[SOLVED] i9-10900k constant temps jump from 70 to 80 ?

[Crimson Morrow]

So I ran into an issue with my 10900k (OC 4.8ghz) where sometimes it would run normally at 70c but then after some time i would run the test again, and it would go up to 80 degrees. This issue was not present before.
I tried reapplying thermal paste (4 times in two days already). I also tried tightening the screws on the heatsync. The last one would help but not for long. The temps would still climb up to 80 sometimes. Prolonged use doesn't have to do anything with it. At least I don't think so. The 10 degrees temp difference would still be present even when idle. Normally it would be 33-36, and 45-46 when the issue is present. I monitored the load, it is 1% loaded by the system thats it. I really don't wanna reapply the thermal paste again, does anybody know what could be causing it? I'm thinking some software issue perphaps? Could it be straight up defective? The processor I mean. Even though I cannot think of anything I could have change recently that could have caused that. One more thing. The first two times I was reapplying the thermal paste I also removed the processor from the socket, just to see if it was sitting right in there (I know it does if it works, but i still decided to check anyways). So um... yeah...



Using it with a Notcua NH-D15
Stress testing it with Aida64

P.s. as to how hard I tightened the screws on the heatsync, I gave it 3 turns after it got set into place.

 

[Dragos Manea]

CPU-Z doesnt stress so much, is very light, prime95 is the heaviest of the bunch. 360mm radiator, d5 ekwb pump and velocity waterblock, cannot get it under 90 at 5.3, 5.0 ghz with prime95 is around 86

 

[Crimson Morrow]

CPU-Z doesnt stress so much, is very light, prime95 is the heaviest of the bunch. 360mm radiator, d5 ekwb pump and velocity waterblock, cannot get it under 90 at 5.3, 5.0 ghz with prime95 is around 86

interesting.. thanks!

 

[Nighthawk117]

Those temps look pretty good to me. Perhaps of not much help, but I have an i9 10850K (a 10900K but 100Mhz slower by default). At 4.8Ghz all core running Cinebench R23 I got a max temp of 76C, that was however with a .095v undervolt on a 240mm AIO liquid cooler. I currently run at 4.5Ghz all core which gives me a typical load temp of 65C, the CPU though is generally very quiet as a result which I prefer.

I found my 10850K undervolted very easily, so you may want to consider doing that if you want to shave a bit off your temps and power consumption.

 

[Crimson Morrow]

Those temps look pretty good to me. Perhaps of not much help, but I have an i9 10850K (a 10900K but 100Mhz slower by default). At 4.8Ghz all core running Cinebench R23 I got a max temp of 76C, that was however with a .095v undervolt on a 240mm AIO liquid cooler. I currently run at 4.5Ghz all core which gives me a typical load temp of 65C, the CPU though is generally very quiet as a result which I prefer.

I found my 10850K undervolted very easily, so you may want to consider doing that if you want to shave a bit off your temps and power consumption.

Thanks!
Yeah... i mean... I also want the power ahah

 

[Nighthawk117]

Thanks!
Yeah... i mean... I also want the power ahah

Sure, you should be able to apply some sort of undervolt even at a high frequency, what you will get out of it though varies from chip to chip. It doesn't seem uncommon for 10th gen i9's to be happy with around a 0.1v undervolt, For clarity I'm talking about a negative voltage offset with voltages set to auto. If you fix the voltage you can obviously fine tune it more, but at the expense of higher power consumption at idle and light loads. If you were to explore it I would just start off low and reduce voltage in small 0.01v increments.

 

[Crimson Morrow]

Sure, you should be able to apply some sort of undervolt even at a high frequency, what you will get out of it though varies from chip to chip. It doesn't seem uncommon for 10th gen i9's to be happy with around a 0.1v undervolt, For clarity I'm talking about a negative voltage offset with voltages set to auto. If you fix the voltage you can obviously fine tune it more, but at the expense of higher power consumption at idle and light loads. If you were to explore it I would just start off low and reduce voltage in small 0.01v increments.

I got mine fixed at 1.220v at all times, seems to be working fine for me. I actually have just created another thread where I go into a deeper detailed about the actual issue that I'm facing rn, you can take a look and well, if you have any thought, you are more than welcome to chip in https://forums.tomshardware.com/threads/10900k-temps-switch-between-70-and-80.3788798/

 

[Karadjgne]

Aida64 and Prime95 (without AVX disabled) can spike power use upto a "virtual" 130% load, far exceeding what is generally considered to be "normal" usage.

AVX is Advanced Vector eXtensions, an instruction set used by the cpu for vector analysis of particles, like all the chunks of debris after something blows up etc, so can be brutal on a cpu instead of the more normal placement of objects in the frame.

So it's going to jack temps up crazy high with any "stress" test that uses such. If all you are after is temps, use Prime95 small fft AVX disabled, as that is a 100% constant load on the cpu, the top which most pc's would consider normal.

Stress tests are for stability, temp tests are for temp. Aida64, occt, ibt, Asus real bench etc use a variable amount, so tests will bounce different load amounts from @ 80-130%, meaning you only see the extreme spike amount, which isn't all that accurate. But good for shaking up any OC/undervolt to see more extreme punishment use.

Good pastes last @ 8 years±, so there's absolutely no benefit to be changing paste with anything close to approaching 2x a day. Pastes will often have a 'burn in' period, as heat cycles and pressure from the cooler seat the connection better and get 'comfortable' being in that state, which often leads to better thermal transmission, lowered temps. Skipping that burn in by constant repaste isn't doing you any good, doing nothing but waste paste.

 

[Nighthawk117]

I got mine fixed at 1.220v at all times, seems to be working fine for me. I actually have just created another thread where I go into a deeper detailed about the actual issue that I'm facing rn, you can take a look and well, if you have any thought, you are more than welcome to chip in https://forums.tomshardware.com/threads/10900k-temps-switch-between-70-and-80.3788798/

Oh ok that's a pretty good vcore, much better than the stock voltage.

 

[Darkbreeze]

Aida64 and Prime95 (without AVX disabled) can spike power use upto a "virtual" 130% load, far exceeding what is generally considered to be "normal" usage.

Prime95 WITH AVX enabled? Yes.

Prime95 WITHOUT AVX enabled? No. It would be a steady state 100% all core load.

Aida64? No. In fact, the following is a statement from Computronix, arguably one of the most prodigious enthusiasts, overclockers and researchers when it comes to Intel Core architecture that you'll find on this, or any other site. And is also the author, as you know, of the Intel temperature guide.

AIDA64's Stress CPU fails to load any overclocked / overvolted CPU to get anywhere TDP, and is therefore useless, except for giving naive users a sense of false security because their temps are so low.

Which, having done tens of dozens of tests myself on a variety of Intel and AMD architectures, I fully agree with, because I've literally seen it proven before my own eyes as truth time after time.


The full monty from Comp. The only thing here I've changed is it saying P95 v26.6 because that was back before there was an option to remove AVX options from testing and 26.6 was the last version that lacked AVX in the testing. Now of course, we can do that, so the version doesn't really matter although you WILL want to be using the latest Prime version when testing especially if you are using a newer CPU.

Steady-state is the key. How can anyone extrapolate accurate Core temperatures from workloads that fluctuate like a bad day on the Stock Market?

I'm aware of 5 utilities with steady-state workloads. In order of load level they are:

(1) Prime95 - Small FFT's (AVX options disabled)
(2) HeavyLoad - Stress CPU
(3) FurMark - CPU Burner
(4) Intel Processor Diagnostic Tool - CPU Load
(5) AIDA64 - Tools - System Stability Test - Stress CPU

AIDA64's Stress CPU fails to load any overclocked / overvolted CPU to get anywhere TDP, and is therefore useless, except for giving naive users a sense of false security because their temps are so low.

HeavyLoad is the closest alternative. Temps and watts are within 3% of Small FFT's.

 

[Crimson Morrow]

Prime95 WITH AVX enabled? Yes.

Prime95 WITHOUT AVX enabled? No. It would be a steady state 100% all core load.

Aida64? No. In fact, the following is a statement from Computronix, arguably one of the most prodigious enthusiasts, overclockers and researchers when it comes to Intel Core architecture that you'll find on this, or any other site. And is also the author, as you know, of the Intel temperature guide.



Which, having done tens of dozens of tests myself on a variety of Intel and AMD architectures, I fully agree with, because I've literally seen it proven before my own eyes as truth time after time.


The full monty from Comp. The only thing here I've changed is it saying P95 v26.6 because that was back before there was an option to remove AVX options from testing and 26.6 was the last version that lacked AVX in the testing. Now of course, we can do that, so the version doesn't really matter although you WILL want to be using the latest Prime version when testing especially if you are using a newer CPU.

this is actually very interesting cause what i'm seeing on my end is that aida64 goes the hardest and brings my temps high as nothing else. I tried prime95 withavx and i can easily stay under 70 with that while aida would be in the 70-76 area.

 

[Phaaze88]

Aida64, FPU ONLY, goes hard.
Cpu only, cache only, or combining all 3, is easy.

 

[Darkbreeze]

Aida64, FPU ONLY, goes hard.
Cpu only, cache only, or combining all 3, is easy.

This.
But Aida64 has introduced AVX options in to their FPU only testing a while back and if that is enabled, then it is not a steady state environment. At least, not from what I've read, been told and seen.

 

[Karadjgne]

@Darkbreeze
Prime95 (without AVX disabled)
= with AVX enabled, which it does by default, it requires you to manually 'disable' AVX

Aida64 and Prime95 (without AVX disabled) can spike power use upto a "virtual" 130% load,

Notice I said 'can spike power use upto'


(from CT)
Notice Aida64 FPU, 115%. Pretty much guarantee that's going to hit TDP at least once, even the Aida64 Stress stands a good chance of hitting hard at least once, which is all that's needed to register a Max Temp, which is what most see, not realizing it's nothing more than a spike and isn't a consistent load.

CT and I have had multiple conversations, trust me when I say I'm in Awe of his commitment to understanding Intel cpu's, and others, and he quite politely (he's Extremely polite and well spoken btw lol) has corrected any misconceptions or misunderstandings I can come up with. But

AIDA64's Stress CPU fails to load any overclocked / overvolted CPU to get anywhere TDP,

I believe could be incorrect. Slightly.

13900k has a TDP of 125w. It's got a Max Turbo that'll hit 298w (according to certain setups, done by pro reviewers). That's not 'overclocked or overvolted', but factory default settings. Even Aida64 Stress at '70' will hit beyond 125w, exceeding TDP. That could also be true of some of the 7000 series Ryzens as well as 5900x/5950x possibly, I have no data to say for certain.

Whether enabling PBO, which doesn't really do anything but raise stock power limits, is considered 'overclocked or overvolted' isn't my call, but doesn't remove the listed TDP of the Ryzens, a 65w Ryzen doesn't become a 105w Ryzen by simply enabling PBO, it becomes more like an Intel with unlimited PL2 with PL1/Tau thrown out the window.

 

[Darkbreeze]

"TDP" is not "TDP" though. The listed TDP for ALL processors is well below what the ACTUAL thermal design power is, and I know you know this. This has been true for Intel since the inception of the Core architecture, or at least since turbo behaviors became "a thing", and has been true for AMD for at least twelve years now.

The actual thermal design power of a CPU is what it consumes when at an all core, full boost, 100% steady state load. And we can postulate exactly what that is by running Prime95 Small FFT with AVX disabled. The TDP we see under those conditions is going to be what the thermal design power actually is. Then, by comparison, if you run the Aida64 CPU stress test, we will see that it does not, in fact, come anywhere near the power consumption/design power spec as when we ran Small FFT.

The Aida64 FPU only test, totally different beast. That is NOT what most people run when they run Aida64 though. 99% of people will just run the Aida64 CPU stress test and think they are good because they don't come anywhere near TDP and therefore no where near thermal limit. And in fact, as Comp indicated, and as I've seen myself, even on an overclocked CPU with a raised voltage, if you are only running the Aida64 CPU stress test it usually still doesn't amount to what we saw running the same CPU at the stock configuration with full load on Prime Small FFT.

 

[CompuTronix]

Guys,

I'd like to jump in here and offer some clarifications.

With respect to CPU test methodologies, while our members and readers are struggling to sort out the information they see on various websites, videos and forums (much of which is conflicting), it's important to emphasize how very difficult it is to maintain a balanced perspective so as to filter out only the most relevant information. Since the number of real-world variables in hardware configurations, environmental conditions and software workloads are nothing less than mind boggling, then it's vital to be highly specific concerning test methodologies, so that variables can be reduced to their lowest common denominators. Only with a controlled approach, can we level the playing field so that valid and repeatable results can be obtained.

Unfortunately, most of our members and readers, as well as the vast majority of gamers and enthusiasts are unaware of this perspective. Testing is instead approached in a randomly haphazard fashion, with little or no understanding of the variables involved such as thermal testing versus stability testing, and steady-state workloads versus fluctuating workloads; cause and effect. Aside from needing a BIOS encyclopedia, the most glaring issues I see are the lack of knowledge that all test utility workloads are NOT created equal, and that Workload = Power = Heat. While the typical gamer is focussed on % CPU Utilization, they don't realize that's NOT actual workload. Power consumption is instead a more valuable metric when Core temperatures become critical, which is best visualized on any monitoring utility that offers a graph.

As CPU Utilization merely indicates processor resource activity, which can not exceed 100%, Power consumption can easily exceed 100% TDP. Those who are running high performance overclocked gaming rigs with 6 or more Cores since 8th Generation are acutely aware of the cooling difficulties involved. To be very specific, "K" processors on "Z" motherboards, by default, do NOT follow Intel specifications in BIOS, which is why, by default, they will exceed PL1 when running Prime95 Small FFTs (No AVX). Aside from finer variables such as VID, Auto Vcore and leakage current, by paying close attention to the details, the lowest common denominators can be revealed, which in turn, levels the playing field so that accurate testing can be performed for apples-to-apples comparisons.

PL1, PL2 and TAU were introduced in 2nd Generation Sandy Bridge, then PL3 was introduced in 5th Generation followed by PL4 in 6th Generation. Although these TDP terms have been with us for many years, they've come under much scrutiny since 8th Generation when CPU Power consumption increased sharply due to the "Core Wars" with AMD, along with increasingly sophisticated levels of Turbo features. Henceforth, as the motherboard manufacturers each attempt to market the fastest products by manipulating Intel's TDP specifications, the relevance of PL1 has been reduced to a marketing ploy. Nevertheless, nothing has changed. The "Watt Wars" has not altered utility workloads for testing CPUs. Results from our favorite utilities remain just as valid now as before.

Prime95 (Small FFTs, No AVX), OCCT (Small, Normal, Steady, SSE) and CineBench R23 (Multi Core) are all 100% workloads within just a very few % of Power consumption from one another, provided that the above settings are observed. Given the extremely high PL2 Power consumption of the most recent Generations of Intel i9 and i7 processors, this makes them practically impossible to cool adequately below Tj Max at 100°C when testing. However, CPU-Z (Bench-Stress CPU) is more consistent with heavy gaming workloads at about 80% TDP. Accordingly, many users have gravitated toward using CPU-Z, which is the recommended go-to alternative for thermal testing recent Generations of ultra-Power-hungry CPUs, while the individual CPU test in AIDA64 is an inadequate workload at about 70%.

CT

 

[Darkbreeze]

the individual CPU test in AIDA64 is an inadequate workload at about 70%.

This. Thanks Comp.