De-Lidding and Overclocking Core i7-7700K with Water and LN2

[Martell1977]

Have you tried replacing the TIM with fluxless solder like they used in SB days? I'm curious what results that would yield with Kaby and the longevity would be better than paste.
Also, what about mineral oil, what would that do for OC?I've seen kits online for this, so just curious.

 

[InvalidError]

It still boils down to the fact that for their top CPU (mainstream), they should put more effort into the TIM and lid.

And Intel likely did. Except that Intel's priority is reducing costs, not achieving unnecessary low temperatures as far as stock reliability is concerned and higher overclocks when overclocking is not officially supported, merely tolerated.

The main reason Intel ditched soldering is because silicon, indium, copper and the substrate all have vastly different thermal expansion coefficients and that puts all sorts of stresses on the indium solder, die balls and the die itself which can crack any one of them and the bonds in-between. With thermal paste, most of it is eliminated along with the expensive soldering preparation steps and soldering process. The indium layer also needs to be fairly thick to reduce the temperature gradient and the risk that the indium layer or chip corners will crack. How thick is that? From what I read, the ideal thickness is 2mm and an indium layer that thick would perform worse than most generic ZnO pastes. Also, the amount of indium necessary to fill that gap (about 1.5 gram) would cost around $5. Indium isn't cheap.

Why the extra TIM thickness under Intel's pasted-on IHS? I'm guessing because if the paste was only thick enough to fill contact gaps between physical contact points as it is for heatsinks, there would be a risk that forces prying the die away from the IHS during thermal cycling could be sufficient to separate the die from the IHS and draw air in. With a thicker paste layer, the vacuum will slowly draw paste on the periphery in as the surfaces pull apart instead of air. The thicker layer may be more long-term stable at the expense of worse thermal conductivity.

 

[cryoburner]

I've read somewhere that aftermarket TIMs need to be re-applied every 6-12 months too, because they tend to stay wet and eventually become inefficient. Don't take my word for it as being the absolute truth, but that would explain why Intel puts a TIM that tends to become more dry, in order for it to be useful for the whole life of the chip.

That's true for many thermal compounds, including some popular ones like Arctic Silver 5, which tends to dry out and lose its cooling performance, and should be replaced after a year or so. Plus its consistency makes it more difficult to work with and clean off the processor, making it all around more difficult and costly to maintain. Some other compounds, like Arctic Cooling MX-2 and MX-4 are said to have 8-year durability, are easier to apply and remove, and provide similar heat transfer at a similar cost per application, except you won't need to re-apply the compound repeatedly. Getting perhaps an extra 1 degree of cooling out of a thermal compound isn't really worth it if that performance ends up worse over time, and requires frequent maintenance.

And yes, I imagine the compound Intel uses under their spreaders is likely designed and tested to last a decade or more, hence why they use it in place of some compound that provides a bit better cooling, but only lasts a year. The heat spreaders are not intended to be removed, so it has to be something that will last for the life of the CPU.

 

[nikoli707]

take the heat spreader off, results will be better.

 

[damric]

I delidded my Skylake a while back. I intended to run without the heat spreader but my cooler just wouldn't sit flush on the CPU die even after I removed all the retention brackets. The good news is that it was about a 20 degree drop at high loads with high overclocks just going to the liquid metal. When I did the same tests with MX-4, the stock Intel grease actually performed better so it's really not all that bad.

 

[InvalidError]

When I did the same tests with MX-4, the stock Intel grease actually performed better so it's really not all that bad.

And I doubt you shimmed the IHS by the same thickness as the original glue when you put it back on, which means that your MX4 layer should be that much thinner than the original paste too, which makes the original paste that much more impressive.

Unlike aftermarket pastes which need to be goopy enough to squeeze out of tubes and syringes for convenient manual application, machine-applied factory pastes can be packed much tighter/harder for better thermal conductivity. It is the fine particles in the grease that carry the bulk of the heat, not the base oil used in the grease. The more densely packed particles are, the less oil there is between particles, the dryer the paste looks and the better it performs since it behaves closer to a solid shim of whatever particles it uses.

You'd much prefer a ZnO TIM to behave closer to ZnO's 50W/mK thermal resistance than its silicone base oil's ~2W/mK but with human-friendly pastes, you only get 7-12W/mK since they are more base oil than solids.

 

[bit_user]

When I did the same tests with MX-4, the stock Intel grease actually performed better so it's really not all that bad.

I've seen other de-lid tests which confirm that MX-4 performs poorly when applied directly to the die. Therefore, it's not a good basis for establishing the performance of the Intel TIM.

The same review established the supremacy of Coollaboratory liquid metal compounds for delid/relid or direct-die cooling. But that was a few years ago...

Anyway, that was bold of you to try direct-die without a shim. I'm glad your CPU still works!

 

[damric]

I don't know. I used MX-4 for other direct die applications like on Radeon graphics cards and it works great. I was really disappointed with the results on CPU die though. The CLU worked great though. I'd definitely dare to do it again. Using a razor was not hard, maybe took 3 minutes.

 

[Co BIY]

Is the Intel "lid" made of Aluminum ? Would a pure copper lid increase performance ?

 

[chalabam]

Try peltier cells. Better than water on long tests, and doesn't run out of nitrogen. Ice remains a problem, sadly.

 

[InvalidError]

Try peltier cells. Better than water on long tests, and doesn't run out of nitrogen.

Peltier elements are very inefficient at generating temperature gradients larger than 10C between their hot and cold side under significant thermal load. If you take a 100W Peltier cooler to cool a 100W chip, you'll need to pump ~100W into the Peltier cooler to merely manage to maintain a 0C temperature difference between cold and hot side and you'd be better off skipping the Peltier altogether than trying to deal with 200W of total dissipation. If you want to achieve significant net cooling, you'll need to pump even more power through the Peltier and somehow manage to keep its hot side near ambient temperature.

Peltiers are neat in theory but not very practical beyond 50W or so.

 

[bit_user]

Is the Intel "lid" made of Aluminum ? Would a pure copper lid increase performance ?

Based on my experience of lapping a Haswell, I think it's copper plated with what could be aluminum.

 

[bit_user]

Try peltier cells. Better than water on long tests, and doesn't run out of nitrogen. Ice remains a problem, sadly.

You need to read this:

http://www.tomshardware.com/reviews/phononic-hex-2-thermoelectric-cpu-cooler,4665.html

...and that basically slams the door on Peltier cooling of CPUs and GPUs. I, for one, am glad they tried it. It's like a Myth Busters experiment, so we now have experimental evidence that it just doesn't work.

They're still good for making things cold, like the image sensors used in astrophotography. But those are comparatively low-power devices, making the thermal gradient much easier to maintain.

 

[damric]

Every TEC/Peltier build work log I have seen on this and a few other forums over the years was either 1) money pit with mediocre performance 2) total failure going down in flames, or 3) all of the above.

I have seen some awesome applications on other electronics in mills and factories as a technician, but never on a PC.

 

[Discorama]

Aren't the lids made out of Aluminum? The article said the Grizzly TP shouldn't be used on Aluminum.

 

[InvalidError]

Based on my experience of lapping a Haswell, I think it's copper plated with what could be aluminum.

It is usually nickel plating, much easier to do than aluminum.

 

[bit_user]

Based on my experience of lapping a Haswell, I think it's copper plated with what could be aluminum.

It is usually nickel plating, much easier to do than aluminum.

I was actually going to guess nickle, as I've read about heatsinks using nickle-plating, but I have no factual basis for saying that my Haswell's lid was nickle-plated. To me, it had the appearance more of aluminum than nickle, but I don't know that nickle couldn't look like that.

Or, being Intel, maybe they used some kind of alloy.

 

[enterprise24]

Nice to see Tomshardware doing extreme overclocking

 

[InvalidError]

To me, it had the appearance more of aluminum than nickle, but I don't know that nickle couldn't look like that.

The appearance of silvery metals changes drastically depending on the type of surface finish. When they are all finely polished though, most people wouldn't be able to tell them apart without touching.

 

[envy14tpe]

Could you clarify on the properties of the glue and what kind of thermal paste would be needed for this? Asking to make sure I do the procedure a tested and true way.

 

[old_newbie]

Stopped reading the article halfway through page 1 after they said "All that's left is add a little glue and reseal the processor." Waaaa???

The whole point of "de-lidding" back in the day was to apply a bigger heatsink directly to the die. I de-lidded all my athlon 64's (before they started soldering them). Never used a shim, never cracked a die, either.

Kids today. No sense of adventure...

 

[InvalidError]

Never used a shim, never cracked a die, either.

The dies and substrates are thinner than they were before and Intel's LGA socket requires pressure more evenly across the whole thing for reliable contact. If you don't use shims on LGA1151, the system will most likely fail to boot because pins near the edges and corners are failing to make contact. The extra flex near the middle of the substrate may also crack solder balls under the die.

 

[bit_user]

Stopped reading the article halfway through page 1 after they said "All that's left is add a little glue and reseal the processor." Waaaa???

The whole point of "de-lidding" back in the day was to apply a bigger heatsink directly to the die.

Where I took exception was to the idea of re-gluing it. While it should helps combat drying of the new thermal compound, it partially defeats the point of delid/relid, which is to decrease the gap between the die/IHS.

 

[InvalidError]

While it should helps combat drying of the new thermal compound, it partially defeats the point of delid/relid, which is to decrease the gap between the die/IHS.

If you press the IHS onto the substrate while re-gluing it, the adhesive thickness will be next to nothing compared to the original glue. Also, most thermal pastes use silicone oil as their base which doesn't dry, it migrates out of the thermal interface as thermal cycling causes particles to pack down tighter and force the excess oil base out, which is why most thermal pastes get better over time and thermal cycles. The problem with "dried paste" is that once the base oil has been squeezed out, there is not enough of it left inside the interface to "heal" it when mechanical stress (ex.: nudging the heatsink) breaks it.

 

[bit_user]

If you press the IHS onto the substrate while re-gluing it, the adhesive thickness will be next to nothing compared to the original glue.

Okay, I guess I was imagining the glue was far more viscous than it apparently is.

BTW, I started to de-lid a haswell-refresh i3, just to see how hard it was & because I didn't care too much if I ended up killing it. I stopped when I found it difficult to jam the razor blade between the lid and the PCB. At that point, I decided the glue was thinner than I was lead to believe & successfully completing the job was going to be more annoying than I thought.