Replacing the thermal interface material in a CPU package?

[OmNominous]

Greetings -

Well, in the last couple of weeks I had a recently-installed used MSI R9 290 erupt into flames (after 2 weeks of use) and a week later a short on my motherboard appeared resulting in another impressive light show. ("I don't remember installing any white LEDs in this case... HOLY $#!%!!") It's not immediately clear if this was caused by something ejected from the R9 290 card or if the power supply (a new EVGA 1200 P2) did the damage, but in any event I'll be buying a new motherboard + CPU since the old gear (MSI 880GMA-E45, Phenom II x2 unlocked to x4 (a great value at the time)) is a bit dated. (And I'll be testing the power supply with a multimeter before using it again.)

I'm planning on switching to Intel for the superior single threaded performance (this system will mostly be used for gaming, with some digital audio workstation duties occasionally) so I was looking at the i5-4690K or the i7-4790K. I'm eyeballing the EVGA Z97 FTW mobo; the ability to connect a multimeter to the various power buses is interesting. DDR4 is intriguing, but it's just too expensive.

I was reading an article on xbitlabs which noted that the "improved" thermal interface material in the devil's canyon CPUs leaves something to be desired and they go on to state "Hardcore overclockers will still have to replace the processor's default thermal interface."

Really? Popping off the integrated heat spreader and applying your own thermal grease is a thing? I've never heard of that (then again I guess I'm not a "hardcore" overclocker) much less seen a guide for this sort of activity. This raises a few questions which I'm hoping the members here can answer:

1) Does this void the warranty? (Pretty sure it does but I have to ask...)

2) Can you just use the same kind of material you would on a heat spreader? (e.g. Arctic Silver)

3) Does the IHS just snap back on? Or do you have to somehow glue it back to the CPU package?

And a related question:

4) Would liquid cooling be more or less impacted by a deficient thermal interface than air cooling?

Thanks

 

[VirtualLarry]

Really? Popping off the integrated heat spreader and applying your own thermal grease is a thing? I've never heard of that (then again I guess I'm not a "hardcore" overclocker) much less seen a guide for this sort of activity. This raises a few questions which I'm hoping the members here can answer:

1) Does this void the warranty? (Pretty sure it does but I have to ask...)

2) Can you just use the same kind of material you would on a heat spreader? (e.g. Arctic Silver)

3) Does the IHS just snap back on? Or do you have to somehow glue it back to the CPU package?

And a related question:

4) Would liquid cooling be more or less impacted by a deficient thermal interface than air cooling?

Thanks

I think that this is one of those things, that if you have to ask, it's not for you.

 

[Idontcare]

Really? Popping off the integrated heat spreader and applying your own thermal grease is a thing? I've never heard of that (then again I guess I'm not a "hardcore" overclocker) much less seen a guide for this sort of activity. This raises a few questions which I'm hoping the members here can answer:

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

 

[DrMrLordX]

I think that this is one of those things, that if you have to ask, it's not for you.

Maybe, maybe not. Lots of people who would never consider delidding have stepped up to the plate and done it with Haswell and other Intel chips that do not have solder love.

OP, you can't get information much more complete than the post linked by Idontcare. tl;dr version: AS5 is not what you want under your IHS after a delid/relid operation. Do yourself a favor and just use CLU.

To address question #4, if you're going to go water (and by water I mean a custom loop, not an AiO unit) and you're contemplating a delid, you may as well just go bare die. Otherwise, well, it's like this: whatever your target temperature, it's going to be harder to hit that temperature if you have bad TIM (or bad gaps) under the IHS. In some ways this is harder on the air guys since they are limited in what they can do (better cooling solutions, such as custom water loops, still do better even with lousy TIM under the IHS), but in other ways this is harder on the water guys since their custom loop won't make as much of an impact as it would on a soldered chip. That being said, most water guys are just going to delid anyway, so the TIM is a hidden blessing in the form of an easier delid.

 

[OmNominous]

Thanks for the informative replies. I checked Idontcare's thread and looked at the youtube videos, which lead me to other delidding videos in which the die did not escape unscathed...

Going with a direct mount of the block onto the die sounds like the way to go, but having to remove material from the block's mounting posts so that the block will sit lower is beyond my comfort level (plus I don't even own a dremel), so I guess I'll leave it to those who are more hardcore than I. :biggrin:

 

[Tristor]

Honestly, the least extreme thing about doing direct-to-die cooling is modifying the socket retaining bracket. Also, the risk is very low of damaging your CPU if you use the vise/hammer method, rather than the razor blade method of delidding. The razor blade method is unnecessarily risky and more time consuming.

Idontcare's thread is an excellent source of information and kind of the definitive one-stop-shop for this sort of thing if you want to understand the results and the best methodology. If you're curious about the actual process of doing it, I took a lot of photos and some video when I was delidding my i5-4670k several months ago.

Album of first part of the delid
post-delid cleanup
lapping IHS
Video of lapping at 220 grit
Applying CLU and test-mount prep
Applying CLU video Part #1
Applying CLU video Part #2
Applying CLU video Part #3


Either way you decide to go, enjoy your new system. I would probably not recommend delidding to you though, since you haven't indicated you've done any previous modding projects. There's lots of other simple modifications you can do which are less involved or risky that are probably a good first step before jumping into something like delidding. That said, a lot of people who have done it were first time modders, as it's really not that difficult, just something that requires patience and care.

 

[Idontcare]

To echo Tristor's multiple excellent recommendations, if you should be interested in trying out the whole delidding thing then my recommendation is to buy a sacrificial throw-away CPU to use as a trial run.

I did this when I first stepped into the arena of lapping the CPU's IHS. Bought a dirt cheap $50 CPU to test it out, make sure I figured out what I was doing, and then did it a second time on my $1k CPU.

In your case, I'd suggest you buy the hardware you are interested in (mobo, 4790K CPU, etc) and also buy a cheapo $47 Haswell Celeron to use as a guinea pig to hone your delidding skills first before you attempt it with your 4790K.

 

[evangel76]

If you plan a mild overlocking like 4.5Ghz, you don 't have to delid ... With a devil canyon, you get easily about 4.5Ghz without any problem.
I always say that delidding is the Russian roulette ... do it at your own risk.

Francois

 

[PliotronX]

The only reason it came to the forefront is because the factories messed up either the bit the mates the heatspreader to the die, the TIM that is between them or both on the Haswell line of processors. I have been delidding since they were reintroduced with the Tualatin Pentium 3. Heatspreaders never offer better temps than directdie except with heatsinks using HDT. Kudos to Tristor on his post, Google Gold right there.

 

[MongGrel]

If you plan a mild overlocking like 4.5Ghz, you don 't have to delid ... With a devil canyon, you get easily about 4.5Ghz without any problem.
I always say that delidding is the Russian roulette ... do it at your own risk.

Francois

Some people have a lot of experience at it, but yeah.

It really doesn't seem that necessary with some of the newer ones.

 

[BonzaiDuck]

Some people have a lot of experience at it, but yeah.

It really doesn't seem that necessary with some of the newer ones.

My skepticism has grown in recent months.

Take, for instance, the i7-4790K. The processor is spec'd at a base speed of 4.0Ghz, and a turbo frequency of 4.4 ghz.

Many of the reports I've read seem to indicate an "OC'ing" range between 4.5 and 4.7. Looking at my old i7-2600K, the spec was 3.4 with 3.8 turbo, and I can clock it to 4.7 without exceeding either sane voltage or thermal limits. In fact, a thought much in my mind these days is whether I want to push the clock to 4.8, or make selections of "Per each core" to achieve 4.9 on one or two cores.

But it's all a matter of perception. The Devil's Canyon at 4.4 is likely to exceed my SB-K's performance at 4.7, since the i7-4770K @ 4.4 beats my SB-K by just a hair in Cinebench.

Do you want the Devils Canyon because it's a great 4-core processor? Or do you want the Devil's Canyon because it's capable of insane overclocks within temperature and voltage guidelines? I'd say yes to the first question, no to the second.

If one delids the processor, you can definitely lower the temperatures by maybe 15C -- with or without a direct-die approach. Certainly -- without the IHS. But whether this will mean better overclocking beyond what's currently being reported? Like I said: My skepticism has grown . . .

 

[aigomorla]

1) Does this void the warranty? (Pretty sure it does but I have to ask...)

2) Can you just use the same kind of material you would on a heat spreader? (e.g. Arctic Silver)

3) Does the IHS just snap back on? Or do you have to somehow glue it back to the CPU package?

And a related question:

4) Would liquid cooling be more or less impacted by a deficient thermal interface than air cooling?

Thanks

1. Most definitely voids warrenty. Even the Performance Overclocking protection Warranty you can purchase for Intel which protects you from Overclocking damages once.
They will not replace ANY delided processor period.

2. Yes, but if u got the IHS off, why would you want to put it back on?

3. You would need to use a special silicon to get it back on and attach it, but again, why would you want to put it back on when you got it off?
Your better of modifying the mount harness, and hard bolting the heat sink over the delided CPU.
But at that point also, your very limited in which heat sinks you can use, because too much torque on die, will CRACK it.

4. Depends on the Liquid Cooling.
I dont consider AIO's true liquid coolers. If your talking about a custom one, then yes. As they are way more efficient then Air coolers, if you built it.
If your talking about a AIO, like a Corsair/Thermaltake/Coolermaster for example... then no... The only way id run an AIO over a standard heat sink is if i was space limited where a good air sink would just not fit.

 

[Abwx]

With a devil canyon, you get easily about 4.5Ghz without any problem.


Francois

That s not true.

OT/

Felicitations quand meme, Francois, pour avoir reussi a decrocher le seul poste de troll officiellement authorisé a s ' exprimer chez Intel, c est dit sans la moindre animosité, bonne continuation a toi.

End of OT/

 

[BonzaiDuck]

1. Most definitely voids warrenty. Even the Performance Overclocking protection Warranty you can purchase for Intel which protects you from Overclocking damages once.
They will not replace ANY delided processor period.

2. Yes, but if u got the IHS off, why would you want to put it back on?

3. You would need to use a special silicon to get it back on and attach it, but again, why would you want to put it back on when you got it off?
Your better of modifying the mount harness, and hard bolting the heat sink over the delided CPU.
But at that point also, your very limited in which heat sinks you can use, because too much torque on die, will CRACK it.

4. Depends on the Liquid Cooling.
I dont consider AIO's true liquid coolers. If your talking about a custom one, then yes. As they are way more efficient then Air coolers, if you built it.
If your talking about a AIO, like a Corsair/Thermaltake/Coolermaster for example... then no... The only way id run an AIO over a standard heat sink is if i was space limited where a good air sink would just not fit.

Some folks had reported a 15C decrease in peak temperatures by using CLU as a thermal interface and reinstalling the IHS! Later, there seemed to be a consensus that the problem with IB and later processors was not so much the TIM, but the variation in distance between IHS and die with the new fabrication process. Either way -- keeping the IHS or direct-die cooling application -- the user would've solved that problem.

But especially for water-cooling, you'd think direct-die is the way to go. If I did it for a heatpipe cooler, I would want to hang my fans somewhere else.

Again -- more of my neophyte pontification: At least with the massive heatpipe coolers, they were designed to occupy known available space on the motherboard (or "volume" to be more accurate.) As I see it, more effective water cooling (better than AiO) means larger radiators and better water-blocks. In fact, I'd think mostly radiator capacity. But with that, you would spend more time and attention in picking the right case. Or matching case and radiators. That could be bigger trouble than choosing RAM with a low profile for a heatpipe cooler. Or -- giving up a PCI-E x1 slot.

But it's certainly the way to go if you see that AiO's are at most slightly better than the heatpipes.

 

[Abwx]

Everyrone is master of his own money but personaly i couldnt advise delidding, it s just not worth it given the 10% or so frequency improvements, as pointed by a member in the Haswell thread one must put thoses efforts in perspective, once more powerfull chips are released the delidding will appear for what it is, a useless and time wastefull operation.

 

[Phynaz]

That s not true.

OT/

Felicitations quand meme, Francois, pour avoir reussi a decrocher le seul poste de troll officiellement authorisé a s ' exprimer chez Intel, c est dit sans la moindre animosité, bonne continuation a toi.

End of OT/

Really? Who isn't getting about 4.5Ghz on DC?

BTW, I reported your post for attacking Francois. Would you ever call Warsam a troll?

 

[BonzaiDuck]

Really? Who isn't getting about 4.5Ghz on DC?

BTW, I reported your post for attacking Francois. Would you ever call Warsam a troll?

Without firsthand experience with the 4790K, I'd say it's highly likely one could get 4.5 Ghz, if the turbo spec is 4.4. But higher? Apparently, some do.

But if the limits are around 4.7 or so, why de-lid? Or -- why overclock? As someone said, why worry about the temperature if you'd be stuck with that sort of overclock anyway, even with lower temperatures?

You're guaranteed 4.4. I don't think a 300Mhz increase over that is worth a lot of modding and trouble. That's a "practical" observation, and I personally don't limit myself to the purely practical. But you have to ask yourself, either for the risk, the temperature, or an overclock that's still voltage-limited.

 

[Abwx]

Really? Who isn't getting about 4.5Ghz on DC?

BTW, I reported your post for attacking Francois. Would you ever call Warsam a troll?

I timely removed you from ignore list since i saw in someone else post that you answered me, so two things before we return to the usual way of dealing wich each other.

First, although 4.5 are possible that s not a rule, Hardware.fr specify that they do not change the overclocking possibilities compared to regular chips and that there s some luck involved in boths segments.

http://www.hardware.fr/focus/98/core-i7-4790k-test-devil-s-canyon-debarque.html

Second if you did understand french, wich i used willingly to not make pollution in this forum, i expressely said to Francois that it was said in a friendly way and i added that i wish him a good continuation, so i guess that you did find some pretense to report something else than what was actualy expressed by reading a single word out of a french written sentence.

If Francois feel offended by my way to congratulate him then i have no doubt that he will either report it himself to some moderator if he doesnt share my way of joking, next time use a better translator before jumping to your indeed twisted conclusions.

 

[DrMrLordX]

In your case, I'd suggest you buy the hardware you are interested in (mobo, 4790K CPU, etc) and also buy a cheapo $47 Haswell Celeron to use as a guinea pig to hone your delidding skills first before you attempt it with your 4790K.

Wouldn't it be possible to buy an older pre-Haswell Celeron for less as a delidding experiment? Or has the presence of the voltage regulators changed up how a delidding has to go?

 

[Idontcare]

Wouldn't it be possible to buy an older pre-Haswell Celeron for less as a delidding experiment? Or has the presence of the voltage regulators changed up how a delidding has to go?

Yes, of course, but I recommend using a CPU that you can pop into the mobo and test after delidding to verify you did not bork it in the process.

 

[DrMrLordX]

Hmm, that is a good point. No sense in delidding if you can't verify the quality of the work.

 

[Abwx]

You're guaranteed 4.4. I don't think a 300Mhz increase over that is worth a lot of modding and trouble. That's a "practical" observation, and I personally don't limit myself to the purely practical. But you have to ask yourself, either for the risk, the temperature, or an overclock that's still voltage-limited.

Hardware.fr managed to do 4.6 at 1.25V, they say that 1.3V is not safe on the long term, there s really nothing to gain with a few hundreds MHz set apart some CPUZ validations, power dissipation wise the chip should be at ease since its thermal density is about 0.5W/mm2, that s very safe value with theoricaly quite some room left even at 4.5, the limitation come from the voltage.

 

[BonzaiDuck]

Hmm, that is a good point. No sense in delidding if you can't verify the quality of the work.

It's sort of a quandary -- Like Morgan Freeman getting ready to open the cardboard box at the climax of "Se7en." "Oh-h-h-h . . . . I don't know . . . "

Some of the Devils Canyon offerings showed decent temperatures without delidding -- maybe it was Anandtech's tests. In other respects, you're not likely getting more than 4.7 Ghz out of those puppies, but you're guaranteed 4.4 with the stock turbo speed.

Ya gotta ask yourself . . . .

UPDATE -- THE BOTTOM LINE:

Here's the Anandtech comparisons for the i7-4790K and 4690K:

http://www.anandtech.com/show/8227/devils-canyon-review-intel-core-i7-4790k-and-i5-4690k/2

Scroll to the tabular results at the middle of the page.

The problem isn't just the temperatures: it's the voltage.

At attainable overclocks, the temperatures aren't all that daunting. Suppose you de-lidded a 4790K (or 4690K)? Various scenarios would give you temperature improvements -- 10C, 15C, maybe more if you didn't flip the IHS back into the equation.

That's not going to buy much in trying to get above 4.6 Ghz.

I'd say -- buy the 4790K, if you'll feel happy with the stock-turbo 4.4Ghz. Certainly, you could bump it up to 4.5. But you likely won't buy much by a de-lidding mod -- unless you kill the processor and have to "buy" a new one.

 

[Phynaz]

Without firsthand experience with the 4790K, I'd say it's highly likely one could get 4.5 Ghz, if the turbo spec is 4.4. But higher? Apparently, some do.

Exactly, some people don't know what "around" means.

 

[Idontcare]

It's sort of a quandary -- Like Morgan Freeman getting ready to open the cardboard box at the climax of "Se7en." "Oh-h-h-h . . . . I don't know . . . "

Some of the Devils Canyon offerings showed decent temperatures without delidding -- maybe it was Anandtech's tests. In other respects, you're not likely getting more than 4.7 Ghz out of those puppies, but you're guaranteed 4.4 with the stock turbo speed.

Ya gotta ask yourself . . . .

UPDATE -- THE BOTTOM LINE:

Here's the Anandtech comparisons for the i7-4790K and 4690K:

http://www.anandtech.com/show/8227/devils-canyon-review-intel-core-i7-4790k-and-i5-4690k/2

Scroll to the tabular results at the middle of the page.

The problem isn't just the temperatures: it's the voltage.

At attainable overclocks, the temperatures aren't all that daunting. Suppose you de-lidded a 4790K (or 4690K)? Various scenarios would give you temperature improvements -- 10C, 15C, maybe more if you didn't flip the IHS back into the equation.

That's not going to buy much in trying to get above 4.6 Ghz.

I'd say -- buy the 4790K, if you'll feel happy with the stock-turbo 4.4Ghz. Certainly, you could bump it up to 4.5. But you likely won't buy much by a de-lidding mod -- unless you kill the processor and have to "buy" a new one.

With Devil's Canyon, I'm of the opinion that unless you are psychologically hung-up on getting that last 100MHz out of your OC, it simply isn't worth your time/effort/expense.

For me personally, I would be psychologically hung-up on getting that last 100MHz if say, for example, I had a Devil's Canyon processor that could get to 4.9GHz without delidding.

Knowing my own personal "trigger points", I'm sure in that scenario I would likely delid my CPU just to get another 100MHz and make my rig turn in that nice satisfying "5GHz" overclock number.

But I wouldn't do it if I was already at 5GHz, not personally rewarding enough to do it just to hit 5.1GHz, or if my CPU was well south of the big 5.0 number, at say 4.6 or 4.7GHz. My brain would not be tickled pink just because I pushed my 4.7GHz chip to 4.8GHz with delidding, not worth the time and effort at that point.

But everyone has their own risk/reward profile and their own threshold trigger points when it comes to psychological reward in OC'ing. So I know my preferences are not likely to be representative of the larger audience.