Peltier subambient cooling i9-10900k over 5.5Ghz all cores?

[tirasoft]

Hello.

I want to try to OC my i9-10900KF towards 5.6-5.7Ghz all cores.

So far , I did try delidded CPU with 360mm AIO (OC speed achievable is 5.5 2cores - 5.4Ghz 4 cores - 5.3 9 cores - 5.2Ghz 10cores) ; also Cooler MAster 360mm Cryo with 5.6Ghz 4 cores - 4.8 Ghz all cores.

I want to try this setup :
delidded CPU + custom water block design for delidded 10th Gen CPU's +
custom water loop : dual D5 pump + 250 mm reservoir + Mo-Ra3 3x420mm rad + 8x200mm push/pull rads
2 x 288 W Laird Peltiers to cool down the water in the main loop
PID custom made Arduino TEC controller (to keep cold side just above the dew point)
each Peltier water cooled with 360mm rad and 3x120mm fans.

Normally, just with a very strong custom water cooling loop 5.5Ghz all cores is achievable (search Luumi on Youtube)
So , with the help of the Peltiers I am wondering if I can up to 5.6-5.7Ghz .

What do you think guys, is it possible with this setup ?

 

[wagex]

You will hit a wall of diminishing returns, if you arent going sub ambient to avoid condensation you're not going to be any better off than just adding more radiator for the water to flow through.

 

[tirasoft]

That's true.
This is why I plan to have 3 water cooling loops : the main one to cool the CPU , and 2 Peltiers each with it's own water cooling loop in order to go subambient.
Dew point here is around 14-16C deg, so this is the temp I want to go down with the water in the main CPU cooling loop. (just above the dew point - to avoid condensation)

 

[Woomack]

Somehow I doubt you will make that with peltiers. You can't count on much better results using single-stage phase cooling and I mean ~5.5GHz would be a limit for a unit that can take ~400W (good luck finding one as most are scaled for ~300-350W max) so about the same as for a higher-capacity water chiller.
In short, 5.6-5.8GHz on all cores is a dry ice area so you need about -70°C to keep the CPU stable. Under load, there will be +30°C anyway.
The main problem is that every 100-200MHz, you have to bump the voltage. Each +0.01V significantly affects the wattage. The voltage is scaling not bad till about ~5.5GHz. Above that, every step is higher (highly depends on the CPU).
Water cooling has a high thermal capacity which is good for 10+ core CPUs, but Intel chips generate a lot of heat on a small surface and the problem will be to transfer it fast enough. So no matter if you add peltiers and expand water cooling capacity, it will still lack the temps on the cores and speed to transfer the heat. This is why for benching you jump from enthusiast-grade water cooling directly to LN2.

I'm not saying it's not possible. It will be just very hard knowing how these chips are acting. Good luck with your tests

 

[tirasoft]

Thanks for your reply @Woomack . If this is all true, than spending the amount of cash to make the cooling happen might be kind of wasted money - and I am reluctant to try.
Probably it will go up to 5.5Ghz all cores for 24/7 use , which is not bad.
In terms of efficiency of heat transfer from the chip, I was thinking that delidding CPU and using a custom made water block (there is a private manufacturer here in UK who make some special waterblocks for delidded 10th gen - which proved to be the best solution - Linus has a video about it) might help.

In terms of exceeding 5.5Ghz on 10th gen, you know there is a Peltier EK waterblock available, tests on Youtube suggests CPU can be clocked all the way to 5.8-6Ghz on single core with serious voltage settings, and CPU temp in the range of single digits celsius - but they couldn't go all cores because Peltier's cooling capacity was too low (max 200-240W).
That is why I was thinking if I increase the cooling capacity and go subambient Peltier way I could achieve more.

- - - Auto-Merged Double Post - - -

PS: on hwbot website there are quite a few Guys who uploaded benchmarks using i9-10900K on custom water cooling 5.6Ghz all cores ...

PPS: and you are right that there is an area of speeds for i9 - between enthousiast water cooling max of 5.5Ghz and what is achievable with LN2 or dry ice - which is like a gap ; I could not see any attempts using strong water cooling and subambient nonnegative temperatures - neither on Youtube or on benchmarking websites .

It could be because it is not possible / or because nobody tried yet (!?)

 

[EarthDog]

It could be because it is not possible / or because nobody tried yet (!?)

...

I'm not saying it's not possible. It will be just very hard knowing how these chips are acting. Good luck with your tests

PS: on hwbot website there are quite a few Guys who uploaded benchmarks using i9-10900K on custom water cooling 5.6Ghz all cores ...

For short benchmark runs, not 24/7 daily driver.

I just don't see the point of making this run that fast as a daily driver. For benchmarking, get on it. But the PITA that is for daily isn't worth the effort (for me) because of the small performance differences.

 

[Woomack]

This project seems like a better idea for current AMD chips as you can achieve more pushing Ryzen 5950X on all cores, even if it won't run at a very high clock. I noticed that even chilled water gives better results than ss cooling and not much worse in general than dice. It's because of high thermal capacity of custom/larger water cooling and large die surface of AMD chips. Some guys in the benching team had good results with water chillers or more standard water cooling with radiators left out the window during winter (eh, guys from Canada ... I haven't seen cold winter for maybe 10 years and I live in Poland).

 

[tirasoft]

Yes, it's making kind of sense what you say about AMD chips.
On the other hand aren't Intel chips much better at scalability ? I mean they respond better to extreme overclocking?

@Woomack
Regarding winters, I understand ... Here in Uk, if there is (very rarely) below 0C and 1-2cm of snow, it is general panic, schools are closing , etc

Using water chillers or leaving radiator out in the cold - same outcome (as I want with Peltiers) = subambient water temperature in the CPU cooling loop.

@EarthDog : indeed I am dreaming of having a RIG both for benchmark and daily use ... Might not worth it in terms of costs and visible daily usage performance (higher FPS in games ; shorter time for Audio/Video editing/rendering maybe ?) - I admit would be more like my personal satisfaction to see 5.7Ghz all core all time.

 

[EarthDog]

I admit would be more like my personal satisfaction to see 5.7Ghz all core all time.

We're here to help!!!

 

[tirasoft]

I mean ~5.5GHz would be a limit for a unit that can take ~400W (good luck finding one as most are scaled for ~300-350W max)
Good luck with your tests

My CPU's all core 5.2Ghz on Linpack test is drawing 360W (1.415Vcore)
So, I was just assuming that if speed is 5.5Ghz and higher, and especially Vcore is higher - power draw will be more than max I have now - those 360W.
Are you suggesting there is a wattage limit for each particular CPU bin as a maximum ?

Doesn't the power draw go higher with Vcore/speed?

 

[Silver Surfer]

That's true.
This is why I plan to have 3 water cooling loops : the main one to cool the CPU , and 2 Peltiers each with it's own water cooling loop in order to go subambient.
Dew point here is around 14-16C deg, so this is the temp I want to go down with the water in the main CPU cooling loop. (just above the dew point - to avoid condensation)

Interesting idea, did you ever follow through?

FYI: There is a lot to discover using Peltiers for cooling in the range you are considering, but asking others opinions and advice that have never actually done it, usually leads to negative comments that dissuade you from going through with your initial ideas. That way you never discover if your idea was good or bad or needed some simple modification to acheive good successful results. The cost of experimenting using Peltier cooling in the below ambient range is usually what stops many from doing it as it is an unknown, until you actually do it.

Consider this, Do you think genius like Einstein, Nikola Tesla, etc. asked others before they implemented their ideas, I doubt they did, they just did it! Ryan

 

[EarthDog]

but asking others opinions and advice that have never actually done it,

My experience with them is that it's a PITA to maintain for such little gains. I (and Woomack) also know these chips better than most here (I'd say). I don't see anything wrong with putting all the facts out there so the OP can make an informed decision. Hell, I even said we're here to help as one of my last posts...

Are you suggesting there is a wattage limit for each particular CPU bin as a maximum ?

Nope.

Doesn't the power draw go higher with Vcore/speed?

That's exactly how it works, yes. GHz requires more voltage which makes more heat you need to remove.

 

[Woomack]

I was testing peltiers some years ago and there is a good reason why people don't do that anymore. Water chillers for 24/7 usage or LN2 for short time benching is the way to go looking at cost/problems and results.
As long as I wasn't testing the 3-peltier combo, then all the specs suggest it would perform like a water chiller with double the power draw. Of course, I'm not stopping anyone from tests.

 

[don256us]

Yeah. I haven't seen anything on peltiers for many years. I can remember when they were picking up steam and a lot of people trying them but water cooling became more mainstream and petiers fell out of favor due to the reality of how they work.