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

Idontcare · Nov 27, 2012

I first became acquainted with Noctua's NT-H1 TIM when I bought my NH-D14 HSF and it came with a tube of the stuff.

I was quite surprised, pleasantly so, to find out that the chemists who work on improving thermal compounds were in fact making progress over the years

Unlike the Arctic Silver products, Noctua claims that NT-H1 does not require a burn-in time to reach optimal thermal conductivity.

Top-performance right from the start
NT-H1 reaches its full performance extremely fast and doesn't require a longer "burn-in" time.

However one serious drawback to this TIM, if the anecdotal reports across various forums is considered, is that it suffers from the pump-out effect when used in direct-die applications.

I haven't experienced this phenomenon yet but I haven't left NT-H1 on my bare-die for more than a week at this juncture so I wouldn't expect to see it in such short time either.

Regardless its unsuitability as a CPU TIM replacement, I tested it anyways just to see how it stacked up against the other TIMs at least from a time-zero perspective before the onset of the pumping-out effect.

I figured for this test I'd have a little fun so I popped open my computer parts toolbox which has some grains of rice in it. What? Doesn't yours?

On goes the NT-H1 and a couple grains of rice just for demonstration purposes.

I know the internet lore is that we are supposed to put on "an amount comparable to a grain of rice", but in this case I put on an amount equal to roughly 3 grains of rice.

And the resulting mounting image?

Pictures of perfection if I do say so myself

And the temperature data were quite good too!

That's about 5°C better than my best IC Diamond results at the high end (4.9GHz).

But, getting back to reality here, the best results are those that can be counted on for months to come and not just the ones that look good for the first few weeks. NT-H1 may produce a superior mount when bare silicon is involved but the pumping-out effect is a deal-killer for any serious consideration of using this TIM long-term with a bare silicon cpu application

tommo123 · Nov 27, 2012

IDC

is this going to be posted as an article on the anandtech page itself? it should.

thanks for this

hokies83 · Nov 27, 2012

Saved Liquid Ultra for last lol.

thilanliyan · Nov 27, 2012

Just to add to the data:

http://skinneelabs.com/2011-thermal-paste-review-comparison/2/

This review shows very little variance between TIMs (max about 5C). Is there any other reason than chemical makeup that could cause the large variance in IDC's data? I'm honestly shocked for example that NT-H1 can perform say 10C better than MX-4...?

EDIT: Maybe better cooling capacity in the loop at skinneelabs exaggerating the performance differences? That exaggeration is mostly useful I think...since most people likely use a H100/H70/etc, rather than a custom loop.

Haserath · Nov 27, 2012

I figured for this test I'd have a little fun so I popped open my computer parts toolbox which has some grains of rice in it. What? Doesn't yours?

No, mine has peas.

rge2 · Nov 27, 2012

thilanliyan said:
Just to add to the data:

http://skinneelabs.com/2011-thermal-paste-review-comparison/2/

This review shows very little variance between TIMs (max about 5C). Is there any other reason than chemical makeup that could cause the large variance in IDC's data? I'm honestly shocked for example that NT-H1 can perform say 10C better than MX-4...?

EDIT: Maybe better cooling capacity in the loop at skinneelabs exaggerating the performance differences? That exaggeration is mostly useful I think...since most people likely use a H100/H70/etc, rather than a custom loop.

Testing Tim1 interface (between die and IHS/heatsink) is not same as testing Tim2 (between IHS and heatsink), as a larger temp gradient will occur across the more critical tim1 interface. Higher variance between different tims at tim1 interface is expected.

thilanliyan · Nov 27, 2012

rge2 said:
Testing Tim1 interface (between die and IHS/heatsink) is not same as testing Tim2 (between IHS and heatsink), as a larger temp gradient will occur across the more critical tim1 interface. Higher variance between different tims at tim1 interface is expected.

True, never thought of that...forgot that IDC's is delidded.

exar333 · Nov 27, 2012

Idontcare said:
I first became acquainted with Noctua's NT-H1 TIM when I bought my NH-D14 HSF and it came with a tube of the stuff.

I was quite surprised, pleasantly so, to find out that the chemists who work on improving thermal compounds were in fact making progress over the years

Unlike the Arctic Silver products, Noctua claims that NT-H1 does not require a burn-in time to reach optimal thermal conductivity.

However one serious drawback to this TIM, if the anecdotal reports across various forums is considered, is that it suffers from the pump-out effect when used in direct-die applications.

I haven't experienced this phenomenon yet but I haven't left NT-H1 on my bare-die for more than a week at this juncture so I wouldn't expect to see it in such short time either.

Regardless its unsuitability as a CPU TIM replacement, I tested it anyways just to see how it stacked up against the other TIMs at least from a time-zero perspective before the onset of the pumping-out effect.

I figured for this test I'd have a little fun so I popped open my computer parts toolbox which has some grains of rice in it. What? Doesn't yours?

On goes the NT-H1 and a couple grains of rice just for demonstration purposes.

I know the internet lore is that we are supposed to put on "an amount comparable to a grain of rice", but in this case I put on an amount equal to roughly 3 grains of rice.

And the resulting mounting image?

Pictures of perfection if I do say so myself

And the temperature data were quite good too!

That's about 5°C better than my best IC Diamond results at the high end (4.9GHz).

But, getting back to reality here, the best results are those that can be counted on for months to come and not just the ones that look good for the first few weeks. NT-H1 may produce a superior mount when bare silicon is involved but the pumping-out effect is a deal-killer for any serious consideration of using this TIM long-term with a bare silicon cpu application

This is why I really like the IC-7 TIM. Knowing I can mount the HSF and not have to do anything other than occasionally dust the HSF area is really nice. REALLY nice for cramped SFF builds where it can be VERY onerous to take the whole thing apart in a reasonable amount of time.

hokies83 · Nov 27, 2012

Only the Liquid Ultra is left...

Remember Wiggle it to make it spread more easy lol..

Also a circle motion can help after the wiggle.

Gillbot · Nov 27, 2012

Idontcare said:
This is an oldie but still a goodie, a classic for us vaporphase cooling enthusiasts because it is designed and rated for use without losing its thermal conductivity at sub-zero temperatures.

I am talking about Artic Silver's Céramique of course!

I could not find any technical specs for this TIM from the manufacturer in terms of thermal performance or viscosity. It is spec'ed for use down to -150°C though

One caveat with this TIM is that per the manufacturer there is a break-in period over which the TIM increases its performance:

Naturally I could not be bothered to wait "a minimum of 25hrs" to see just how much better the temperatures became for this TIM. I tested it straight after application the same as all the other TIMs, no special treatment here

I'm sure the temperatures improve in time, but they are pretty good at time-zero. Bottom line though is that I'm not about to start using Céramique as my 24/7 TIM of choice so I'm not exactly concerned with nailing down its exact performance to 3-9's precision This is intended to be a "good enough for gov'ment work" assessment, at best, and in that regard I feel like I have given Céramique a fair shake.

By eye Céramique is no more viscous than AS5 or NT-H1 (or MX-4 for that matter), it is white in appearance and very easy to apply. It is also very easy to cleanup and remove. It gets a 10/10 for all of the ease-of-use categories

Obligatory pre-mount photo-op:

Post mounting images:

^ as you can see in the post-mount images, this stuff spread out beautifully when mounted. Such a crisp and clean mount.

Actually it spreads out so well that I'm concerned that with long-term use this TIM might suffer the same "pump-out effect" that AS5 is noted to suffer from in this same application.

Temperature results were quite good, not to speak out of turn but they actually proved to rival those of IC Diamond; which surprised me except for the fact the mount was just so darn awesome how could Céramique not deliver good temps?

A key difference though between IC Diamond and Céramique is that IC Diamond is reportedly designed to prevent the so-called pump-out effect, so in that sense Céramique may look good at first but in the end Céramique simply won't work for long-term use in a bare-die mounting application.

I can confirm that it will "pump out" over extended periods of time. I've been using it as my 24/7 TIM for some time, basically since it's initial release. On machines that have been in service for over ~18 months tend to start gradually climbing in temperature. I initially contributed it to "dust buildup" but even after monthly cleanings the temps still seemed to creep up. After re-applying the TIM, they were fine once again.

thilanliyan said:
Really never expected ceramique to do so well! I'm using mx4 myself for both cpu and gpu...maybe I should consider something else next time I take my loop apart.

Since it's initial release it yielded results many didn't expect. I've been using it as my only TIM since. With the typical few C spread difference from this TIM to the "best" TIM, I decided I liked ceramique the best for it's ease of application, low cost and overall performance per $ spent.

bononos · Nov 27, 2012

Its shocking to see the MX4 perform relatively poorly

Does it (MX4) or the AS5 suffer from the pump out effect though? And the pump out effect is mitigated by spring loaded clips/tensioners right?

Edit- I saw comments about the AS5 suffering from pump-out effect but the website says:
Absolute Stability:
Arctic Silver 5 will not separate, run, migrate, or bleed.

meloz · Nov 27, 2012

IDC, you have put more effort in this exercise than most commerical websites do when they test gear. Brilliant thread. The photos and narration really help us in understanding what you are doing.

I am in awe.

Torn Mind · Nov 27, 2012

I am utterly and totally fascinated by the contents of this thread. Keep up the good work.

Idontcare · Nov 27, 2012

meloz said:
IDC, you have put more effort in this exercise than most commerical websites do when they test gear. Brilliant thread. The photos and narration really help us in understanding what you are doing.

I am in awe.

Torn Mind said:
I am utterly and totally fascinated by the contents of this thread. Keep up the good work.

:$ thanks for the kind words, I'm really glad to hear you enjoyed it. That was the hope and the goal.

Idontcare · Nov 27, 2012

When it comes to these liquid-metal TIMs, my experience up until recently was limited to Indigo Xtreme. Indigo Xtreme delivers pretty much unparalleled results because it is essentially a low melting-point solder. If you truly desire to replace the CPU TIM on your Ivy Bridge with a solder equivalent to become more like a soldered Sandy Bridge CPU then you need look no further than Indigo Xtreme.

On the other end of the spectrum of liquid-metal TIMs is Coollaboratory Liquid Ultra which is a true liquid metal alloy at room-temperature and remains a liquid at all operating temperatures above 8°C. (it doesn't boil until reaching 1350°C)

The one advantage of using a true liquid metal TIM versus the solder-like Indigo Xtreme TIM is that you can more easily remove your HSF from the IHS (or the exposed CPU die in my case) the same as you would if you used any other TIM such as AS5 or MX-4. Removing Indigo Xtreme from my 2600k and H100 required a razor blade, a hammer, and some 200 grit sandpaper :0

The primary component in Liquid Ultra is the Gallium. According to Coollaboratory there are a few more alloying elements involved:

Chemical characterization:
Alloy of the metal components gallium, indium, rhodium, silver, zinc and stannous, bismuth; suspended in a graphite-copper matrix

It flows like a liquid, it pours just like water. It is also electrically conductive, so care must be taken to ensure it does not come in contact with anything that might be electrically biased.

That last comment may make it sound like a bad idea to put it on your CPU but the CPU silicon itself is electrically insulated by an insulator oxide material. Provided we steer clear of the socket and the landing pads on the top of the PCB we shouldn't be in any danger of electrically shorting out the CPU.

Applying Liquid Ultra is similar to applying any other TIM, you squeeze some out of the tube onto the surface in question.

There is one minor step involved, not sure if it is really required but I did it anyways, which is to use a provided brush to spread-out the Liquid Ultra prior to mounting the HSF.

Here's what it looked like after I brushed the Liquid Ultra around on top of the silicon die:

^ the TIM looks brownish/copper in color in this image because it is actually much like a mirror and it is reflecting the copper color from the nearby H100 waterblock.

After the tests, unmounting the H100 to see the die post-mount:

(again, the brownish color is an artifact of reflecting the light coming off the copper of the nearby H100, the TIM itself is very much a mirror-like silver)

^ the imprint on the H100 is nice, exactly what you want to see. The Liquid Ultra remained confined to the area local to the CPU, it did not wick across the CPU PCB or the H100 copper surface.

Cleanup was rather painful, to put it mildly. While all the Liquid Ultra can be removed, mostly, it takes a bit of work.

Here is what you get to look at as you start to wipe away the Liquid Ultra from the copper block:

Further cleaning up the H100 with IPA and paper towels removes nearly all of the Liquid Ultra:

Same deal with the CPU itself, removing the liquid ultra was a complete PITA to put it mildly :/

Really laying into the H100 with IPA soaked towels and scrubbing the bejebus out of it resulted in the following:

^ the silver sheen masking the copper color is an indication of the fact that there is still a thin layer of Liquid Ultra adhering to the H100. The only way I could fully remove all the Liquid Ultra was to re-lap the H100 with 800 grit sandpaper (followed by 1000/2000/3000 grits).

^ boom, pristine H100 surface restored

(Liquid Ultra results are continued in the next post, VB restricts me to a max of 10 images per post)

Idontcare · Nov 27, 2012

(Liquid Ultra results continued...)

Removing the last remnants of Liquid Ultra from the CPU posed to be an even greater challenge because I couldn't fall back on using sandpaper to remove the remaining layer of TIM.

In the end my poor 3770k CPU didn't fare as well as my H100. Between the abrasion caused by the IC Diamond and the hazing (remnants of Liquid Ultra that could not be removed), this CPU is starting to look a bit worse for the wear.

^ compare this "after" photo to the nice pristine surface I started with in the following "before" photo:

Like high-school prom night, sometimes the goods aren't returned to their owner in that original factory-new condition

Alright then, I know you didn't come here for the double entendres and Pulitzer Prize caliber witty writing, so here are the results from the Liquid Ultra tests:

These results are basically ~5°C better than IC Diamond at the higher end of the clockspeed range (4.9 GHz). Surprisingly, really surprisingly to me, the NT-H1 results were comparable to the Liquid Ultra results.

Liquid Ultra still beat NT-H1 by a very slight margin, and of course Liquid Ultra wins out in terms of actually being usable as a long-term solution with bare-die mounting whereas NT-H1 will give you grief with its pump-out effect issues.

Idontcare · Nov 27, 2012

Pulling all the results from the nine different TIMs together, we arrive at the following table:

I rank-sorted the table from left to right in order of decreasing thermal performance. The closer to the left-hand side of the chart the better the TIM performed in the tests.

As others have noted, I am at a loss to explain why MX-4 performed so poorly in these tests. Out of disbelief I remounted and reran the tests multiple times and the results were repeated to within a degree each and every time.

The poor showing for MX-4 is either due to some unforeseen interaction with the silicon die of the cpu (not exactly the target application for MX-4

) or the rigorous thermal environment that is created by the much higher concentration of heat at the surface of the CPU die itself versus that of the IHS-to-HSF interface is highlighting what would otherwise be a very small performance delta between MX-4 and the other TIMs.

There are other possibilities too, for instance there is a chance that my particular tube of MX-4 (I only have one tube) is somehow borked.

However, in testing MX-4 vs. NT-H1 from the exact same tubes using the exact same H100 and same 3770k (albeit with the IHS still in place), the MX-4 performed just 1°C poorer than the NT-H1. They were practically indistinguishable in terms of thermal performance in that application.

So this tells me that these seemingly anomalous MX-4 results appear to indicating that MX-4 just doesn't react well to be placed in direct contact with a bare silicon die.

OK, getting back to the big picture view of the results here, perhaps a graph will better enable digestion of the results:

Note I intentionally zoomed in on the temperature axis to show as much differentiation between the various TIMs as possible.

Surprisingly most of the TIMs performed within a degree or two of the Liquid Ultra, although for some of these TIMs that result is the uninteresting one because we probably wouldn't want to use the TIM in this application anyways owing to the pump-out effect.

The two TIMs that would appear to merit use for this direct-die application are (1) Liquid Ultra, and (2) IC Diamond. Neither are expected to suffer from the pump-out effect and both provide excellent cooling results for this bare-die application.

Unfortunately there is a downside to both of these TIMs, when it comes to cleanup and removal of the TIM you are looking at a situation that is decidedly more challenging than that of your everyday TIM. Between the copper staining and gallium residuals, both requiring lapping to fully remove, and the irreversible alterations to the surface of the exposed silicon die, if you choose to pursue using either of these TIMs in your setup just be prepared to more or less live with things looking used in a way that you'd never expect to see from the other TIMs like NT-H1 and AS5.

That said, I personally feel the upside more than outweighs the downside and would not hesitate to use either IC Diamond or Liquid Ultra on my H100 or CPUs again.

I will most likely settle on using the IC Diamond for my long-term CPU TIM at this point in time just because it is not electrically conductive and the temperature difference between the two is not enough to make all that much of a difference in terms of power-consumption or overclock potential. Just a personal preference at that point, I can appreciate that others would choose differently given their specific situation and preferences as well.

And that about wraps it up. I do have one more test to complete and post about, the ridiculously excessive NT-H1 test

:twisted: I may have gotten just a tad too much NT-H1 on there

LOL, I'll post up the results on this funny test tomorrow. Feel free to contribute your own bare-die results to this thread if you like. I'm curious to know if people have observed pump-out effect with Liquid Ultra, or confirmation that it hasn't been observed for x-number of months and so on.

hokies83 · Nov 28, 2012

You used to much Liquid Ultra Bro

use 1/4 of that and try again XD

Half a rice grain. <-- and i think that is to much.

Liquid Ultra is made to Spread does not work as well if u let it spread it's self.

http://youtu.be/3N3D1zaeJoU

Watch the video of how much he used on a Huge Amd IHS.

GPz1100 · Nov 28, 2012

IDC, thanks for your efforts.

It is interesting to note how high the temps jump between 4.5 and 4.7ghz all across the board. Seems a wall of sorts is hit a that point, and alot more Vcc is needed to cross over.

Any reason you didn't do a 4.6ghz?

hokies83 · Nov 28, 2012

I just took these pictures afew mins ago.. after i made the post above to show the amount needed to cover a Whole IHS.

Your goal is no puddles your trying to put on like paint pretty much.

dma0991 · Nov 28, 2012

Great job IDC, didn't realize that AS5 still managed to be competitive with newer TIMs. Now I have a strange desire to buy the NT-H1 to get a few °C lower. :hmm:

Kenmitch · Nov 28, 2012

IDC did you do any testing with HT on/off? Was wondering how much HT effects temps on IB's.

Haserath · Nov 28, 2012

Now watch as IDC's NT-H1 crazy test gives the best results. /TIM demand skyrockets!

Nice info. Thx IDC!

Denithor · Nov 28, 2012

Gillbot said:
I can confirm that it will "pump out" over extended periods of time. I've been using it as my 24/7 TIM for some time, basically since it's initial release. On machines that have been in service for over ~18 months tend to start gradually climbing in temperature. I initially contributed it to "dust buildup" but even after monthly cleanings the temps still seemed to creep up. After re-applying the TIM, they were fine once again.

Since it's initial release it yielded results many didn't expect. I've been using it as my only TIM since. With the typical few C spread difference from this TIM to the "best" TIM, I decided I liked ceramique the best for it's ease of application, low cost and overall performance per $ spent.

Pretty much the same experience here. Been (quite happily) using Ceramique since it first launched and have been through quite a few tubes over the years. I have seen the same thing - gradual climb in temps after about 1-1.5 years mounted. This pump out phenomenon happens, but it is definitely not a fast thing, certainly not something that should stop you from using for relatively long-term periods of time on bare die. It doesn't just stop working suddenly one day, you just notice that your load temps start to gradually increase and take that as a reminder to reapply at that time.

Nice to see how well Ceramique holds up against the best out there - exactly on par with ICD and only a few degrees below the absolute best. I'll just stick with it for now.

Thanks for the work, awesome read!

Zap · Nov 28, 2012

Great thread! Thanks, IDC!

Idontcare said:
I devised a simple tool for quickly verifying that the stacks of washers were indeed the correct height:

A toothpick which has been marked in blue to the desired height and washer count. With this handy tool I could quickly touch it next to each of the four stacks of washers and visually verify no stack was missing a washer or had too many washers.

Why didn't you use your calipers? Doesn't it have a pin that extends out as you increase the measurement? Use that part.

Idontcare said:
The second TIM tested was IC Perihelion.
...
What I observed in using this stuff is that it is akin to a thermal pad in a tube. This stuff is thick, viscous, not very pliable. It is kind of like putting a piece of cheese on your CPU in terms of consistency.

And not too surprisingly this TIM was the most difficult to get a good mount. It was very very difficult to get enough on the CPU so as to cover the die while at the same time avoid getting so much on the die that it would form a 1mm thick pad when I mounted the H100.

Sounds like Shin Etsu stuff. It is useful in some specialized circumstances where you don't have anywhere near a perfectly mating surface or a gap that needs bridging, but otherwise a PITA to use.

Idontcare said:
I had heard of the game "52 card pickup", and now I've played the game "50 washer pickup". Will not be making that same mistake twice

How permanent is your washer setup? Maybe you can glue them together as a stack? Get your desired stack prepped in a clamp (maybe cover clamp in a plastic baggie) and then drip some super glue gel into it.

Kenmitch said:
I'm kinda shocked it spread out so good....That stuff if pretty thick and kinda hard to work with from what I remember. Last time I used it was on my GPU water block a couple years ago....Seems to be holding up just fine to this day.

I've had a big tube of Ceramique that was super thick, obtained from someone else and not purchased new. I have also used new tiny tubes of the stuff that was much thinner. Made me wonder if the stuff thickened with age, or if they changed formulations over the years.

Gillbot said:
I can confirm that it will "pump out" over extended periods of time. I've been using it as my 24/7 TIM for some time, basically since it's initial release. On machines that have been in service for over ~18 months tend to start gradually climbing in temperature. I initially contributed it to "dust buildup" but even after monthly cleanings the temps still seemed to creep up. After re-applying the TIM, they were fine once again.

:hmm: I rarely go over a year without upgrading something (better HSF, faster CPU, replacing entire system) and often much sooner. I'm not a "build once, use five years" kind of guy.

Guess I won't worry about pump-out effects.

Gillbot said:
With the typical few C spread difference from this TIM to the "best" TIM, I decided I liked ceramique the best for it's ease of application, low cost and overall performance per $ spent.

:thumbsup:

I like my women as I like my TIM. Cheap, reasonable performance and easy to use. :biggrin:

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

Elite Member

Platinum Member

Senior member

Lifer

Senior member

Member

Lifer

Diamond Member

Senior member

Lifer

Diamond Member

Senior member

Lifer

Elite Member

Elite Member

Elite Member

Elite Member

Senior member

Senior member

Senior member

Platinum Member

Diamond Member

Senior member

Diamond Member

Elite Member