Overclock.net banner

new custom loop temp problems

2K views 70 replies 18 participants last post by  Jpmboy 
#1 · (Edited)
ISSUE WAS SOLVED LOL THANKS. ,just recently switched over from a corsair h115i to a custom loop with 560mm gtr rad in push pull with 8 noctua 3000rpm fans, d5 pump, 924ml koolance reservoir and heatkiller iv pro block. everything mounted open air bench table. problem is my temps are averaging 11c higher than previously. and with this gear it is very unusual that a 280mm aio performed 11c colder than this current setup. I already tried re applying thermal paste and remounting the block. used kpx first mount and kryonaut 2nd time no difference. also ran the loop for 12 hours over night at max pump speed to bleed air. took about 1.8-2 liters of coolant to fill the loop. used 75% distilled water and 25% mayhems xt1 nuke. going to attach a pic of setup. any ideas on why this is running 11c hotter? ambient temp the same.
 

Attachments

See less See more
3
#2 ·
What are the temps on idle ? What are the temps when on full load ?
 
#5 ·
Idle temps are 30c and 1 run of cinebench r15 max temp is 79c. where previously idle temps on the h115i were 27c and max temp in cinebench was 63c. thats a big difference if you asked me already remounted 4 times. twice with kpx and the other two times with kryonaut. and the last two remounts i made sure the heatkiller
was tightened all the way down. no difference. Dunno what else to do.
 
#6 ·
Koolance claims the reservoir has redundant ports. so i go out of the res into the d5pump with heatkiller top, out of the pump into the heatkiller block (the port with the arrow which i am assuming is the inlet) than out of the block into the rad and out of the rad back into the res.
 
#4 ·
Try using an additional washers under the mounting springs for more mounting pressure.

But yes, making the flow is properly aligned with the inlet would be helpful.
 
#8 ·
Did you check the cold plate to see if it was perfectly flat?

I'd be suspicious of air bubbles if it checked out ok. I raised my reservoir to the highest point in the loop, opened the vent on top and ran it that way for a while. Surprising how long it took to stop seeing bubbles come up through reservoir.
 
#14 ·
Did you check the cold plate to see if it was perfectly flat?

I'd be suspicious of air bubbles if it checked out ok. I raised my reservoir to the highest point in the loop, opened the vent on top and ran it that way for a while. Surprising how long it took to stop seeing bubbles come up through reservoir.

You should be just hearing trickling water which is a nice cool sound we want under water.
 
#9 ·
...I had a similar scenario with a 5960X - custom loop was way more powerful (re. rad space, fans, cooling media volume),but temps were much worse than H150i Pro AIO. It turned out to be the IHS of the CPU, which - hard to believe - was both concave and convex at the same time...the worst I have ever seen. The AIO happened to have a better 'matching fit', but after some serious lapping of the CPU, it finally worked (with the custom setup much cooler than AIO at idle and under load).

With your testbench, you can try another 'trick', which is rotating the CPU block by 45 degrees, than 90 degrees, than 180 degrees and checking each time for idle and load temps. After that, another step would be the spin method > spinning the CPU on its IHS on a flat surface...it really shouldn't spin very much at all if it has a nice flat surface.

Finally, apart from correct loop sequence (in, out etc) mentioned above by others, also make sure that the custom CPU block does not have debris in it, either from a previous setup, or even via a new rad. Every single new rad I ever got had a lot of little black and brown bits in it from manufacturing...lot's of flushing in the bathtub until no more debris comes out...

Good luck :)
 
#12 · (Edited)
so i think i solved it. i rotate the block 90 degrees and so far seeing way better temps. looks like the block was sitting on 4 capacitors that werent letting it make full contact with the ihs. now that i rotated it as suggested all seems good. And great news at max load so far not even reaching 60c. temps improved greatly with this setup compared to the AIO thanks everyone for the help. thanks j7sc for the suggestion.
 
#18 · (Edited)
so here are some statistics. Testing was done with 1 hour of realbench with the corsair h115i and then after with my new custom cpu loop which consists of; 1. HW Labs 560mm GTR w/ 8 noctua 3000rpm industrial fans in push/pull, 924ml koolance reservoir with their base and top, d5 pump ran at 4800rpms with a heatkiller d5 top, heatkiller iv pro cpu all copper block. Total performance increase of -7 degrees celcius at load. here are some screenshots. first pic is h115 2nd pic is custom loop. ps only 4 of the noctuas were connected in push for this test still need to buy some splitter to connect the other 4. 8700k @ 5.1core-1.395vcore-4.8cache, ram @ 4266mhz 17-18-18-38-328-1t 1.46vdram, 1.25vIO, and 1.3vSA so its overclocked pretty well. also delided.
 

Attachments

#37 ·
Try bringing down the speed of your pump to ~3000 rpm and see what type of result you get. Faster is not always better in this scenario, especially with the D5 being cooled by the liquid in the loop. I see a several degrees better at 2700 rpm than at 4800 rpm in my loop, as an example. Play with it a bit and you may find a pump speed that drops both noise and temps.
 
#19 ·
That's about what I gained going from an H-100 to my 480mm loop on the 1800X rig.
 
#20 ·
Nice going with the lapping. What are you temps now ?
 
#21 ·
I didnt lap it I just had the block mounted incorrectly so the capacitors on the board got in the way of it sitting flush on the ihs. I resolved this by rotating the block 90 degrees. previous temps after running realbench for 1 hour with the h115i were 68c max and 54c avg. Now with 1 hour of realbench they are 61c max and 57c avg. So i gained 7c of headroom.
 
#22 ·
Honestly, you should lap it lol. I lapped my 9900K to bare copper mirror finish with 3000 grit final stage.

It runs at 5.1/4.8 @ 1.248v VR VOut and is cooled with a EK Phoenix 280 kit and I went from 96c hottest core in Prime95 AVX/FMA3 small FFT 1 hour to 84c lol. Mine was convex as all heck so the paste (Prolimatech PK-3 is my go-to paste) all puddled in the middle and didn't spread really well.
 
#23 ·
Hi,
Lapping will flush 3 year Intel warranty down the drain.
 
#24 ·
Well, legally no. Unless Intel can prove that the modification directly caused the chip to fail, federal law protects your right to modify any product without voiding the warranty. But its really up to how willing you are to draw out a court battle over a $500 CPU :p
 
#40 · (Edited)
so i did 5 runs of cinebench with pump speed at 4800rpm, 3600rpm, and 2400rpm. and these are the results.

4800rpm 1st run core max 56c, 2nd run 58c, 3rd run 58c, 4th run 58c, and 5th run 58c
3600rpm 1st run core max 57c, 2nd run 58c, 3rd run 58c, 4th run 58c, and 5th run 58c
2400rpm 1st run core max 57c, 2nd run 58c, 3rd run 58c, 4th run 58c, and 5th run 58c

ps my d5 pump is 3 inches from the fans on my radiator so it gets ample airflow and is never really warm to the touch. testing was done with 4 rad fans in push @ 800rpm
20.3c ambient air temperature measured with a fluke thermometer with the probe placed next to the system. 8700k @ 5.1 core - 4.8 cache - 1.39vcore no avx offset
 
#41 ·
Ok, so no change in temps with speeds from 2400-4800 rpm. At least you can quiet it down with no loss in performance if you like.

Your D5 pump would not get warm to the touch, nor would it's location near the fans make any difference. It is cooled by the liquid in the loop, not externally by air. DDC pumps are air cooled, not D5s.
 
#44 · (Edited)
The problem can be poor contact. A CPU is either concave or convex.

Apart from lapping or removing an IHS, a different block can help. The reason is this. Some blocks are machined to be slightly concave or convex to better match certain CPU's. If you place a concave block on a convex CPU, it'll make good contact.

In an ideal world, an IHS should be flat, which would make things simple for block manufacturers.
 
#53 ·
you'd have to drop below .5gpm to see any loss. .5-1gpm is the sweet spot, above that gains are minimal.


maybe I missed it, what're your ambient temps?

I'd also use prime over cine, it'll give you a better idea of what your actual max temp is. If the temp "issue" is immediate its between the block, & the chip. Specially if you've already ruled out flow. if the temp "issue" slowly shows up, then its a different story.

sometimes its beneficial to rotate the block 90 degrees, and inspect the internals to confirm the correct jet plate is installed.
The temp issue was solved a couple of days ago - the block was hitting board components. Rotating the block solved the issue.
 
#45 ·
you'd have to drop below .5gpm to see any loss. .5-1gpm is the sweet spot, above that gains are minimal.


maybe I missed it, what're your ambient temps?

I'd also use prime over cine, it'll give you a better idea of what your actual max temp is. If the temp "issue" is immediate its between the block, & the chip. Specially if you've already ruled out flow. if the temp "issue" slowly shows up, then its a different story.

sometimes its beneficial to rotate the block 90 degrees, and inspect the internals to confirm the correct jet plate is installed.
 
#46 · (Edited)
The difference between his loop and his AIO was a night and day different from the get go. He rotated the block which helped resolve the issue. It is definitely the water block and IHS making poor contact. OP suggests the block was hitting caps on the MB, reducing contact ability.
 
#47 ·
must be another Ryzen + EK block user. oh, that makes way more sense. I loaded directly into the last couple posts -.- silly OCN.
 
#48 · (Edited)
It is a Heatkiller IV PRO, but I don't have a clue what CPU it is. There was talk about Intel warranty a few pages back in relation to lapping the IHS, so it is probably a socket 1150/1151 CPU.

My 9900K doesn't make great contact with my EK block (the older style) so I know how the OP feels.
 
#49 ·
i guess it would help if I put the effort into actually going back & reading the OP, since OCN thought I could just skip to the end. It's a bit strange to hear its a heatkiller. also strange that the AIO out performs the loop.

pretty sure intel's OC warranty won't even cover lapping. IHS must be legible, cuz its totally the only way they can confirm authenticity of the chip. There was one little caviat I got by on with my 4930k, but IDR what it was.

-.- I'm the worst. i've put nearly zero effort into this 9700k build. I CLU'd my Supremacy, dialed in a 5.2, and went about my business. I have zero clue what temps it runs at :p
 
#50 ·
OP checked the block and found it was hitting some caps on the MB, so he rotated it and now the custom loop is doing 7c better than the AIO.

Hey, you might want to have a quick look at those 9700K temperatures when you get a minute. :p

My loop is 360 + 420 Rad's, EK block, D5 pump, EK res, 5GHz HT @ 1.3v, ambient 23c, RealBench highest temp 76c. Not very good if you ask me. Deffo the contact to blame. Might try rotating my own block at some point.
 
#51 ·
so rotation was the solution after all, as it appears the block was resting on some bits leading to lack of pressure.
 
#57 ·
Kryonaut is the best paste, you could try orienting the block 90 degrees different. I had a rig where, because of the layout on the CPU and the way the water flowed in my block, it made a huge difference in cooling. LOL I just read another post recommending basically the same thing...

I'm a little puzzled why you went full custom loop when you already had decent temps ... I've never had trouble with my AIO's.
 
#63 · (Edited)
I didnt see that the problem was fixed, good luck
 
#65 ·
Warranty is no longer valid if you:

1) Remove IHS.
2) Lap the IHS.
3) Replace thermal solution.
4) Reinstall IHS.
 
#66 ·
5) remove markings from IHS in any way shape or form, say like, while using the scotchbright pad that CLU used to come with. (may still)
 
#71 ·
same here. was tough getting good pictures of the q-core thing for Intel to scan.
 
#69 ·
Hi,
Glad that's all settled :)
 
#70 ·
also known as ... "QED"
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top