95 because I hate seeing red numbers on my CoreTemp app lol

PUBG does not check if the calculations that are done are correct. You only see a result when the error is big enough to cause a crash aka you are blinding you self for small rounding errors.
Those small rounding errors can slowly over time corrupt data you are working on

just an examaple let say pubg is decoindg a texture does it matter if a color becomes 234.245.236 or 235.245.236 . naah no crashing and you would no notice this slifht difference in color on the texture
But does it matter that a byte changes from 234 to 235 when you ntfs records get updates... highly


I ran Cine R16 a few times, can that count for a good stress test.
opionse does not really change how things works.
This is a horrible test. again due to lack of error detection


I don't like using P95 because I hate seeing red numbers on my CoreTemp app lol
Self denial is really not a way to handle technical stuff
 
guess you better make a new stress tester then, so you can pimp it every chance you get...

uhh should i be sorry for actually providing technical information rather than sassy comment to appear cooler on a forum ?
 
.
Those small rounding errors can slowly over time corrupt data you are working on

This! Thank you, I have been trying to find out exactly what an unstable OC can do besides the obvious things like freeze/hang, BSOD etc. Do you have any more info on this, I tried to google it but cannot find info like what you just mentioned.

And if C16 is a terrible stresser, is there anything else I should use that won't make my temps skyrocket like P95 or do I have no choice but to bite the bullet and do a torture test for hrs those conditions?
 
This! Thank you, I have been trying to find out exactly what an unstable OC can do besides the obvious things like freeze/hang, BSOD etc. Do you have any more info on this, I tried to google it but cannot find info like what you just mentioned.

And if C16 is a terrible stresser, is there anything else I should use that won't make my temps skyrocket like P95 or do I have no choice but to bite the bullet and do a torture test for hrs those conditions?

This is the actual reason i got into "proper" stress testing.
my systen would run Quake3 fine for hours (the game of the time)
but i was unable to make winrar archives. dues to calculations errors
Reasson as i clocked my cpu down a bit again then winrar worked as planned

That what told me CPU "instabilites" os more than just "not crashing"
 
This! Thank you, I have been trying to find out exactly what an unstable OC can do besides the obvious things like freeze/hang, BSOD etc. Do you have any more info on this, I tried to google it but cannot find info like what you just mentioned.

And if C16 is a terrible stresser, is there anything else I should use that won't make my temps skyrocket like P95 or do I have no choice but to bite the bullet and do a torture test for hrs those conditions?


The issues is you need the temps to go up to verify you system is correct

let say you stress it with "What you use" and you complete you 5 games for 16 hours no issues becuse the temp is not high

new patch comes out and one of you games that was core speed bottlenekc now get a vulcan support and cna use more cores
temps goes up and you are now unstable during the game because you did not test you system for worstcase scenario

as son as you doe not test for worstcase scenario you "leave an area" of unknown behind you that might contains instability but becauseyou havent wenture in there yuo dont know it it it can surpsie you down the road


That the technical part of it to answer you question (Even though i think its a bad idea)
You can use prime and disable AVX
if this is not enough refer to my list above and just find something that is weaker.
it contains both temp and power load measurements

looking at the 4.5 +20v result we see only prime95 and linpack based test discovered the error
moving to next colum you can find the next best things

under 4.6 +.20v
we can see that asus relbench steps into the ring
folding at home and heavy load as well ( again might not have error dectetion just crashing)
OCCT also show that i i part of the second best choices ( this is the weakest Linpack test ive seen)
old prime 26.6 ( most likley due to no avx)
Intel XTU
Soo the above is what i would consider the second best options
 
Back
Top