Long-term result of CLU (liquid metal TIM)

Yuriman

Diamond Member
Jun 25, 2004
5,530
141
106
I've been running with CLU between my bare-die Ivy's core and my copper waterblock for close to 9 months now (same mount). In the past month, I found I was getting instability at voltage/frequency combinations that used to be stable despite my reported temperatures being as good as ever (under 75c IBT @ 4.6GHz) and I feared that the CLU had infiltrated the die perhaps, or that my chip was experiencing degradation.

Yesterday I pulled everything apart and found that the CLU had somehow clumped and formed round, lumpy, crystal-looking structures between the die and the waterblock. I'm unsure as to how this could have happened as I have pretty high mounting pressure, and the CLU should have been liquid at normal operating temperatures. There were solid CLU lumps stuck on both the die and the bottom of the block.

Removal of the CLU was difficult - the steel-wool-like scrubby pad that was included with the CLU was effective at getting it off of the die (though it left visible microscratches on the surface), but the waterblock had to be lapped to get the surface flat again, as not even a razor blade was able to get the CLU off effectively. Even after lapping there is still a silver stain on the bottom of the block.

I switched over to Noctua NT-H1 which I use on everything else and reassembled, and found that my previously stable overclock is now stable again.

My best guess is that perhaps when the CLU hardened/crystallized, it left gaps in places between the die and the waterblock, causing hot spots and instability despite the temperature sensor reporting that all was well.
 
Last edited:

wasabiman123

Member
May 28, 2013
132
1
81
Man, if only Intel kept using good ol' solder enthusiasts wouldn't have to jump through hoops doing this. I read it might be because the pressure from soldering could damage these smaller and smaller chips, but the tradeoff in TIM weaknesses is frustrating...
 

Homeles

Platinum Member
Dec 9, 2011
2,580
0
0
Man, if only Intel kept using good ol' solder enthusiasts wouldn't have to jump through hoops doing this. I read it might be because the pressure from soldering could damage these smaller and smaller chips, but the tradeoff in TIM weaknesses is frustrating...
Enthusiasts would still be interested in using it on the IHS.

OP, did you not take photos? That would have been really interesting to see.
 

Yuriman

Diamond Member
Jun 25, 2004
5,530
141
106
You can't really see the "crystals" in these pics because I had already distorted/destroyed them before taking these pics.

After initially trying to rub the CLU off with a paper towel:

CFwt7aS.jpg


After vigorous scrubbing with an abrasive:

zdPuXLX.jpg


The stuff you see on the die here could actually be felt with a fingernail:

xGfwhI3.jpg



After lapping (and taking off a non-negligible amount of copper) there was still a stain, suggesting that the CLU penetrated and permeated through the copper.
 
Last edited:

Yuriman

Diamond Member
Jun 25, 2004
5,530
141
106
Is that an Apogee?

Yeah, GT I believe. I have an 1155 mounting bracket but my board has holes for both 775 and 1155 and I like the 775 mounting bracket more. I've been using the same loop, relatively unchanged, since early 2007.
 

GAO

Member
Dec 10, 2009
96
1
71
You left some PII (personally identifiable information) in that first photo! THe FBI might be happy. :D
 

BonzaiDuck

Lifer
Jun 30, 2004
15,699
1,448
126
Galvanic corrosion from the dissimilar metals potentially?

Perhaps it was to be expected. Somewhere I thought I'd read that the CLU more or less amalgamates with other metals -- specifically the copper.

I can't remember how "second-best" experiments by IDC turned out -- with the diamond paste. I remember suggesting that it wouldn't just ooze out between the IHS and the die.
 

GreenChile

Member
Sep 4, 2007
190
0
0
Perhaps it was to be expected. Somewhere I thought I'd read that the CLU more or less amalgamates with other metals -- specifically the copper.

I can't remember how "second-best" experiments by IDC turned out -- with the diamond paste. I remember suggesting that it wouldn't just ooze out between the IHS and the die.
Diamond paste on bare die is a horrible idea. The diamonds will gouge right through the protective layer on the backside of the chip and open it up to copper migration from the water block. Copper diffuses very rapidly through pure silicon and will kill the die.
 

BonzaiDuck

Lifer
Jun 30, 2004
15,699
1,448
126
Diamond paste on bare die is a horrible idea. The diamonds will gouge right through the protective layer on the backside of the chip and open it up to copper migration from the water block. Copper diffuses very rapidly through pure silicon and will kill the die.

I still don't see how that would happen if the paste isn't used on the die like rubbing compound on your car's paint-job. If the particles don't move around, how would that gouge the protective layer? How would the copper move across the layer of particulate diamond?

I'm sure IDontCare had done tests with this. But -- hey -- with the CLU problem, maybe we could expect more of the same.

It still goes back to the initial conclusion: Intel's choices for IB and Haswell have really put a damper on "enthusiast enthusiasm."

UPDATE: Here's the link to IDC's ICD test:

http://forums.anandtech.com/showpost.php?p=34293887&postcount=50

He notes that there is some slight abrasion of the die. The cooling results are decently good, and as promised by Citarella's company, there isn't any "pumping out."

But there's no long-term conclusion about this.
 
Last edited:

GreenChile

Member
Sep 4, 2007
190
0
0
I still don't see how that would happen if the paste isn't used on the die like rubbing compound on your car's paint-job. If the particles don't move around, how would that gouge the protective layer? How would the copper move across the layer of particulate diamond?
When you seat the heat sink into place you are spreading the diamond particles across the die and smashing them into the film. This film is very thin compared to the size of diamonds and is easily scratched. You are opening thousands of doors for copper to diffuse through. Copper migrates very easily through many different materials so it is not a big stretch for it to migrate through the paste and into the silicon.

IDC's 3770K died shortly after his bare die testing experiments which included IC Diamond. There is no way of knowing if it was killed from electrostatic discharge or copper contamination but the circumstances and timing are suspicious.
 

BonzaiDuck

Lifer
Jun 30, 2004
15,699
1,448
126
When you seat the heat sink into place you are spreading the diamond particles across the die and smashing them into the film. This film is very thin compared to the size of diamonds and is easily scratched. You are opening thousands of doors for copper to diffuse through. Copper migrates very easily through many different materials so it is not a big stretch for it to migrate through the paste and into the silicon.

IDC's 3770K died shortly after his bare die testing experiments which included IC Diamond. There is no way of knowing if it was killed from electrostatic discharge or copper contamination but the circumstances and timing are suspicious.

Thanks for the objective reporting. So what now? What good is de-lidding if you can't significantly improve the heat transfer to IHS and heatsink?

Did I read somewhere here that Haswell-E would resolve this problem of IB and Haswell CPU manufacture? Because otherwise -- look at the types of threads in this forum now. Definitely not the enthusiasm of over-clocking results for Sandy Bridge.
 

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
NT-H1 is a very nice material, not the very best in terms of thermal performance, but as far as easy to use, trouble-free, long lasting TIMs go, I believe it to be among the very best.
 

GreenChile

Member
Sep 4, 2007
190
0
0
Thanks for the objective reporting. So what now? What good is de-lidding if you can't significantly improve the heat transfer to IHS and heatsink?

Did I read somewhere here that Haswell-E would resolve this problem of IB and Haswell CPU manufacture? Because otherwise -- look at the types of threads in this forum now. Definitely not the enthusiasm of over-clocking results for Sandy Bridge.
I use CLU between the die and IHS and find it to be very effective. No degredation of performance so far and it's been several months.

I could be mistaken but I heard all the E chips are still using solder. I also heard a rumor that Haswell refresh will go back to solder but don't know if that comes from a reputable source. It would be nice though.
 

Ferzerp

Diamond Member
Oct 12, 1999
6,438
107
106
NT-H1 is a very nice material, not the very best in terms of thermal performance, but as far as easy to use, trouble-free, long lasting TIMs go, I believe it to be among the very best.

NT-H1 lasts about 2-3 months between the die and IHS. No, I don't know why, but it ends up clumping up and suddenly temperatures are a lot higher.
 

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
NT-H1 lasts about 2-3 months between the die and IHS. No, I don't know why, but it ends up clumping up and suddenly temperatures are a lot higher.
My experience has been different so far, but I will keep an eye on my existing installations.

Sure would like to find a "set it and forget it" TIM that performs well.
 

BonzaiDuck

Lifer
Jun 30, 2004
15,699
1,448
126
My experience has been different so far, but I will keep an eye on my existing installations.

Sure would like to find a "set it and forget it" TIM that performs well.

Well, that was the Holy Grail of the long search IDontCare started maybe 18 months ago. So far, I've seen one person (here, starting this thread) with misgivings about CLU. Nobody else that I know tried IC Diamond, except for IDC. And I'm probably a "bad" person for waiting to see what happened to others -- although IDC had undertaken the experiments on his own.

For all we know, with so many experiments, he may have simply mishandled the I7-3770K. But we just don't know. All the "non-Anandtech" sites which offer de-lidding advice steer toward CLU.

And frankly, this Sandy Bridge I built in 2011 may have a very long practical life. I'm not going to sell the components if I build something that has any risk of short-term loss. So I'm not in any great hurry starting this year to buy Haswell and Z87 components. Still -- I'm building the component list and reading reviews.
 

Yuriman

Diamond Member
Jun 25, 2004
5,530
141
106
Thanks for the objective reporting. So what now? What good is de-lidding if you can't significantly improve the heat transfer to IHS and heatsink?

Did I read somewhere here that Haswell-E would resolve this problem of IB and Haswell CPU manufacture? Because otherwise -- look at the types of threads in this forum now. Definitely not the enthusiasm of over-clocking results for Sandy Bridge.

My temperatures are still significantly reduced. I'm able to run at 4.8GHz / 1.37v and stay under 100c IBT loaded. It'd be a cold day in hell when you can do that without delidding.