3950x watercooled temp issue.... please come in

newls1

Supreme [H]ardness
Joined
Sep 8, 2003
Messages
4,607
LEt me start off an state that ive never had a cooling issue EVER... been watercooling everything since the late 90s and this is the first time im clueless and need help. Here is the exact issue im having:

My 3950x is cooled with the following components..... EK 60mm thick 420mm rad in push / pull using very well known yate loon fans. Heatkiller PRO IV all copper waterblock for AM4, D5 pump/res combo ... all 150% proven parts and actually pull outs from another build in my pc room.

Current OC on my 3950x is all core 4.35Ghz @ 1.3125vcore / 1.1 SOC and as soon as I load aida64 stress test, cpu spikes to 92/93c in a 21c room. Here is the suspicious part..... As soon as I "END" the stress test, the temps SLOWLY drop down, like 90-88-80-75-73-70-65- etc... over about a 15sec time frame. Never have I seen this, ive ALWAYS had it when I stop a stress test, the temps immediatly drop to mid 30's or 20's, etc... not over a 15 sec time frame. So this tells me a bad mount. so i confirmed mount a few times and the mount is full contact with tons of pressure. So do amd cpus read core diode temps differently then intel and this is normal behavior, or is there possibly something else at fault here. Heatkiller blocks have never let me down, but im not ruling this out. was thinking about picking up a EKWB velocity at microcenter to "TRY" to see if this has different results but that is an expensive 110$ "try"....... help please? whats your input.... thank you
 
I'm not too familiar with Heatkiller blocks, but I do know the Alphacool's were cooling less than optimally for AMD's 3000 series (something to do with die placement on them). I believe, last I checked anyway, Alphacool was going to work on an updated design of their waterblocks to address this.

Whether that's the same for your Heatkiller blocks - I dunno.
 
SMT off.Do you really need 1.3125 vcore?

Few other posts read.Need 1.225v at 4300. SOC 1.0875. 73 max temp prime 95.

Maybe there ambient is ultra low :).
 
Last edited:
what's the temps between the chiplets? there should be a sensor on each one is there a huge delta between them? usually would give another idea on if it's a bad mount or not as well. could also be thermal paste.. my guess is that due to the CCD layout there's not enough microfins over the actual chiplets and looking at images doesn't look like you could turn it 90 degree's. i know that was a problem i had with my block where the microfins were longer than they were wider so i had to turn it to actually cover the chiplet on my cpu.
 
I'm not too familiar with Heatkiller blocks, but I do know the Alphacool's were cooling less than optimally for AMD's 3000 series (something to do with die placement on them). I believe, last I checked anyway, Alphacool was going to work on an updated design of their waterblocks to address this.

Whether that's the same for your Heatkiller blocks - I dunno.
this is eactly what i was thinking and EKWB's velocity block cooling channel goes the opposite direction as heatkillers, so this may prove to be a good idea to buy one and try... and report back
 
what's the temps between the chiplets? there should be a sensor on each one is there a huge delta between them? usually would give another idea on if it's a bad mount or not as well. could also be thermal paste.. my guess is that due to the CCD layout there's not enough microfins over the actual chiplets and looking at images doesn't look like you could turn it 90 degree's. i know that was a problem i had with my block where the microfins were longer than they were wider so i had to turn it to actually cover the chiplet on my cpu.
im using coretemp to report my temps, and that program only displays 1 temp... should i read temps with a different program? maybe this is my whole issue
 
im using coretemp to report my temps, and that program only displays 1 temp... should i read temps with a different program? maybe this is my whole issue

check in HWinfo64 you should see both chiplet tdie temps.
 
check in HWinfo64 you should see both chiplet tdie temps.
thank you SO MUCH, this might very well be my issue. core temp could be reading wrong sensor or something.. when I get home sunday, i will check this out. Figured something weird was going on, cause if i really had a core temp of 92/3c my water temps in rad would be alot warmer then 38c
 
I'm cooling a 3900x with the same block and am having none of the issues reported by OP. I'm not overclocking, but top out at 60c and don't see the "slow drop" after a stress test.

I don't believe the design of the Heatkiller block is at fault here.
 
LEt me start off an state that ive never had a cooling issue EVER... been watercooling everything since the late 90s and this is the first time im clueless and need help. Here is the exact issue im having:

My 3950x is cooled with the following components..... EK 60mm thick 420mm rad in push / pull using very well known yate loon fans. Heatkiller PRO IV all copper waterblock for AM4, D5 pump/res combo ... all 150% proven parts and actually pull outs from another build in my pc room.

Current OC on my 3950x is all core 4.35Ghz @ 1.3125vcore / 1.1 SOC and as soon as I load aida64 stress test, cpu spikes to 92/93c in a 21c room. Here is the suspicious part..... As soon as I "END" the stress test, the temps SLOWLY drop down, like 90-88-80-75-73-70-65- etc... over about a 15sec time frame. Never have I seen this, ive ALWAYS had it when I stop a stress test, the temps immediatly drop to mid 30's or 20's, etc... not over a 15 sec time frame. So this tells me a bad mount. so i confirmed mount a few times and the mount is full contact with tons of pressure. So do amd cpus read core diode temps differently then intel and this is normal behavior, or is there possibly something else at fault here. Heatkiller blocks have never let me down, but im not ruling this out. was thinking about picking up a EKWB velocity at microcenter to "TRY" to see if this has different results but that is an expensive 110$ "try"....... help please? whats your input.... thank you

That's got obstruction written all over it in my opinion.
 
I'm using an Alphacool Eisblock on my 3900X on the test bench and I'm not seeing temps like the OP at all. I'm not even using a great TIM. I'm rocking a graphite thermal pad and its still doing OK. The Alphacool block I'm using is one that I had back in the Ryzen 2000 series days, so its not even an updated design. Something's definitely wrong with your setup.
 
thanks everyone for the ideas... when i get home sunday i will check the pump
 
thanks everyone for the ideas... when i get home sunday i will check the pump

Not just the pump. You have an obstruction somewhere in yer loop. The pump is the last place I'd look. The milled fins of a cpu/gpu block make great sources of obstruction, hint hint.
 
ill have to check rad then too cause everything else is new
 
New does not mean it can't be the problem. You might be able to feel the heat in hoses on output side of waterblock to figure out where restriction is. If it's in block the heat won't get out into hose very far. Of course, the longer system (pump) runs the farther from source the heat will move, even if very slowly.
 
Did you clean/flush everything before assembling yer loop?
yes of course... thats why im really not thinking this is the issue... but ill double chk everything when I get home. built this loop coming off a 48hr shift, so many things could have been overlooked. i try to NEVER build pc's coming off a long shift like that for reasons like this, but i was to excited to start playing with the 3950... ill surely update this thread when i get home off this 48 and dismantle everything
 
yes of course... thats why im really not thinking this is the issue... but ill double chk everything when I get home. built this loop coming off a 48hr shift, so many things could have been overlooked. i try to NEVER build pc's coming off a long shift like that for reasons like this, but i was to excited to start playing with the 3950... ill surely update this thread when i get home off this 48 and dismantle everything

Oh ok. Sorry just had to ask to confirm. Good luck on yer teardown.
 
ok, little update.. physically spent ALL DAY tearing apart this 1 day old loop and "RE-CLEANING" rad, pump, and even bought a EK VElocity WB to see if temps changed any cause the heatkiller block flows water 1 way over the cores, where the EK flows water the other way across the cores, and after re-assembly, temps are very similiar. this time I used hwinfo64 to read both die temps and this seems to be a little lower, mostly in the 80's under full aida64 load. So ive eliminated any possible thought of debris/restriction in loop, and pump and res fully clean, along with the rad. so either im going to live with these temps, or figure out how to do a oc based on per ccd. I would like to find out how to find which ccd is my strong one and try to go to 4.5ghz on that one, and 4.3 on the other ccd.... or something like that.
 
Well you dropped your max temps 15c, that is a freaking lot of C's to drop. The biggest positive you netted is that you are now under the throttle temp which is somewhere around 85c depending on bios. Also, I suggest you upload a pic of your loop. Btw, your loop is cpu only with just the EK block and an AK XE420mm rad? Test your loop with two apps Cinebench R20 and Prime95 (at least 10minutes). That will give me a reference point temps wise to my setup.

As for the ratio oc, you can follow my suggestions in that other post.
 
Well you dropped your max temps 15c, that is a freaking lot of C's to drop. The biggest positive you netted is that you are now under the throttle temp which is somewhere around 85c depending on bios. Also, I suggest you upload a pic of your loop. Btw, your loop is cpu only with just the EK block and an AK XE420mm rad? Test your loop with two apps Cinebench R20 and Prime95 (at least 10minutes). That will give me a reference point temps wise to my setup.

As for the ratio oc, you can follow my suggestions in that other post.
ill upload pics of the setup in just a few. I have 2 loops going. 1st loop is cpu only and 2nd is gpu only. HWinfo64 die temps do NOT match what coretemp reads, so that explains the temp difference. I appreciate all your feedback so far.
 
Is this your first time doing water-cooling with an amd cpu? Is the pump at full speed or less than 50% speed?
not hardly.... my first AMD watercooling adventure dates back to the barton core days, were you even around? Im very well versed in watercooling, just never cooled a "many core" cpu before and was shell shocked by temps
 
Just got my build together yesterday - 3900x with copper Heatkiller 4, Alphacool 420x45 + 280x60 + EK D5 - so sorta similar. I don't think I saw any issues like what you're describing, but I'll run aida when I get home to compare.

The only "weird" thing I noticed - I had a leftover burr from manufacturing on my IHS on two adjacent edges. A razor blade with light confirmed the "non-flatness" I would've had significant gap for the TIM to fill had I not sanded those edges over.

I would think that, in general, quick rises in temperature and slow drops after no load point to a contact problem. Block/IHS, TIM, or IHS/Silicon contact

I think with flow problems you'd see slower going up and slow going down since you still have the thermal mass of the block to heat going up.

A der bauer video delided a ryzen pretty easily and replaced the thick indium with LM which reduced the gap between the silicon and IHS - maybe you have a bad indium sheet application from factory?
 
Last edited:
can you tell the flow is pretty quick from your res? also after apply thermal paste, does the paste spread evenly throughout the CPU heat spreader after clamp down the block. whats temp when using stock air cooler with and w/o OC? does the CPU temp normal under stock operation?
 
Alright so for comparison's sake, keep in mind I haven't got any real overclock, just PBO, RAM at 3800c14, FCLK at 1900.

In Aida Extreme (CPU, FPU, cache) I'm averaging 1.23 Vcore 1.075 SoC and 4.04Ghz at 73C over the course of about 10 minutes.

It hops right on up to 70 in the matter of a second or 2 at the start. When I end the test it takes about 10 seconds to get quite close to idle temp after 10 more its back to idle temp.

Also forgot to mention, I'm using conductonaut as a TIM and I have my block mounted with input at the top.
 
Pointelss to get a water cooling for a CPU that cant overclock worth jack.... It comes out of the box optimized and OCed to where it should be. Most you get is 200Mhz in a OC, So why you need 5 rads I have no idea..........
Watercooling isn't only about increased overclocking headroom.
 
Most of the peeps I know using expensive custom loops don't push overclocks much if any higher than peeps with top tier air cooling do. Most water coolers' enjoy building and having H2O systems. Many readily admit a top tier air cooler in case with good airflow supplying cooler with air only 2-5c warmer than room at full load cools about the same as their water cooled system.

I used to be water cooled, but got tired of the extra weight, time, and expense of building and maintaining water compared to time it take to install and maintain air cooling. A good case with filtered intakes only needs filters cleaned (I vacuum mine when I vac the room). System inside of case is still clean in 3 year old system. ;)

newls1
I agree with others. Symptoms sound like heat transfer from chip to waterblock or coolant flowrate issue
 
Pointelss to get a water cooling for a CPU that cant overclock worth jack.... It comes out of the box optimized and OCed to where it should be. Most you get is 200Mhz in a OC, So why you need 5 rads I have no idea..........

I wouldn't go as far to say pointless, but yeah Custom loops are hardly ever a "need". I do plan on adding a waterblocked 2080ti in short order which will better use that capacity. I'm also a fan of the silence, it's all packed into an R6 case.

I did almost go with noctua-12s, saving the money and effort. It turned out so cool though, I have no regrets. :D
 
I have no regrets. :D
hqdefault.jpg
 
Back
Top