• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Direct Die with AIO cooler with rockitcool 10th gen - 10900k?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
M

markm75

Registered
​
Joined
Sep 28, 2008
I have the 10900k and thermaltake aio 360 (might be switching to cool master 360 to see if helps with a newer aio), realizing that a real water cool loop is more ideal but i'm not up for that just yet.

My temps reach 90-95C during stress tests (1.305v LLC4 at 50x, 51x slightly unstable, Drdram 4k, 1.26 SA/CIO, 1.49v), but real world loads dont exceed 87C, usually 68-79 on average unless i'm rendering. I hit about 87C on med occt sse. Barring the new aio not helping i was considering delid, but realizing delid might only gain 5-7C i spotted the delid and direct die rockitcool kit which in the yt video i found they had good results 12c drop

https://rockitcool.myshopify.com/co...cts/10th-gen-direct-to-die-frame-kit-complete

Does anyone know if there is an issue attaching a standard aio water cooled solution to the chip? Can the standoffs that come with the aio work or will some 3rd party solution be needed to line things up and is liquid metal applied to both the die and the block (or center of the block maybe).
Thanks in advance
 
Last edited:
Please update your signature so we know what you are working with. :)

Why is your SA/IO voltage so high? What speed memory are you running? Up to around 4000 MHz, you shouldn't need much more than stock/1.15V,1.05V...that may save you a couple of C by lowering it.

These CPUs are using the solder TIM, delidding isn't going to get you much. Going bare die is also pretty risky, though with the kit that mitigates the issue some, surely. 12C isn't going to do much for you... if you're lucky, you may get another 100 MHz out of it, so make sure you're doing this for the right reasons. If you're benchmarking competitively, have at it... otherwise... not sure I would take the risk for the meager gains.
 
EarthDog said:
Please update your signature so we know what you are working with. :)

Why is your SA/IO voltage so high? What speed memory are you running? Up to around 4000 MHz, you shouldn't need much more than stock/1.15V,1.05V...that may save you a couple of C by lowering it.

These CPUs are using the solder TIM, delidding isn't going to get you much. Going bare die is also pretty risky, though with the kit that mitigates the issue some, surely. 12C isn't going to do much for you... if you're lucky, you may get another 100 MHz out of it, so make sure you're doing this for the right reasons. If you're benchmarking competitively, have at it... otherwise... not sure I would take the risk for the meager gains.
Click to expand...

i was pretty sure at lower volts it wasnt passing stress tests sa/io, which would be the 1.15, sa? will retest. I guess if i just live with the occassional render that causes 87C then i probably shouldnt worry about delidding or direct die to be safe. Its not like i run those every day, maybe 3-4 times per month at most for several hours each max. Otherwise it might peak at 83C on a load of xplane 11 for instance but after that stays in the 70s.

I updated my sig :)
 
Thanks for updating! I'm confused. YOu have DDR4 4000 RAM at 1.49V and running its stock (4000) speeds? I'd imagine 1.35V for the DRAM voltage is plenty... they're tested that way.

87C is nothing! That's (part of the reason) why you stress test... to see worst case temps (and stability of course). I think you are fine, personally. :)
 
Just a heads up to the OP, you're not supposed to have links in your signature. Only exception is for Senior Members, and links within OverClockers.com and or these forums.
 
EarthDog said:
Thanks for updating! I'm confused. YOu have DDR4 4000 RAM at 1.49V and running its stock (4000) speeds? I'd imagine 1.35V for the DRAM voltage is plenty... they're tested that way.

87C is nothing! That's (part of the reason) why you stress test... to see worst case temps (and stability of course). I think you are fine, personally. :)
Click to expand...


I tried vcsa 1.15 and cio at 1.10 even, when i did this i couldnt boot for some reason.
I noted that many suggested 1.25, but maybe too high?
However, if i leave those to auto, sa ends up around 1.31 and cio around 1.24

I left ram at 3200, volts 1.4, auto on sa/cio, cpu i was able to set to 1.29 this time.. ran p95 blend 12 hours fine, even did small ffts for around 30 min

I then set ram to 4000, left rest as is (auto cio etc).. tried realbench 2.56 set to 32gb and ran it... bsod (usually watchdog errors) after 26 mins.
So even at auto this failed.

I think meaning needs more vcore since cio/sa were auto.

My temps hit about 88C (realbench)

Other notes..

I'm using v51 case, i have the radiator up top, i have the exhaust as the 1080ti hybrid output, i have 2 fans on the front pulling air in. My case however sits above my desk (around 4 feet off floor i'd say), so that doesnt help.
I've thought maybe putting the thermaltake 360 in the front my be better and push air in across.
 
Auto always overvolts in favor of stability. If your chip needs it, it needs it. That said, I just noticed it is 32GB so, makes more sense now.

Again though, you keep mentioning 3200 and "volts" at 1.4.... your kit runs at 1.35V. There shouldn't be a need to increase that.

If you change the rad setup, you may get a couple degrees better temps on the CPU, but it isn't running hot...
 
EarthDog said:
Auto always overvolts in favor of stability. If your chip needs it, it needs it. That said, I just noticed it is 32GB so, makes more sense now.

Again though, you keep mentioning 3200 and "volts" at 1.4.... your kit runs at 1.35V. There shouldn't be a need to increase that.

If you change the rad setup, you may get a couple degrees better temps on the CPU, but it isn't running hot...
Click to expand...

Well the specs on the ram i have actually have it at 4000 and 1.40v according to newegg and the sticker anyway.

I had things at 1.305v and 1.26 cio/sa before and it was still unstable at 4000, so i guess given i had 4000 and auto on cio/sa and realbench failed at 26 mins, i should probably try around 1.31v with auto cio at least and see if i make it 1 hr on real bench (being at 1.29 now), though that 88C will probably be about 91C i think, not ideal but ok

I view realbench as a pretty much equivalent to much of the rendering i do at times, i had similar temps to realbench (avx processing).
I also need to re-set the ring to 47x, its on auto, 47x seemed ok before.

In the msi bios there is "dynamic" and "fixed cpu" for the ratio setting, this an additional setting besides "all core", i have it on fixed, not sure the difference.
Of course right now LLC is still on auto, before i had it on LLC4.
Ill probably nail down the vcore then maybe try to set fixed sa/cio, unless leaving on auto is fine (or maybe too high).

edit: bsod even at 1.29 and 3200 ram setting about 50 mins into realbench so i guess not stable at 1.29 either (and 3200)
 
Last edited:
Well the specs on the ram i have actually have it at 4000 and 1.40v according to newegg and the sticker anyway.
Click to expand...
Ahh.... ok, yes. Some kits are like that. I thought more commonly they are still 1.35V (I have two sets that are). My fault!

You may want to isolate testing a bit as realbench does a lot of everything, it's hard to pin down. Run small FFT for CPU (prime 95) and see if it works... then run blend on it for more memory. If those run for a couple of hours, it should pass realbench.
 
EarthDog said:
Ahh.... ok, yes. Some kits are like that. I thought more commonly they are still 1.35V (I have two sets that are). My fault!

You may want to isolate testing a bit as realbench does a lot of everything, it's hard to pin down. Run small FFT for CPU (prime 95) and see if it works... then run blend on it for more memory. If those run for a couple of hours, it should pass realbench.
Click to expand...

No worries :)
I ran a slightly older p95 small fft earlier (avx off), in about 1 min time temps hit 99-100C, then a watchdog bsod, but if i run occt small it seems to stay a tad cooler enough to test anyway.

Definitely wasnt stable in realbench even at 3200 at 1.29, failing at 50mins, so more vcore.
 
EarthDog said:
Leave memory to default, 2133 or w/e and see if it still fails.
Click to expand...

Yeah i guess i can drop it back to 2133 from 3200, i guess if it doesnt fail then more vcore either way (at 3200-4000) Hoping my difficulties here arent because i chose this msi over typically asus (extreme II was an option). I think its just chip lottery though. And stock is nil on switching to another one right now, so kinda stuck.
 
Last edited:
First it passed at 1.29v with 51x and 2133 and ram auto on everything else without xmp.

Then, it passed at 1.29v 51x with auto on everything but the vcore and ram xmp/4000. This on small occt for 30 min, 1plus hours of realbench etc.
Soon as i set ring cache from auto to 47x i get page fault bsod crashes during realbench.
So i guess to have cache above 43x it needs more vcore. My temps during realbench hit about 88c with 47x ring cache and 1.29v (with llc at auto).
 
markm75 said:
First it passed at 1.29v with 51x and 2133 and ram auto on everything else without xmp.

Then, it passed at 1.29v 51x with auto on everything but the vcore and ram xmp/4000. This on small occt for 30 min, 1plus hours of realbench etc.
Soon as i set ring cache from auto to 47x i get page fault bsod crashes during realbench.
So i guess to have cache above 43x it needs more vcore. My temps during realbench hit about 88c with 47x ring cache and 1.29v (with llc at auto).
Click to expand...

When increasing cache frequency, you need to raise vcore to accommodate for it.
 
ShrimpBrime said:
When increasing cache frequency, you need to raise vcore to accommodate for it.
Click to expand...
Eaxctly right.

@mark - make sure your work benefits from increasing the cache speeds. In some (many?) cases it doesn't matter much , especially considering the voltage/temp increase that comes with it.
 
I ended up successful at 1.29v and 46ring (i think around 87C realbench 92c prime95 blend). Now i'm trying to see if i can manually lower the cio/sa. Sa automatically ends up at 1.320 and 1.243 cio.. i think the sa is a tad high as is. Going to try fixed 1.25 sa and 1.24 cio as a start.

For the avx type graphical rendering integrations i run, i think higher cache might be of use. Not sure if higher ring helps with say flight sims such as xp11 and msfs2020 with fps, guessing not by much.
 
Last edited:
Well, trying 1.25 sa and 1.24 cio failed. I guess I'll leave on auto though sa is high at 1.32ish at 4000. 4200 with those on auto also failed so I think maybe there I just need to tweak timing
 
Well I went back to my stable settings and now even those are suddenly failing realbench with page fault in non paged area. Thry were fine for hours on various tests but now thry aren't. At a loss here
 
You must log in or register to reply here.
Back