Bare-die testing: A delidded 3770k, an H100, and 9 different TIMs

[crashtech]

^ IDC showed that excellent contact is had between die and IHS once the adhesive at the perimeter of the IHS is removed. The photo documentation of this is somewhere in this thread, if I remember correctly, there is even a video of the IHS spinning around on the die, graphically demonstrating contact only with the die and not with the PCB.

 

[Idontcare]

One of the most interesting things re: delidding is what I read not too long ago:

That one reason for high temps might not be so much the TIM or how it is applied to the die, but rather how the IHS is glued (back) on. It's all about the gap between the IHS and the die. The thickness of the glue can prevent that the IHS touches the die evenly as it should.

This explains how some people did otherwise perfect CLU jobs but discovered afterwards it didn't work, exactly how you described where you did a perfect "flat" CLU job but needed to apply LOTS more for reasonable temps. (Because otherwise there was poor contact between die and IHS)

This would also mean that sanding down the IHS itself (at the base) would be interesting to achieve a nice, flat CLU application that actually touches the IHS. (As opposed to having to use TIM as a "filler")

Hi flexy! I won't claim to be the source of whatever article you read long ago, but there is a small chance that you are recalling a thread and post I wrote some 2+ years ago in the following thread:

Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

The specific post of relevance here being: Proof that the benefit from Delidding is entirely due to reducing the CPU-to-IHS gap

(personal note - can't believe that was over 2 years ago! Dang how the time flies!)

 

[DrMrLordX]

I have not done conclusive testing of any kind to determine exactly what kind of contact there was between the die and IHS before my delid, nor have I done any to test the same thing post-delid (bear in mind that I left the epoxy on the PCB to make it easier to reseat the IHS before mounting the HSF. In fact, I've pretty-much documented the whole process).

From what I saw of the die shortly after I removed the IHS, it looks like there was a high-pressure mount, but the TIM distribution looked spotty and uneven. It is entirely possible that the underside of the IHS is simply not all that flat. When I did a tear-down to see what was wrong after my first faulty application of CLU between the die and IHS, the contact pattern of the TIM showed that there was only a tiny bit of contact between the die and IHS in the dead center of the die, hence the awful temps.

 

[flexy]

Hi flexy! I won't claim to be the source of whatever article you read long ago, but there is a small chance that you are recalling a thread and post I wrote some 2+ years ago in the following thread:

Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

The specific post of relevance here being: Proof that the benefit from Delidding is entirely due to reducing the CPU-to-IHS gap

(personal note - can't believe that was over 2 years ago! Dang how the time flies!)

Wherever I read that, I am sure they were referring to exactly THAT fantastic article and the findings of yours. Excellent!

Ok I did some more reading now.

But I am unsure about a good choice of TIM to use *on die* now for various reasons:

* You say ICD would be your choice, but then I read negative things how it scratches the die...
* NT-H1 suffers from pump-out, so it's not that great
* MX-4 seems to be a catastrophe (for whatever bizarre, unknown reasons) for on-die
* CLU would be my choice, but it's electrical conductive and probably a mess to work with.

Sort-of unsure what to do now. Right now I only have MX-4 here, but amazingly I also read that some people did a simple delid, put a MX-4 (!!!) on the die, put cap back on (no glue)..and saw a MASSIVE drop in temps.

Really confused now how to proceed...

 

[ehume]

Has anybody thought to use 'articulating paper' between CPU and IHS?

Wikipedia: "Articulating paper is a diagnostic tool used in dentistry to highlight occlusal contacts and the distribution of occlusal forces. That is, it marks those points on the teeth where the teeth contact during biting and grinding.[1] Articulating paper is made of a thin, non-adhesive paper strip covered in fluorescent ink or dye-containing wax.[2] A strip of articulating paper is placed between the teeth while the desired mandibular movements are performed."

We used to all experience this when we went to the dentist, and he (it was always a "he" in those days) would have us grind our teeth on this before he ground the amalgam (remember that?) filling.

 

[Idontcare]

Has anybody thought to use 'articulating paper' between CPU and IHS?

Wikipedia: "Articulating paper is a diagnostic tool used in dentistry to highlight occlusal contacts and the distribution of occlusal forces. That is, it marks those points on the teeth where the teeth contact during biting and grinding.[1] Articulating paper is made of a thin, non-adhesive paper strip covered in fluorescent ink or dye-containing wax.[2] A strip of articulating paper is placed between the teeth while the desired mandibular movements are performed."

We used to all experience this when we went to the dentist, and he (it was always a "he" in those days) would have us grind our teeth on this before he ground the amalgam (remember that?) filling.

I don't remember which forum it was exactly, but I have seen posts and images from people using articulating paper in exactly the manner you are suggesting. It basically showed that once you remove the black adhesive the IHS makes 100% contact across the entire surface of the die.

Meaning the underside of the IHS itself does not need to be lapped or polished, unlike the exterior surface of the IHS which mates to the HSF.

 

[Idontcare]

But I am unsure about a good choice of TIM to use *on die* now for various reasons:

* You say ICD would be your choice, but then I read negative things how it scratches the die...
* NT-H1 suffers from pump-out, so it's not that great
* MX-4 seems to be a catastrophe (for whatever bizarre, unknown reasons) for on-die
* CLU would be my choice, but it's electrical conductive and probably a mess to work with.

Sort-of unsure what to do now. Right now I only have MX-4 here, but amazingly I also read that some people did a simple delid, put a MX-4 (!!!) on the die, put cap back on (no glue)..and saw a MASSIVE drop in temps.

Really confused now how to proceed...

MX-4 will experience the same pump-out issues that NT-H1 experiences when used on direct-die applications.

My #1 recommend for direct-die application is CLU. Just don't go crazy on the amount you add and you will have nothing to worry about. The web is full of walk-throughs on delidding and using CLU, my own threads show you how and many other people's threads show you as well.

 

[xorbe]

Has anyone tried to quantify the gap between the die and inner lid? I was going to try to estimate it with anything I could use as shims for measurement (and my caliper).

 

[Idontcare]

Has anyone tried to quantify the gap between the die and inner lid? I was going to try to estimate it with anything I could use as shims for measurement (and my caliper).

Sure, see this thread for an example.

(note the links in the OP of that thread)

 

[xorbe]

Sure, see this thread for an example.

(note the links in the OP of that thread)

Thanks! Should I use my CLU, or try to save and reuse the Intel stuff?

 

[Idontcare]

Thanks! Should I use my CLU, or try to save and reuse the Intel stuff?

CLU, do not attempt to reuse Intel's stuff!

See: http://forums.anandtech.com/showpost.php?p=33785983&postcount=53

 

[jj109]

I saw this on OCN and was reminded of this thread. Hey IDC, there's that pump out you were talking about.

 

[Idontcare]

I saw this on OCN and was reminded of this thread. Hey IDC, there's that pump out you were talking about.

Dang, look at that temperature creep! Great find, thanks for posting and sharing :thumbsup:

 

[rumpleforeskin]

Pretty interesting, that would be 180 cycles crammed into 24 hours. So for the average user who turns their PC on in the morning and off at night this could approximate a 6 month outcome.

 

[jj109]

Pretty interesting, that would be 180 cycles crammed into 24 hours. So for the average user who turns their PC on in the morning and off at night this could approximate a 6 month outcome.

I counted about 200 spikes, so that's a little over 24 hours of testing. It's hard to translate this to real world lifespan of a NT-H1 application on the CPU die, but one temperature cycle per day is probably low balling.

On the other hand, I've had excellent results from NT-H1 on water-cooled GPU die despite countless # of Witcher 3 induced temperature cycles. I wonder why it doesn't degrade like on the CPU die. Lower temperature differential? Or maybe it's the expansion IHS itself causing issues?

 

[PliotronX]

Meh, I have NT-H1 on my GTX 780 and that sees several cycles per day by loading up games. No temperature creep in 1.5 years.

 

[Idontcare]

Pretty interesting, that would be 180 cycles crammed into 24 hours. So for the average user who turns their PC on in the morning and off at night this could approximate a 6 month outcome.

That is very close to my own experience in terms of practical lifetime of an NT-H1 mount on bare CPU silicon die. Obviously it is end-user specific as it depends on their actual compute needs and thermal cycling of the CPU.

I counted about 200 spikes, so that's a little over 24 hours of testing. It's hard to translate this to real world lifespan of a NT-H1 application on the CPU die, but one temperature cycle per day is probably low balling.

On the other hand, I've had excellent results from NT-H1 on water-cooled GPU die despite countless # of Witcher 3 induced temperature cycles. I wonder why it doesn't degrade like on the CPU die. Lower temperature differential? Or maybe it's the expansion IHS itself causing issues?

The pump out effects will be largely surface property dependent. Meaning the actual chemical composition of the exposed surface layer (surface energy, gibbs free energy, etc.) will determine to first order the rate of adhesion or separation of the viscous (semi-fluid) TIM under pressure.

A silicion surface that was passivated at TSMC will most certainly be different relative to the silicon surface which has been through an Intel fab. A TIM which works well on one surface could quite possibly perform rather poorly on the surface which exited another foundry. Was the surface nitrided, polyamide, oxide, other?

 

[DrMrLordX]

The orientation of the die with respect to gravity may also have an effect. In a conventional tower PC, the narrow edge of the die is pointing in the general direction of gravitational force, whereas the die of a GPU is normal/perpendicular to gravitational force.

 

[PliotronX]

The orientation of the die with respect to gravity may also have an effect. In a conventional tower PC, the narrow edge of the die is pointing in the general direction of gravitational force, whereas the die of a GPU is normal/perpendicular to gravitational force.

That sounds too logical.. nevermind sorry. Carry on.

 

[crashtech]

My hypothesis is that pump-out may be exacerbated by higher than anticipated clamping force per unit area on the die, which might be expressed as PSI or kPa. The bare die to IHS contact patch has a much smaller surface area than the IHS to HSF contact patch, yet is subject to the exact same clamping force if the adhesive is removed during delidding. This results in far greater pressure on the bare die TIM application.

 

[soccerballtux]

I saw this on OCN and was reminded of this thread. Hey IDC, there's that pump out you were talking about.

do we know this isn't the system just warming up?

edit: let me expand the picture and read the title of it first, heh

 

[soccerballtux]

My hypothesis is that pump-out may be exacerbated by higher than anticipated clamping force per unit area on the die, which might be expressed as PSI or kPa. The bare die to IHS contact patch has a much smaller surface area than the IHS to HSF contact patch, yet is subject to the exact same clamping force if the adhesive is removed during delidding. This results in far greater pressure on the bare die TIM application.

really? at 75C it's going to be so viscous. I'd call it leak out honestly

 

[crashtech]

really? at 75C it's going to be so viscous. I'd call it leak out honestly

But there is no explanation why there is a difference between the two types of mounts, why one pumps out and the other does not.

 

[KingFatty]

Difference (i.e., mismatch) between the two different materials making the interface where the TIM goes?

I could see where you have the heat spreader still in place, it's a metal similar to your heatsink, so they expand/contract at similar rates.

But when you go bareback, the bare silicon die might expand/contract differently than the metal heatsink, so you get enhanced pumping effect?

Just speculation...

 

[crashtech]

I think Intel puts the CPU/IHS assembly in a jig with a predetermined (fairly low) amount of pressure applied while the adhesive around the IHS perimeter sets. The cured adhesive acts as a buffer or spacer to prevent the high pressure required to seat all 1150 lands plus the pressure needed to seat the HSF from basically crushing all the TIM out from between the IHS and the die.