[SOLVED] What's the difference/what am I missing?

Phaaze88

Champion
Ambassador
7820X, delidded and under a Direct Die Frame
View: https://imgur.com/sq3RtMN

NH-D15S and Alphacool Eisbaer Aurora 360 are what I have on hand for the cooling.

1080Ti
View: https://imgur.com/pYnFXLr

Kraken G12 + Fractal Celsius S36

Both are direct die cooled now, but is it still necessary to use liquid metal on the cpu at this stage(no more IHS)?
Paste(NT-H2) has worked just fine for the gpu before and after changing the stock cooler. Thermals have been great for it as well; never sees over 40C under loads.

@rubix_1011 , @CompuTronix
 

CompuTronix

Intel Master
Moderator
Update:
Welp, after running an 8hr long Realbench last night, the delta between the hottest and coolest cores was 12C...

Yep. Paste does not work effectively on direct die cpus like it does with direct die gpus...
During my delidding / direct-die testing, I found that pump-out with conventional TIM could be seen in as little as a matter of several hours, to a few days, to a few weeks, (depending on the brand), where Core temperatures gradually begin to increase. Further, it's important to emphasize that when seating the IHS or when seating the cooler on direct-die, the more precisely the IHS or cooler is geometrically seated in place, the less likely that the liquid metal will be "smeared", and thus the better your results. So moving / sliding/ twisting or otherwise adjusting the position of the IHS or the cooler on direct-die should be avoided in favor of exact positioning on the initial attempt. This is where attention to detail is critical, so that an otherwise perfect delidding / direct-die procedure isn't compromised during the final steps.

Even so, don't expect identical results twice, as slight differences between seatings are normal. Keep in mind that Intel's specification for DTS (Digital Thermal Sensor) accuracy is +/- 5°C, (which in my opinion is sloppy), so during an all Core steady-state 100% TDP workload (P95, Small FFTs, no AVX), the deviation between the hottest and coolest Cores shouldn't exceed 10°C ... IF each sensor exhibits linear behavior. However, as the inner Cores tend to run hotter due to being insulated by the outer Cores, as well as variables in both internal and external TIM application, your results can be skewed from the specification. Nonetheless, 12°C is a bit much. Fortunately, you'll find that when you repeat the procedure with Conductonaut, you should see lower as well as more consistent Core temperature values.

CT :sol:
 

rubix_1011

Contributing Writer
Moderator
Direct die, you can use any quality paste, does not have to be liquid metal.

On my GPU blocks, I use a bit of plastic wrap on my finger, drop a dot of TIM, and dab it around to fully cover. GPU blocks usually provide a lot oc clamping force due to several screw mounts. Just make sure to provide full coverage and enough to provide good contact like you normally would. Just be careful if you decide to use anything conductive, but I've had great results with most quality TIMs, including MX-4.

Thermal Grizzly disintegrated under my GPU block after several months of Folding At Home 24/7, so I cleaned and went with MX-4 few months ago.

For CPU, I would take the same approach for direct die cooling, and should be good for clamping force due to backplate mount design.

BTW, I'm going to be starting a revival of the thermal compound testing/roundup in the next few days - it should be out (maybe) in December. Will take a bit to cover 10 compounds, 2 different coolers (big air/big AIO) and different clamp forces of each (using inch/pounds converted also to newton meters).
 
Reactions: Phaaze88

Phaaze88

Champion
Ambassador
Well, that's one way to spread paste... might be a little less wasteful; I've found that NT-H2 doesn't spread as easily as the old one did.

Thermal Grizzly disintegrated under my GPU block after several months of Folding At Home 24/7, so I cleaned and went with MX-4 few months ago.
The heck, is there some kind of QC issue with this stuff?
Well, it was F@H, but still... that seems way too short.

BTW, I'm going to be starting a revival of the thermal compound testing/roundup in the next few days - it should be out (maybe) in December. Will take a bit to cover 10 compounds, 2 different coolers (big air/big AIO) and different clamp forces of each (using inch/pounds converted also to newton meters).
Neat! Looking forward to it.
 

rubix_1011

Contributing Writer
Moderator
I was kind of disappointed with TG (I think I used Kryonaut), primarily with longevity. Initially, it worked great and thermals were excellent. However, I noticed my 2080 just started running hot and when I rebuilt my loop I pulled the GPU block and to my disbelief, the TG was nearly all 'gone'....like....vanished. With CPUs, I use the dots (lentil size) but for GPUs, I usually use the spread method.

I cleaned the die, re-applied with a glop of MX-4 and it has been running well for a few months now.

Also pulled the jet plate out of the EK CPU block to improve flow rate and ran Y-fittings to run parallel coolant runs to CPU+GPU at same time.

Flow rates are very, very high now.
 

Phaaze88

Champion
Ambassador
Kryonaut, huh? :unsure:

Looking around for either of these seems to suggest there's some QC issues.
Kryonaut: has fakes out there. Just WUT.
Kryonaut: arriving old and dried out.
Kryonaut: cures and is abrasive when it's advertised not to do either - related to the fakes, I believe.
Kryonaut: missing applicator.
Kryonaut: short shelf life.
Kryonaut: arriving with broken or damaged tubes.
Conductonaut: opening the TG Conductonaut pouch, only to find a tube of Kryonaut instead... can't return it either.
Conductonaut: corrodes, even with copper and nickel - though I'm going to assume those people don't know the difference between corrosion and simple staining... unless, there's also fakes of this out there, geez.
Conductonaut: missing applicator/syringe.
Conductonaut: damaged tube; skeet-skeet erry-where.
Conductonaut: short shelf life, which is strange to me, because that's what Silicon Lottery uses, and I ran this for over a year with no issues before finally opening the lid myself... and it was still there.
 

faalin

Judicious
That's odd I've been running TG Kryonaut paste on my 8700k since june 2019. Delidded with liquid metal under a full copper IHS and TG Kryonaut between IHS and Cryorig R1 ultimate cooler. Running 5.1Ghz for over a year without any issues, computer stays on 24/7 with about a handfull of shutdowns over the year.
 

CompuTronix

Intel Master
Moderator
MX-4 is my CPU / GPU TIM of choice for its viscosity, spreadability and longevity.

However, for CPU delidding or direct die, conventional paste will fail in a relatively brief period of time. Since CPUs typically have such a small surface area compared to GPUs, thermal cycling causes “pump-out”, which expels conventional TIM from between the Die and IHS, whereas liquid metal is very resistant to pump-out.

Several years ago I tested a wide variety of conventional TIM in both delidded and direct die configurations over the course of many months. I found that pump-out is indeed a serious problem. It's also interesting to note that although Intel's paste is formulated to resist pump-out, it still degrades over time, as the thermal bond with the Die slowly deteriorates. This is evidenced during delidding older processors, where very close visual examination of the Die surface reveals the bonding pattern, or lack thereof.

Pump-out has been discussed at length in other forums such as Overclockers.net. I've never encountered any longevity issues with liquid metal products, whether from Thermal Grizzly or CoolLaboratory. I've also queried the folks over at Silicon Lottery. Although they changed from CL to TG a few years ago due to Conductanaut's higher thermal conductivity, they claim that their CPUs aren't returned for thermal problems due to liquid metal degradation.

CT :sol:
 
Reactions: Phaaze88

Phaaze88

Champion
Ambassador

Darkbreeze

Retired Mod
MX-4 is my CPU / GPU TIM of choice for its viscosity, spreadability and longevity.

However, for CPU delidding or direct die, conventional paste will fail in a relatively brief period of time. Since CPUs typically have such a small surface area compared to GPUs, thermal cycling causes “pump-out”, which expels conventional TIM from between the Die and IHS, whereas liquid metal is very resistant to pump-out.

Several years ago I tested a wide variety of conventional TIM in both delidded and direct die configurations over the course of many months. I found that pump-out is indeed a serious problem. It's also interesting to note that although Intel's paste is formulated to resist pump-out, it still degrades over time, as the thermal bond with the Die slowly deteriorates. This is evidenced during delidding older processors, where very close visual examination of the Die surface reveals the bonding pattern, or lack thereof.

Pump-out has been discussed at length in other forums such as Overclockers.net. I've never encountered any longevity issues with liquid metal products, whether from Thermal Grizzly or CoolLaboratory. I've also queried the folks over at Silicon Lottery. Although they changed from CL to TG a few years ago due to Conductanaut's higher thermal conductivity, they claim that their CPUs aren't returned for thermal problems due to liquid metal degradation.

CT :sol:
I knew there was a reason you had recommended only using liquid metal for direct to die cooling, which is why I mentioned it in the members systems gallery thread.

To be clear, regarding Thermal Grizzly, while I have not used it for any direct to die applications, I can say that for normal applications I have never experienced what I've seen a few people here complain about with it being hard, difficult to spread/poor viscosity, crumbly, doesn't last, or any other issues. I think that is likely ALL related to getting FAKE product. I've used maybe twenty tubes of Kryonaut and haven't had any of those issues with any of them. I have however purchased them all from the same source, here:

https://www.outletpc.com/jf5378.html

I think buying it through Amazon or Newegg is problematic because many if not most of their Kryonaut listings have generally been from third party sellers, but fulfilled by the retailer, which obviously opens things up to the potential for considerations of shipping very old or fake product in some cases. Much less likely to happen when sold AND fulfilled by the retailer, and when the retailer is an official vendor for that manufacturer. It's the only reason I can think of.
 

Phaaze88

Champion
Ambassador
Update:
Welp, after running an 8hr long Realbench last night, the delta between the hottest and coolest cores was 12C...

Yep. Paste does not work effectively on direct die cpus like it does with direct die gpus...
 

Darkbreeze

Retired Mod
Saying I told you so is rude, so instead I'll only say ¯\(ツ)/¯ LOL.

Obviously, first hand experience is always preferred though, and now you have that. Helps to confirm for me as well when making recommendations in similar circumstances in the future too.
 
Last edited:

Darkbreeze

Retired Mod
Not really. There is no "deserved" here. I agree with the idea that getting first hand evidence, proof, is always desirable even when somebody I know is knowledgeable tells me a thing because I know that sometimes you can get different results with very similar circumstances and variables. The only time it ever gets my goat is when the one involved HAS that proof, for themselves, and it matches what others have said, and they STILL want to dispute it by pretending like they didn't get that result. That's when it makes a mofo mad. LOL.

In your case, it's a good thing, because now you can tell others who ask the same question, look, we have this proof and recommendations from these other sources PLUS I have seen this myself.
 
Reactions: Phaaze88

CompuTronix

Intel Master
Moderator
Update:
Welp, after running an 8hr long Realbench last night, the delta between the hottest and coolest cores was 12C...

Yep. Paste does not work effectively on direct die cpus like it does with direct die gpus...
During my delidding / direct-die testing, I found that pump-out with conventional TIM could be seen in as little as a matter of several hours, to a few days, to a few weeks, (depending on the brand), where Core temperatures gradually begin to increase. Further, it's important to emphasize that when seating the IHS or when seating the cooler on direct-die, the more precisely the IHS or cooler is geometrically seated in place, the less likely that the liquid metal will be "smeared", and thus the better your results. So moving / sliding/ twisting or otherwise adjusting the position of the IHS or the cooler on direct-die should be avoided in favor of exact positioning on the initial attempt. This is where attention to detail is critical, so that an otherwise perfect delidding / direct-die procedure isn't compromised during the final steps.

Even so, don't expect identical results twice, as slight differences between seatings are normal. Keep in mind that Intel's specification for DTS (Digital Thermal Sensor) accuracy is +/- 5°C, (which in my opinion is sloppy), so during an all Core steady-state 100% TDP workload (P95, Small FFTs, no AVX), the deviation between the hottest and coolest Cores shouldn't exceed 10°C ... IF each sensor exhibits linear behavior. However, as the inner Cores tend to run hotter due to being insulated by the outer Cores, as well as variables in both internal and external TIM application, your results can be skewed from the specification. Nonetheless, 12°C is a bit much. Fortunately, you'll find that when you repeat the procedure with Conductonaut, you should see lower as well as more consistent Core temperature values.

CT :sol:
 

Phaaze88

Champion
Ambassador
Well, the Conductonaut arrived, and it was legit - a code was provided in the packaging for verification on TG's site.
I got to work...
Turned it on, and shortly after, the fans all roared up.
Aww, crap... Sure enough, 78C in bios.
Was it because I didn't apply a layer on the cooler cold plate?
I turned it off and got about to applying LM to the cold plate.

Put it back together, and... Same result.
Maybe I spread the LM too thin?
I'm posting from my phone and not the PC, BTW.




Update: Ok, I appear to have fixed that.
It looks like I didn't apply enough LM? Like, it was too dry or something? IDK. It's idling around the 30s on all cores currently.
Even after all the videos and threads warning about putting too much, but unlike paste, too little also doesn't seem to work.

I will be testing thermals later tonight.
 
Last edited:
Reactions: CompuTronix

Phaaze88

Champion
Ambassador
Update:
4.4ghz
View: https://imgur.com/jasmYal


4.5ghz
View: https://imgur.com/hKtWleU


4.6ghz(Silicon Lottery bin)
View: https://imgur.com/IQbw3n3

I should be able to work with this and get it a little lower. The 1.237v from SL can't be the lowest I can get this; there's been several bios updates(performance) since then.


Well, that's that. My next task is to work on trying to skim that Vcore some more.
I reckon this thread was solved in regards to paste pump out.
I'm going to continue using paste for the gpu though; it's doing ~just fine.

Thanks everyone!
 

CompuTronix

Intel Master
Moderator
Nice job, Phaaze!

It looks like you have a consistent deviation between the highest and lowest Cores of only about 5°C, which is a huge improvement from the 12°C you had before, as well as a healthy decrease in overall Core temperatures. Well done!

CT :sol:
 
Reactions: Phaaze88

ASK THE COMMUNITY

TRENDING THREADS