i9-12900k hit 100C. with H115I Elite Capellix

[ji-loong]

Is this normal? running i9-12900K hit 100C. at full load while using H115I Elite Capellix.

Please advise, if it not normal what should I do or how should I investigate this issue.

My build

Case: meshlicious SSUPD 
Motherboard: ASUS Z690-I
CPU: i9-12900K
RAM:  G.Skill Trident Z5 32GB DDR5 6000 
HHD: Western Digital WD BLACK SN850 2TB x 2
PSU: Cooler Master V850 850Watt 80 Plus GOLD
GPU: ASUS TUF RTX 3080-TI

Sandwich build (left side) CPU

Sandwich build (top view) Left CPU with stand-off, Right GPU

Idle Load

Cinebench multi-core full load 1 min...

Cinebench multi-core full load 5 min...

Cinebench multi-core full load 10 min.

[LeDoyen]

your motherboard unlocks the boost time. The CPU does not fall back to 125W under long load. that's why it heats up.

Also one of your fans is  almost completely blocked, but that's minimal compared to what the motherboard settings do, and you can't do much about it in a SFF case.

Definitely look at re-enabling intel power limits to get back to normal. I am still on 10th gen so i can't help much regarding where it is in bios on a Z690

[ji-loong]

6 minutes ago, LeDoyen said:

your motherboard unlocks the boost time. The CPU does not fall back to 125W under long load. that's why it heats up.

Also one of your fans is  almost completely blocked, but that's minimal compared to what the motherboard settings do, and you can't do much about it in a SFF case.

Definitely look at re-enabling intel power limits to get back to normal. I am still on 10th gen so i can't help much regarding where it is in bios on a Z690

You mean this motherboard do some overclock automatically? am I understand correctly?

[ji-loong]

Look like you mean this one the maximum turbo power of the i9-12900k. it is unlocked by the Asus motherboard while running Cinebench. But this is what we should expected from AIO to cool down the CPU. isn't it?

Edited December 30, 2022 by ji-loong

[LeDoyen]

the CPU should go up to 241W on short duration load then fall back to 125W.

On the motherboard, often, those limits are simply overridden, letting the CPU boost as hard as it can until it bumps on thermal limits to keep maximum performance (looks good on benchmarks).

To be honest, maybe only the biggest AIOs on the market can keep the CPU in check under those curcumstances, or custom loop cooling, and still.. with good airflow on the massive radiators.

you can see you only have one or two cores throttling lightly so it's almost there. every degreee you can gain on water temperature will be seen as a degree saved on the cores. but that difference between CPU and water temp will pretty much be the same on other coolers. That difference is mostly governed by the power sent to the CPU (vcore) more than by the cooler.

And even if you manage to bring the water temp down, the motherboard may still allow the CPU to boost higher, and still show throttling.

To really let a 12900k stretch its legs with no power limits, you'd need more than a H115i, which is not feasible in this case.

restoring the normal spec in power limits would totally solve the problem. Alternatively, you could put your own power limits, and tweak to have as much performance possible without throttling.

[c-attack]

Your CPU is cooled by physically conducting heat from the CPU to somewhere else, regardless of the cooling method type, brand, size, etc.  Most of the AIO (components (liquid, radiator, fans) are in the waste heat removal business and this is what differentiates one type from another.  However, if you cook the CPU underneath at the pins from your voltage level, no cooling method can save you other than somehow reducing the socket area temperature by 30C+.  That is what extreme overclockers do with LN2.  

 

Your problem is not the AIO.  You are 35C coolant temp before starting the load.  In the next shot 60 sec later you are at 97C with an increase in coolant to 37C.  That means only 2C of that 97C is the result of the AIO.  The most efficient water cooling can keep you is at the same coolant temp you started at.  It cannot bring you below the temp level of the case.  So rather than fan settings, pump speeds, fan direction, etc. (all of which matter for longer operation), the 5 seconds later = 97C tells you your problem is right at the CPU level.  This could be less than perfect contact between the CPU lid and cold plate on the CPU block, but the more likely place to start is with the voltage.  I can see 1.31v under what is likely to be heavy Vdroop.  So where does that put you for normal loads?  Maybe 1.48v with a looser LLC?   That's a bit high, but also not uncommon for a lot of these overclocked out of the box motherboards that apply the turbo frequency across all cores and voltage to compensate.  

 

The next logical step is try a stress test with a different instruction type.  R20/23 can be tough for some combos without tuning.  Use the simple CPU test in the "bench" tab of CPU-Z.  It's linear and without AVX or complex instruction.  If that also hits 97C 5 seconds in then I think you are looking at physical contact issues, which should also be giving you trouble at the desktop with browsers and low load stuff.  On the other hand, if it's 20C lower than this is likely just early set-up tuning required if you really need to be running AVX loads, Prime, or other stuff like that.