10900k real world temperatures, what is too high/expected? msi z490 Settings

[markm75]

I have the 10900k overclocked on all core to 51x bios 1.305v and vccio and sa at 1.25 with dram at 1.48(4000mhz). LLC on 4, TVB is on, eist disabled, speed shift disabled

I have the thermaltake 360 cooler which on my 8700k at 5ghz did well, i'd say 78c max on most things. My fans are on the top of the case for the radiator and on the front i have 2 or 3 120's i think blowing air out though i'd need to confirm that, along with a push out on the 1080ti hybrid at the rear (thermaltake v51 case).

Now with the 10900k, prime95 blend is hitting 90c (no avx), medium occt hits 84c, real world windows 70-80c max, gaming 83c, and rendering in my one app that i think uses avx it hit 87c, this is with ambient fairly normal or slightly below room temp even. Idle im 30s to 40.

Any thoughts on what is "safe" for 24/7 here with the new chip? I was thinking maybe if i up the vccio/sa (though i thought about 1.25 wasnt good 24/7) then maybe i could lower the vcore from 1.305. If i lower vcore anymore, i fail stress testing pretty quickly.

I see others with similar radiators/aio's and this chip but they seem to have lower temps and even lower vcores in some cases. I suppose maybe i did a bad job putting the cooler back on, i used pea sized arctic silver 3 and applied the cooler.


My settings are below, 12 hrs stable p95 blend (90c max)

Thanks in advance


Ram: F4-4000C17D-32GTRSB (16x2)


Bios 1.305 and auto mode (not override)
All Core, 51x
Eist disabled greyed out/Turbo greyed out
Ring Set to 47x
Thermal velocity Boost is set to defaults auto
AVX offset -1
SA 1.25 (1.25 max safe 24/7?)
IO 1.25
Ram 1.48
Ram xmp and 4000
LLC4
LN2 Extreme
Cstate Enabled
Speed shift Disabled
PCIe Spectrum Enabled

 

[EarthDog]

90C for stress testing temps is fine. It will start to throttle at 100C.

 

[markm75]

90C for stress testing temps is fine. It will start to throttle at 100C.

Well for avx type rendering i'm at 87C already though, i would think for normal use or semi regular use thats too high?

 

[EarthDog]

Well for avx type rendering i'm at 87C already though, i would think for normal use or semi regular use thats too high?

Nope. 87C is still less than 90C.

 

[wingman99]

Under 100c is fine for longevity.

 

[markm75]

for completion sake since another thread related to cooling/die i posted my results, my mostly final results:

So i'm still unsure how i managed stability at 1.29 previously, but now that i relied on auto volts i've found real stability:

Auto volts in bios on all, LLC auto, xmp and 4000 for memory and auto volts there too. 51x and auto ring (if i set ring to 47 windows wouldnt boot even on auto volts)This results in an idle of around 1.301 initially in windows. Under extreme load this is 1.315 (sa 1.321 and cio 1.242).

Realbench initially hits 81C by 3min 89C, by the end 94C (too hot), this is with the case side off too. Now if i blast the ac window unit i can get it down to around 90C.I was "realbench stable" all night, about 6 hours worth. I consider this ok now.I rebooted into normal windows (I use an isolated test windows for stress tests) and tested a few activities, ambient is slightly cool right now though.Xplane 11 sim i only hit about 78C (probably 82 when room is warmer). Fs2020 only 65C. A quick 4k video render, about 75C. AAn (avx) integration hits 85C in the somewhat cool morning ambient edit 89c with case side on. I dont know many other workloads to test, but given this it might be more than fine for what i do.
*ambient where the case sits above the desk is around 76-77F, so around 25iC ish (fans on 360 blow out the top, 1080ti hybrid rear and 2 120's in the front pull air in, in the v51 thermaltake case) , it would be about 23c (74f) if i had the room to put it more floor level

Cinebench : 6506 (default clocks 6350 at 49x)
Passmark cpumark: 27868 (default clocks 27031 at 49x), no change in fs2020 or xp11 fps regardless here, my integration time did go from 12 min to 9 minutes though (avx).

I guess Ill be leaving cio/sa at auto, seems ok. I will probably move the cpu to fixed 1.315 (the most under load) and actually try to figure out adaptive/offset, guessing i set adaptive to 1.301 and +.014 offset. Tho the cpu mode is set to "auto" not override or adaptive right now, i believe it just still down volt under no load regardless.

The last debate is whether to try the asus ROG Maximus XII Hero which i ordered to arrive tomorrow in case i couldnt figure things out. I'm not convinced that the "better" board would yield lower temps under load, but maybe it would. If temps are higher than this for any other activity (IE: above 88C, ill probably delid, i feel i can do it safely and drop 7c). edit 89c with the case side on during one process, thats borderline too much.

 

[EarthDog]

A better board won't do anything for you unless the voltage is different. You're on auto and its pulling off the cpu so it shouldn't be any different.

Seemsike you're ready to rock and roll!

 

[markm75]

A better board won't do anything for you unless the voltage is different. You're on auto and its pulling off the cpu so it shouldn't be any different.

Seemsike you're ready to rock and roll!

Yea, figured that. I set it to 3.15 fixed. Just need to decide on adaptive if it saves any extra energy. Guessing not. Eist is on

91c on latest avx integration at 25c ambient. Not happy about 91. Would feel OK at 85. Might try the kraken v73 360 and see if I can go down a few degrees. No open loop just yet.

 

[wingman99]

I did some thermal testing with my i9 9900k. It is the 8 cores 16 threads that don't dissipate heat fast enough. not the die shrink earthdog. What I did with i9 9900k is run at 5.0GHz 6 cores, disabled HT, set the core voltage to 1.296v exactly same as my 5GHz i5 8600k 6c/6t. Ran prime95 small FFT on both, no AVX. So the temperature was exactly the same 87c

 

[EarthDog]

I did some thermal testing with my i9 9900k. It is the 8 cores 16 threads that don't dissipate heat fast enough. not the die shrink earthdog. What I did with i9 9900k is run at 5.0GHz 6 cores, disabled HT, set the core voltage to 1.296v exactly same as my 5GHz i5 8600k 6c/6t. Ran prime95 small FFT on both, no AVX. So the temperature was exactly the same 87c

You can't really compare like that...but I'm not entirely sure what you're getting at (or why in someone else's thread - this is about a 10900k).

Edit: to give a quick answer... the 8600k is 149mm die with 6c and doesn't use the solder TIM. 9900k is is 180mm die with 8c/16t and uses solder tim and has more cache. That's 2 more cores (33% more) in 20% more space. Transistor density seems to have increased as well. Those are unlike things.

(Sorry for the edit)

 

[wingman99]

The i5 8600k, i9 9900k, i9 10900k are on 14nm. I did the post to show the OP that cooling trouble is not enough cooling so he would know that the CPU cooling he is thinking of will help. Open loop is needed for a few c with all the cores and threads of 10c/20t

AIO is not going to cut it. Stock the i9 1090OK is TDP (thermal dynamic power) 125w of cooling, who knows how high for overclocked. My i9 9900k is 95w TDP cooling stock.


Transistor density is the same 14nm.

 

[EarthDog]

FYI, I run a 10900k at 5.2 ghz all c/t with a corsair h115i 3x120mm aio. No avx offset (aida64 stress test, cpu/fpu/cache @ ~1.32V. Its at the limit (temps reach 90C+), but it works. These thing need 3x120s...125W is laughable..

View attachment 211884

I know they are both 14nm... there are more transistors on the 9900k than 8600k due to the additional cores. Its also a tweaked 14nm process..14nm+ to ++. My point is simply that you can't compare those two processors as if they are like with temps.

 

[wingman99]

Did you read this article intel TDP link: https://www.anandtech.com/show/1458...ng-an-interview-with-intel-fellow-guy-therien When you clock the processor anything from default BIOS on some motherboard removes the power limit, some you have to adjust the power limit. How Stock TDP works

Guy Therien: Sure. So the general concept is that we have frequency limits – whenever a core is active, such that every CPU has a single or dual core turbo frequency that we publish. Beyond that, depending on the number of cores are active, we have a maximum limit on the frequency. But for that system limit, that’s a frequency limit, and what we really end up doing is enforcing TDP. TDP is an average power that is forced over time. This is important – it’s not just a singular value. As a result, what turbo is really doing is controlling power rather than frequency.

What I believe stock TDP gates the power to clock. My motherboard does not follow intel's TDP guide lines.

I was comparing the processors to show it is the same TDP.

 

[EarthDog]

Correct. Boards are different in how they behave. Some, a few, adhere strictly to Intel boost limits and others, many, do not. If they are removed or longer/higher varies by board. Nothing new there.

 

[Woomack]

TDP is nowhere close to the real requirements. It supposed to be a suggestion of what CPU cooler is required but recent Intel and AMD processors are far from these values. The same will tell you any cooler manufacturer and that's why for example Noctua removed their TDP list from the website.
To keep 10900K at "non-throttling" temps, you need at least 250W TDP cooler and it supposed to be 125W TDP CPU.
My tests are pretty much like this:
200W cooler = 99°C in 1 min of full load and thermal throttling / CPU wattage ~220W, Intel specs and power limit
220W cooler = 99°C in 1 min of full load and thermal throttling / CPU wattage ~220W, Intel specs and power limit
250W cooler = 95°C
300W cooler = 90°C
350W AIO = 85°C

I guess that some reviews are showing 270-280W power draw on the 10900K when motherboards are far from Intel specs. I had up to 250W (more like 248W in hwinfo64 readings) on ASRock, MSI and Supermicro on my 10900K ES with unlocked power limit.

Here is one more comparison as I was playing with underclocking:
i5 10500 auto settings, 6 cores+ht = ~95W max
i9 10900K auto settings besides manual ratio x42 for all cores, locked 6 cores+ht = ~80W max
Simply, you can't compare even the same generation.

 

[Sumguy]

10850k.. So i have to ask i set mine to 48 on all cores @ 1.345 in bios, but in windows under a load its 1.25 or something like that.. where does the concern on vcore lie.. is it the idle vcore or the load vcore.. bc for me to get it stable at 50 it needed like 1.425 in bios and loaded out at like 1.34 i think. i thought that was too high so i dropped it down.

 

[EarthDog]

Load voltage is important.

Use LLC to compensate for Vdroop.

 

[Sumguy]

I have it set to ll4, wont that load vcore also degrade the chip. id like to get to 5.0 but not going to murder the chip for 200mhz
edited

Core temp shows way lower vcore in idle mode then cpuid does.. vid in coretemp shows 1.229, cpuid shows 1.345 - and its set to 1.345 in bios.View attachment 211913

 

[EarthDog]

1.35V 24/7 is fine.

If you are still sagging on load, raise the LLC up to minimize that difference. To confirm voltage, use the MSI program (Command Center) to confirm voltages. Coretemp is generally off and others depend on different factors. CPUz is likely correct.

 

[Sumguy]

so what is the max vcore for daily use.. if i wanna try and get to 5