Just can't cool down that Threadripper 3970x. Please send help!

[EvilVir]

Hi, I'm new here so hello everyone! :)

 

I'm looking for help with my rig, no matter what I do I can't cool that thing down at all. Plus there are some other (minor?) issues, that maybe someone will be able to help me with.

 

So specs first:

• CPU: AMD Threadripper 3970x
• GPU: Gigabyte Windforce GeForce 970 (yeah, that's not a gaming machine)
• RAM: 8x32GB GSkill RipJawsV 3200MHz (yes, I know 3600MHz would be better but I learned this too late)
• PSU: Corsair HX1200i
• Drives: 3x 1TB Corsair MP600
• Case: Phanteks Evolv X
• OS: Windows 10 Enterprise
• Cooling: continue reading ;)

I'm quite happy with this thing but not with temperatures.

 

I went through variety of coolers:

• Fractal Celsius S36 AIO
• Noctua NH-U14S TR4-SP3 Air Cooler
• EKWB EK-Velocity sTR4

Let's get Noctua out of the picture first - idle temps were around 45-50C with fan on 100% RPM during winter days. Lower the fan and you're looking at 60-70C in idle. Next Celsius S36 - I was running this thing for last four months getting idle temps around 40-50C (50C were spikes but general lower half of 40C in idle). I could live with it but summer came and now with ambient around 25-27C my idle temps with Fractal's S36 jumped up by around 10C, to low 50-ish C. Start Visual Studio or a mere browser and I'm at 60C with fans ramping up.

 

So it came time for big guns. I was like "ok, this thing costed me so much already, no cutting corners, let's go big". So I bought bag of EKWB's goodness:

• EK-Velocity sTR4 Water Block
• EK-Quantum Reflection Evolv X Distro Plate (D5 PWM  pump included)
• EK-CoolStream Classic PE 360 Radiator (it's largest I can fit in that case... barely)
• 3x EK-Vardar EVO 120ER Fans
• Clear CryoFuel, soft tubes and bunch of connectors, ports, valves, leak testers and what not.

For price alone I would buy another computer but hey, it looked so damn good on pictures and promised so much performance. I've done and checked everything: rinsed the radiator (it took like two days alone to take all the debris out of there), removed as much air bubbles as I virtually could (there still are some that I can't do anything about). Leak testing with air, leak testing with water etc.

 

Long story short: this whole EKWB thing is performing WORSE than Fractal's Celsius S36. My idle temps are around 47-55C, run anything and say hello to 60-65C. Cinebench R20 - 77-83C after 60s run. This is ridiculous! I didn't expect miracle but at lest these 5-10C less in idle and more stable under light load (like web browser or Visual Studio). Tuning pump and radiator fans doesn't do much, of course if I set down pump to lowest settings temps are all over the place, so I need to keep it around 75-100% (3500-4000RPM) just to get 5-10C worse temps than cheap AIO gave me!

 

At this point I'm almost giving up. Around $800 (good thing I can save on taxes as it would get well beyond $1000 otherwise) went down the drain and I'm close to throwing this junk to garbage bin and get back to Fractal's AIO. Not to mention it looks way worse than slick two tubes, no plate & no pump design Fractal has (also my EK branded tubes, with EK's own clear CryoFuel turned opaque, what you can see on attached photos), the production quality is mediocre (scuff on water block, visible scratches on inside portion of distro plate etc.).

 

So you guys are my last resort I think. What I could do wrong? I'm a noob if it comes to custom loops, the tubing isn't the top-notch like seen on YouTube, but I can see the flow is quite rapid. Pictures of it all attached below.

 

I know this question will come, so as for the thermal paste I used only Thermal Grizzly's Kryonaut for all setups mentioned above. With different application methods: 9 dots, even spread with "credit card", 12 dots, lines, blobs, you name it. Today I was to resocket the CPU and reapply paste and block again but run out of thermal paste again, so that has to wait till delivery early next week. I'm also thinking about ordering another bottle of coolant liquid and reflushing this whole thing (only thing that holds me out is yet another air bleeding process).

 

Again: please help me out if you can, any directions, any tips what I can do to make this EKWB stuff running at least as good as Fractal's AIO if not little better, would be welcome.

 

This is how it looks like:

 

 

 

ps. As you can see I'm trying to cool only CPU, hence the short-circuit on the GPU ports. If I would hook up GPU to this thing also, it would surely catch fire.

 

[SavageNeo]

2 hours ago, EvilVir said:

-snip-

i mean idle temperetures do not matter os lest get them out of the pic.

 

yes those are little over normal temps for custom cooling but if you are under 95c you will be just fine

Edited July 18, 2020 by TheRandomness
Please snip long quotes.

[EvilVir]

Just now, SavageNeo said:

i mean idle temperetures do not matter os lest get them out of the pic.

 

yes those are little over normal temps for custom cooling but if you are under 95c you will be just fine

Thanks, but to stay under 95c means that I have to spin up fans and pump, as this is my daily workstation it's hard to listen to all that noise for 8-10 hours each day.

 

That's why I decided to buy EKWB gear - it was to be cooler and more quiet but for now turned out to be quite opposite  

[Jurrunio]

Maybe stock voltage (both idle and load, yes they are different) are just too high? Consumer parts with single digit amount of cores do around 1.3-1.45v when power unlimited, a 24 core like yours shouldnt be allowed to go above 1.3 for thermal's sake even at idle.

[narrdarr]

Firstly nice build.

 

This are some things that come to mind for me. 

 

Idle temps are going to be the same with almost any cooler air or water. Somewhere between 40 50 is normal.

 

Voltage maybe be a bit high. Try capping voltage. start at 1.4v and work your way down toward 1.3. At some point you will your boost clocks will not boosting as high. At that point you have 

Hit the threshold for undervolting.

 

Water cooling setup:

On you distro block

I see your rad is connect 

One intake one outlet looks good

Your cpu block 

One intake one outlet looks good

You have a line that runs from one side of the distro to the other. Which is technically fine but for testing id move the cpu line there and cap the other inlet/outlet holes. Giving max flow to the cpu.

What is the extra run coming off the distro block? Just a drain / fill port.

Might have air trapped in the system. If air is trapped in the cpu block it could be causing issues. Try shaking the system while it is running and/ or squeezing the return line. 

 

TIM

You said you tried a bunch of methods

However don't be scared to more. Also don't worry spreading let the paste do the work for you. Let it fill the areas that it needs and push the rest out.

 

You can also try a different cpu block. 

 

Another thing you can do is lapped the cpu. But before I'd lap a cpu like that I'd exchange the cpu. Might have a contact issue between the die and IHS. 

 

 

 

 

[EvilVir]

Thanks for hints @narrdarr, very useful!

 

That extra line there is because this plate is meant to be CPU + GPU loop. Thing is that if nothing was connected to GPU ports then there was no flow between two halves of the plate. I'm not sure if that's the case if I connect CPU to GPU ports and just lock CPU portion with caps - I asked EKWB support about that, still waiting for them to get back to me about this. I can test myself but I would need to buy another coolant bottle and that takes 3-5 days to arrive + it means whole drain/fill/bleed operation + I don't want to wear the acrylic threads. If EKWB will confirm that it's possible to run only CPU on only GPU inlet/outlet I will surely drain the loop and do it as I really don't like how it looks now (here's one to EKWB's designers for really poor ports placement on that tank, I wanted to connect this with really short tube and two 90 angle fittings but ports are too close to each other).

 

There are two extras there - at top there is tube going above the radiator to fill port (again ports could be really done better as I had to use lower fill port and so had real problems to top-off the tank with fluid). At the bottom there's just a valve directly attached to the drain port (planned to route that down into actual drain hole in the case but it wasn't sturdy enough for me and gave problems with cables management).

 

As for TIM, yes after I removed the block today I saw that it wasn't even spread. Like top half of the block was covered with thinner layer than rest of it (I could see nickel plating through that thinner part in some places). I ordered new 11g tube of Kryonaut that should arrive at Monday. I will try and go with credit-card spread again, maybe also will waste some TIM to do a test-run to see if I can get good coverage with that method this time. Of course all this moving and removing block gave it some two or three hair-like scratches at the cold plate, so I'm hoping it will not affect performance even further.

 

Anyway for now the TIM spread is my only hope to make this thing usable. Lapping is too risky IMO, maybe I would find someone to do it for me but that's a far option. CPU exchange would be hard due to how law works in such cases in Poland, and it's always a gamble what you'd get - this unit is stable and really good performing. I read in other topics that somebody have issues running one virtual machine, I can throw three-four of VMs at this puppy, alongside with dozen of Docker containers, Visual Studio debug and AngularJS debug and I don't feel like running anything really. It's interesting though what you and @Jurrunio said about voltage - I will def. check that once I have this thing running again this Monday. I'm not doing any overclocking whatsoever, apart from PBO set on enabled in BIOS, as well as ensuring that RAM to Infinity Fabric ration is 1:1, as advised by AMD. Maybe one of these is giving me hard time, I'll double check that - thanks for the tip!

 

 

BTW: does anybody find tubes colour odd? This meant to be clear tube with clear coolant. But you can see how it went milky. I've done an experiment this morning, cut off like inch worth of tube, put it in a jar and drown in very same coolant. 10h later I can see it is getting little foggy. Coolant itself is still clear (same in loop - I can see fluid inside tank and it is clear) just the tube is opaque now. This is EKWB's soft tubing with EKWB's premixed coolant. You think I have bad batch of tubes or it's normal? Maybe some other brand would do better?

[narrdarr]

1 hour ago, EvilVir said:

Thanks for hints @narrdarr, very useful!

yea NP

 

1 hour ago, EvilVir said:

That extra line there is because this plate is meant to be CPU + GPU loop. Thing is that if nothing was connected to GPU ports then there was no flow between two halves of the plate. I'm not sure if that's the case if I connect CPU to GPU ports and just lock CPU portion with caps - I asked EKWB support about that, still waiting for them to get back to me about this. I can test myself but I would need to buy another coolant bottle and that takes 3-5 days to arrive + it means whole drain/fill/bleed operation + I don't want to wear the acrylic threads. If EKWB will confirm that it's possible to run only CPU on only GPU inlet/outlet I will surely drain the loop and do it as I really don't like how it looks now (here's one to EKWB's designers for really poor ports placement on that tank, I wanted to connect this with really short tube and two 90 angle fittings but ports are too close to each other).

 

There are two extras there - at top there is tube going above the radiator to fill port (again ports could be really done better as I had to use lower fill port and so had real problems to top-off the tank with fluid). At the bottom there's just a valve directly attached to the drain port (planned to route that down into actual drain hole in the case but it wasn't sturdy enough for me and gave problems with cables management).

 

I think this is affecting the performance the least.

However

Capping them should be fine. you can use straight distilled water from the store to get you through testing. Then add your biocide afterwards.

---

I see.

 

 

1 hour ago, EvilVir said:

As for TIM, yes after I removed the block today I saw that it wasn't even spread. Like top half of the block was covered with thinner layer than rest of it (I could see nickel plating through that thinner part in some places). I ordered new 11g tube of Kryonaut that should arrive at Monday. I will try and go with credit-card spread again, maybe also will waste some TIM to do a test-run to see if I can get good coverage with that method this time. Of course all this moving and removing block gave it some two or three hair-like scratches at the cold plate, so I'm hoping it will not affect performance even further.

If the TIM is general thinner on half then you probably have a mounting pressure issue or a block issues. Your not the first to run in this. this is one of two main issues second being voltage but ill get there. This is probably why your aio was preforming better

This you might fine interesting

https://www.xtremerigs.net/2018/08/29/amd-threadripper-cpu-block-review-round-up/all/1/

scroll down to the part about TIM spread struggles

 

Still i don't think spreading the TIM is going to help you. By not spreading you will be able to see how well it does spread between the block and IHS

 

1 hour ago, EvilVir said:

Anyway for now the TIM spread is my only hope to make this thing usable. Lapping is too risky IMO, maybe I would find someone to do it for me but that's a far option. CPU exchange would be hard due to how law works in such cases in Poland, and it's always a gamble what you'd get - this unit is stable and really good performing. I read in other topics that somebody have issues running one virtual machine, I can throw three-four of VMs at this puppy, alongside with dozen of Docker containers, Visual Studio debug and AngularJS debug and I don't feel like running anything really. It's interesting though what you and @Jurrunio said about voltage - I will def. check that once I have this thing running again this Monday. I'm not doing any overclocking whatsoever, apart from PBO set on enabled in BIOS, as well as ensuring that RAM to Infinity Fabric ration is 1:1, as advised by AMD. Maybe one of these is giving me hard time, I'll double check that - thanks for the tip!

 

Voltage seems to be your second biggest issue. Not going into to much detail. but PBO will push 1.41-1.45v even when a single thread is being used. I would try running without PBO. then lowering voltage. People are doing manually all core OC 's around 1.35v. 

Ram looks good. i wouldn't touch that.

some more info

 

It about OC TR but just some info

https://www.anandtech.com/show/15152/asus-rog-zenith-ii-extreme-review/8

PBO /OC info

https://www.gamersnexus.net/guides/3491-explaining-precision-boost-overdrive-benchmarks-auto-oc

https://www.youtube.com/watch?v=Q5WQAMSFPA4&t=1222s

 

 

1 hour ago, EvilVir said:

BTW: does anybody find tubes colour odd? This meant to be clear tube with clear coolant. But you can see how it went milky. I've done an experiment this morning, cut off like inch worth of tube, put it in a jar and drown in very same coolant. 10h later I can see it is getting little foggy. Coolant itself is still clear (same in loop - I can see fluid inside tank and it is clear) just the tube is opaque now. This is EKWB's soft tubing with EKWB's premixed coolant. You think I have bad batch of tubes or it's normal? Maybe some other brand would do better?

 

I thought you got frosted tubes. LOL

TBH im not sure. Maybe someone else here knows. I would ask EK.

 

 

So your block mounting evenly and all the way and voltage seem to be the biggies.

 

[EvilVir]

Thanks, I just opened another support ticket with EK, about these tubes. Below is the picture how tube looks after just 12h soaking in the jar with coolant (top piece, bottom is for reference and never touched coolant):

 

 

Looks like I'm up to disassembly all that loop, buying new tubes and putting I all together again. I will get new TIM tomorrow and will try to get good coverage (got to do it in four tries as that's about for what 11g tube will last lol :), will get some distilled water as well, drain the loop and swap the tubes to other ports. Thinking about ordering new set of tubes and replace all of them but probably I'll wait for EK to get back in touch (they lately take plenty of time to respond sadly) and in the mean time will do some more tests on the scraps.

 

And then PBO, I'll hate to loose it as it bumped my Cinebench score by almost 1000 once, but if the price for PBO is inability to cool this beast to some acceptable levels, then I'll live without it I think. Too bad I can't check voltages today but surely I will post them here once I will be able to run this PC again in few days.

 

Thanks again for all your input, it's really helpful and makes me still have hope that this whole effort and cash will not go to waste.

[narrdarr]

18 minutes ago, EvilVir said:

Thanks, I just opened another support ticket with EK, about these tubes. Below is the picture how tube looks after just 12h soaking in the jar with coolant (top piece, bottom is for reference and never touched coolant):

 

 

Looks like I'm up to disassembly all that loop, buying new tubes and putting I all together again. I will get new TIM tomorrow and will try to get good coverage (got to do it in four tries as that's about for what 11g tube will last lol :), will get some distilled water as well, drain the loop and swap the tubes to other ports. Thinking about ordering new set of tubes and replace all of them but probably I'll wait for EK to get back in touch (they lately take plenty of time to respond sadly) and in the mean time will do some more tests on the scraps.

 

And then PBO, I'll hate to loose it as it bumped my Cinebench score by almost 1000 once, but if the price for PBO is inability to cool this beast to some acceptable levels, then I'll live without it I think. Too bad I can't check voltages today but surely I will post them here once I will be able to run this PC again in few days.

 

Thanks again for all your input, it's really helpful and makes me still have hope that this whole effort and cash will not go to waste.

If you can improve contact of the block to the IHS i using PBO will be ok.

you can do this by increasing mounting pressure with some washers round the screws so the screw can go through but not the spring. this should work well for you.

 

 

 

[EvilVir]

Ok, so month passed. That was busy month so it took me a good while to finish building this thing. But it's done... and quite depressing. Not much improvement, still idling on around 50C with ambient 26C. But this time around I can have fans and pump at around 30-50% speed, no need to keep them on 100% which is an improvement from first attempt where they had to stay on max to keep same temps. So that's something. Still all this bulky hardware, large block, hefty radiator, more than 0.5L of coolant and it's on par with cheap AIO! This is not right at all.

 

I'm out of the ideas at this point to be honest. Any advice would be helpful.

[narrdarr]

5 hours ago, EvilVir said:

But this time around I can have fans and pump at around 30-50% speed, no need to keep them on 100% which is an improvement from first attempt where they had to stay on max to keep same temps

that sounds like a big improvement, but your pump speed seem a bit low.

 

What changes have you made?

 

[EvilVir]

Well ramping up the pump doesn't give any lower temps in idle, just in load so I keep it on 40% to reduce noise up until 70C, then it goes up. It almost looks like this CPU is totally immune to any cooling in idle and always stays in 45-55C range. AIO? Same. Custom loop? Same. Custom loop and household fan pointed at the CPU? Same. Pump, seven case, fans on 20% or 100% speed? Same.
 

I wonder if it would be the same if I'd just glue a folded tinfoil with a piece of chewing gum to it lol.

 

As for changes, well I drained the system, stripped and exchanged all tubing, resocketed the CPU, reapplied TIM, removed the inlet and outlet tubes, bought EKWB's special filling bottle and used it to slowly fill the system from top-most inlet. Replaced the GPU bridge tube with flow indicator.

 

Still I don't understand why I can't go below 40-50C in idle. Everywhere I see people are like "with custom loop you can OC this thing to 4.5GHz", they are chilling both this CPU and GPU and getting 36C in idle (DaPoet from YouTube for example). And then there I am sitting here, with very same rig and looking at 45-55C on Windows' desktop... I just don't get it

[narrdarr]

what monitoring software are you again?

[narrdarr]

also what are the load temp?

[EvilVir]

HWInfo64, MSI Creator Center, Ryzen Master, BIOS itself.
 

DaPoet, I mentioned above also used HWInfo64 in his video and it showed nice 36-39C for him. I know about problems with monitoring software causing the "observer effect" but even if I use none, I can tell that temp is in some range by noise of the fans ramping up in predefined points.

[EvilVir]

Load temp (like Cinabench 20 10 minutes test) goes to around low 80-ish range. In normal work, like compiling the code etc. it usually spikes to around 68C and then drops down after done.

[narrdarr]

temps are in check, but i agree it should be better. The more i stare at this the more i see and what i see is the distrop plate is probably your main issue and secondly i still that the block doesn't contact well compared  to others.  This distro plate distributes water equally based on volume between the rad, cpu, gpu (if included). Its not a one way flow system.

 

As a test i would reconfigure the distro plate to act more as a res.

Close all inlet/outlets except for 2 (one inlet one outlet)

run from distro------rad----cpu----distro or distro-----cpu----rad----distro.

[EvilVir]

Hm that's a good idea, will give it a shot in coming days. I need to think how to pull this off though because this distro plate has required inlet/outlet pairs to be used. For example I don't have GPU in the loop so I had to do a bridge (eventually I've put flow indicator there) so water can pass from one side to another. Right now my config is distro -> CPU -> distro -> flow indicator -> distro -> rad -> distro. Heck, it might be easier to just buy some cheap reservoir, even used, and test on that. I'll see what can be done here.

 

I'm affraid that the block isn't even or something else is wrong with it. These temps should be at least 5-10C lower IMHO. This or the CPU or Mobo is malfunctioning and giving too much current to the CPU (in BIOS I see 1.44V but then in Windows it depends on the tool I use to meassure - Ryzen Master shows 1.2-1.3 in idle mostly, but HWInfo shows 1.4-1.45).

[narrdarr]

 

1 hour ago, EvilVir said:

Hm that's a good idea, will give it a shot in coming days. I need to think how to pull this off though because this distro plate has required inlet/outlet pairs to be used. For example I don't have GPU in the loop so I had to do a bridge (eventually I've put flow indicator there) so water can pass from one side to another. Right now my config is distro -> CPU -> distro -> flow indicator -> distro -> rad -> distro. Heck, it might be easier to just buy some cheap reservoir, even used, and test on that. I'll see what can be done here.

 

I'm affraid that the block isn't even or something else is wrong with it. These temps should be at least 5-10C lower IMHO. This or the CPU or Mobo is malfunctioning and giving too much current to the CPU (in BIOS I see 1.44V but then in Windows it depends on the tool I use to meassure - Ryzen Master shows 1.2-1.3 in idle mostly, but HWInfo shows 1.4-1.45).

Hold on

3 hours ago, narrdarr said:

This distro plate distributes water equally based on volume between the rad, cpu, gpu (if included). Its not a one way flow system.

This statement is wrong. When i was looking at i thought the distro plate was in a parallel config, but is not. It is indeed in series and does flow one way. 

 

However if you would like to try to bypass most of the distro plate. you would go (based on the EK diagram)

 

"gpu inlet" run to cpu block-----rad------"rad outlet"

 

This will bypass everything else on the plate. you shouldn't even need to cap the others.

 

 

my OG concern was the block. after seeing my mistake on distro plate layout. im back to that being the primary concern.

 

 

[Guest]

@EvilVir

 

Hi! 

I want to help you cool your cpu a little better. 

 

The CPU and VRMs are close to each other.

Both transfer BTU through the tracings.

 

Put a fan like 80mm or a couple of 60mm fans right over the VRMs. This should really help thermals.

 

As you see, there will be some counter productivity with the rad fan pulling, but the point is that you cool stuff by blowing on it. No different than blowing on your mashed potatoes before you seer the top of your neck hole.

 

No, the rad fans sucking air from the general area doesnt really help cool the components.

 

Back in the day, I bought this Asus board. It came in the box with a blower style fan that clipped to the VRM heat sink. It was strongly recommended to use a fan because the stock cooler cooler wasn't there blowing down on the board from use of a water block.

 

Anyhow, enough of all that. This picture should describe my above ramblings.

 

Good Luck!

 

 

[EvilVir]

@ShrimpBrime thanks man, I appriciate that and will try. Yesterday I pointed a household fan, that I use to cool myself, on the motherboard and no change at all. I will try to point some smaller but directly on the VRMs.

 

This thing is really weird, just like it doesn't care about ambient and whatever you throw at it. All the temps I gave you guys before were taken when ambient temp was about 24-26C, during night, with window open. Right now I'm sitting here, window closed, it's bright sunny day, Ambient is about 28-29C now in the room and guess what - no change. Nothing, still going between 45-55C, with occasional blimps to as high as 62C and as low as 43C. Insane. I start to think that one of the thermal probes inside CPU is broken lol It's just not physically possible that whatever you throw at this thing it behaves exactly the same.

 

I will try and cool these VRMs though as well as RAM sticks (as there are 8 of them, closely packed together in all available slots, generating some amount of heat on their own).

[Guest]

The idle temp spikes are a direct result of XFR single core boost clocks. Generally with a higher v-core between 1.4 and 1.5v which is very normal.

 

And also, we can touch base some conversation about temperature gradients. 

 

Further more, using HWInfo64, there's one particular temp you want to watch which is considered the actual temp. To get a better understanding of that, read this link.

 

Answer is the 2nd post by the developer.

https://www.hwinfo.com/forum/threads/cpu-temp-sensors-explanation.5597/

 

You have a monster cpu. Worry mostly if it throttles IMO unless your manually overclocking.

 

According to the white sheets, 70c is the cpu high temp alert to the mainboard. If running a stock cooler, the cpu fan would be 100% at and after this temp. The alert is only canceled when core temp reaches 69c or less. 

Throttle temp is 95c. Stay under this on stock system, that's just fine.

 

Touch on TDP.

This is the max pstate number. So 105w is not a reflection of boost clocks, that cpu is easily 50w hotter under a full load if not more.

 

Edit: I need to make a correction as I used a poor TDP number but was just an example.

The 3970x is a 280w rated chip. It's over 300w when under full load.

 

So 280w (pstate) is 955 btu/hr. This is just to clarify approximately how much heat you need to move.

[EvilVir]

@ShrimpBrime I read that post by HWInfo developer yesterday and yes, number I'm watching and giving above is TCtrl/TDie (I have both on same row, no separation). Right now, after couple of hours idling, I have only few background programs (like Enpass, Docker Desktop etc.) + Clockify, Skype, Windows Mail and Edge Chromium opened and I'm seeing it jumping around 54-62C. This is way too hot IMHO, like 10C too hot to be comfortable. Also spinning pump or fans up doesn't make any perceived difference here (right now all are set on about 50%, with going to 100% on the 70C mark).

 

I just can't understand how the EKWB's block that covers whole CPU die is performing same as standard off-shelf AIO? Is that even possible if everything is connected properly? Because that's the first point which I think should be rulled out - is this custom loop on par with AIO and then indeed I should admit that all I got for these c.a. $1500 I've already put into it is just questionable look (frankly I liked AIO's look better) and focus on finding problem elsewhere (like VRMs, airflow in case etc.) or this is not possible and I should focus on finding problem with the loop itself first (maybe water block doesn't make good enough contact still, is uneven or something else is wrong? should I RMA it? go back to AIO?). Any idea how I can check if the loop is doing its best and there is no room for improvement there?

 

I got impression that custom loop with dedcated waterblock will always work at least a bit better than AIO with smaller coldplate and with less fins. Everyone - LTT, GamerNexus, JayZTwoCents, just to name few, is like "oh yeah, this is one big waterblock, it will chill out your CPU, two GPUs and you can even overclock all of your components" or like "this Asetek based AIO is good enough but if you really want to run this thing cool then you need to go with custom loop". Even skipping the fact that all the tests they do are on open bench in air conditioned rooms, there MUST be some difference right? I mean this EKWB's water block alone weights probably more than Fractal's AIO as a whole (with rad, pump, tubes and fluid). And also let's not forget that AIO has pump, that is generating heat on its own, inside the water block! IDK, I'm again on the road this weekend but I think on Monday I will take out the water block and see if I can improve contact somehow, how the paste is spread. Last time it wasn't even, like lower half of the block was fully covered with paste and upper half was clean as it would barely touched the TIM at all (or otherway around - like the "clean" half was making so good contact, that it squeeze all the TIM into lower half where there was some air pocket - I don't know how to read that, maybe you do?).

 

What I'm hoping here is that there is some flaw in the loop, that once fixed will bring me down some nice 5-10C and then on top of that I can work towards cooling down components around the CPU to gain even more. If I can achive stable 5-10C in same scenarios where I'm right now looking at 55-65C range (so light, desktop type load) then I would call it a success and go drunk AF out of happiness

[EvilVir]

ps. On the side note, this is not the only computer I have. In fact this is one of four. Two others are on Intel CPUs (7700T and 2600K) and another one is some Athlon 200GE. Athlon is not yet completed (it will do as NAS) but 7700T and 2600K are running right now, they have these small stock coolers, are closed in small cases with no additional fans at all (7700T in ITX and 2600K in MiniTower) and I know these are much lower powered, 4 cores processors. But still they manage to sit around 43-48C right now.

 

IDK if this information is helpful, maybe would give you some perspective into operational environment I'm in right now. I remember that this 2600K thing, when it was acting as my main desktop, it had some big thermaltake air cooler installed and had like 10C less temps than it has now.

[Guest]

Haha 200ge (which I have and also a 220ge both I've delided) and then compare this chip to the monster thread ripper. 

 

Anyhoot, I wonder if the HeatKiller IV Pro would be a better water block for this chip.

Full copper no nickel plating. Nickel by the way is super poor for heat transfer. 

 

However I dont think your temps aren't that bad really. Could be better, sure. Could be worse too.

 

I'd mention to lap the cpu, but that's an expensive chip. Wouldn't do it personally if couldn't be helped.

 

There's that one thing we cant check on. The solder under the IHS plate. I wonder if an RMA and try a different chip, perhaps the thermals would be very different.