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

Page 39 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

taq8ojh

Golden Member
Mar 2, 2013
1,296
1
81
Right now I am at 4.4GHz, vcore averages at 1,150V (as reported by Hwinfo64), and hottest core says 75°C. The heatsink is just warm (not hot at all) though, so I assume there is room lots of improvement IF ONLY the pressure on the IHS was higher.
Of course, maybe it's the new Noctua fan at work here and is much better than I thought. It's spinning at just 1100RPM. I do have PWM control on btw, it goes down to 500 or so in idle. The best part is it's absolutely quiet at 1100, I really cannot hear anything.
 
Last edited:

Ed1

Senior member
Jan 8, 2001
453
18
81
Right now I am at 4.4GHz, vcore averages at 1,150V (as reported by Hwinfo64), and hottest core says 75°C. The heatsink is just warm (not hot at all) though, so I assume there is room lots of improvement IF ONLY the pressure on the IHS was higher.
Of course, maybe it's the new Noctua fan at work here and is much better than I thought. It's spinning at just 1100RPM. I do have PWM control on btw, it goes down to 500 or so in idle. The best part is it's absolutely quiet at 1100, I really cannot hear anything.
Ok, then it looks very good to me for 4.4ghz .
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
20,839
3,174
126
guys make sure the sink your attaching to is also flat to begin with.

A lot of heat sink manufactors as well as waterblock makers BOW the blocks a little.

This causes a focus point on the die and it can be very bad.
Also if the sink is bowed u can get uneven temps because of the uneven surface.

Allign the sink next to a piece of grid paper and inspect your sink is flat and not bowed.
A Flat sink will show the grids as stright in the reflection.
A Bowed sink will show curves on the reflection.

Always Grid test b4 u put any sink on a naked die.
 

Ed1

Senior member
Jan 8, 2001
453
18
81
guys make sure the sink your attaching to is also flat to begin with.

A lot of heat sink manufactors as well as waterblock makers BOW the blocks a little.

This causes a focus point on the die and it can be very bad.
Also if the sink is bowed u can get uneven temps because of the uneven surface.

Allign the sink next to a piece of grid paper and inspect your sink is flat and not bowed.
A Flat sink will show the grids as stright in the reflection.
A Bowed sink will show curves on the reflection.

Always Grid test b4 u put any sink on a naked die.

that can be true on HS that have block type bottoms but his 212evo is a direct contact heat-pipe type sink . I don't think it be good to try and lap them .
lapping the core/IHS might help a little bit if its off a lot .
 

Garzhad

Junior Member
Jun 1, 2013
6
0
0
Has anyone done this and used a high-end air cooler like the Phanteks TC14PE?
http://www.newegg.com/Product/Produc...82E16835709001
I've got that and an i5 3570k that i'm contemplating delidding after I get my system built.
Probably going to at least lap the IHS top and the HSF(I think it's the kind of HSF you can lap, i don't recall seeing anything about it being direct touch?) But I really don't see any feasible way to lap the bottom of the IHS properly.

I'll most likely try the vice-and-hammer method of delidding; grandfather has one in his garage, just have to lay down some blankets in case something goes wrong and it falls, wouldn't want the die to hit the concrete floor and crack.

That said, I know IDC and most of the others just lock the unsecured IHS down with the retention bracket, but has anyone tried gluing it back down afterwards and recorded what, if any difference in temps, the re-glued IHS makes compared to the free-floating one, and what material they used?

I would figure that as long as you made sure the adhesive didn't lift the IHS off the die before it hardened it wouldn't make a difference, but would help keep the IHS corners from bending around the die from the pressure exerted by the HSF mounting bracket.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Hi Garzhad, welcome to the forums! :thumbsup:

Has anyone done this and used a high-end air cooler like the Phanteks TC14PE?
http://www.newegg.com/Product/Produc...82E16835709001
I've got that and an i5 3570k that i'm contemplating delidding after I get my system built.

I don't recall if anyone here has reported their delidding results with that specific HSF, but it is in the same caliber as a NH-D14 and results with that high-end air cooler are available throughout this, and other, threads in the forum.

I'll gladly link you to specific posts of interest if that is something you are interested in seeing. But if you are specifically looking for results with the TC14PE then we'll just have to hang out a bit and see if another member chimes in with some links or results.

Probably going to at least lap the IHS top and the HSF(I think it's the kind of HSF you can lap, i don't recall seeing anything about it being direct touch?) But I really don't see any feasible way to lap the bottom of the IHS properly.

People have reported success in lapping the inside of their IHS by taking a tiny stick of wood (0.25" on edge) and covering the end with sand paper and then swishing it back and forth to remove the rough surface and get down to copper.

But it doesn't appear to deliver any sort of benefit. You get 80% of the benefit of delidding just from eliminating the gap, another 20% can be had in picking a great TIM like Liquid Ultra. Getting beyond that, lapping the inside of the IHS, starts giving you single-digit returns but at the expense of out-sized levels of effort on your part.

Personally I never bothered to lap the inside of my IHS. For two reasons - the effort did not seem justified based on reports from people who had done it. And secondly I am actually not all that keen about putting copper directly on the bare silicon die. Yes I did it for my tests, but that was done in the name of generating data and not done in the name of creating a 24x7 machine capable of functioning for 3-4 yrs.

Copper on bare silicon is a recipe for early chip death. So you have to weigh that added risk when deciding what is best for you. OC'ing is not without risk, adding voltage and delidding is not without risk. So in the grand scheme of things the risk-adder of putting copper on your die may not be all that much incremental risk.

That said, I know IDC and most of the others just lock the unsecured IHS down with the retention bracket, but has anyone tried gluing it back down afterwards and recorded what, if any difference in temps, the re-glued IHS makes compared to the free-floating one, and what material they used?

I would figure that as long as you made sure the adhesive didn't lift the IHS off the die before it hardened it wouldn't make a difference, but would help keep the IHS corners from bending around the die from the pressure exerted by the HSF mounting bracket.

Having played around with delidded Ivy Bridge CPUs for a while (three 3770k's so far) I personally have not had any reason to glue the IHS back down onto the delidded processors. Once they are latched into the socket retention mechanism they are truly going nowhere.

But if I were truly intent on gluing it back down then I would use the socket retention mechanism to my advantage in pressing the IHS down with the needed downforce to eliminate as much of the gap as possible.

So what I would do is prepare the CPU and IHS as a delidded CPU but I would add a bead of adhesive to the IHS lip before placing the IHS back down onto the CPU PCB (while the CPU is in the socket). Then I'd lock the socket clip, pushing that IHS down close to the PCB right where I want it. Then I'd mount my HSF to add even more down pressure.

Then I'll know the gap is as minimal as possible, and whenever I do remove the HSF and take the CPU out of the socket then the IHS will be stuck fast to the PCB.

Generic superglue is probably good enough for the task IMO.
 

gockelmax

Junior Member
May 11, 2013
2
0
0
Has anyone already heard anything about a delidded Haswell?
If not, IDC there may be a new project for you^^

By the way I love this thread, it truely is a fountain of knowledge!
 

Garzhad

Junior Member
Jun 1, 2013
6
0
0
Hi Garzhad, welcome to the forums! :thumbsup:

But it doesn't appear to deliver any sort of benefit. You get 80% of the benefit of delidding just from eliminating the gap, another 20% can be had in picking a great TIM like Liquid Ultra. Getting beyond that, lapping the inside of the IHS, starts giving you single-digit returns but at the expense of out-sized levels of effort on your part.

Personally I never bothered to lap the inside of my IHS. For two reasons - the effort did not seem justified based on reports from people who had done it. And secondly I am actually not all that keen about putting copper directly on the bare silicon die. Yes I did it for my tests, but that was done in the name of generating data and not done in the name of creating a 24x7 machine capable of functioning for 3-4 yrs.

Copper on bare silicon is a recipe for early chip death. So you have to weigh that added risk when deciding what is best for you. OC'ing is not without risk, adding voltage and delidding is not without risk. So in the grand scheme of things the risk-adder of putting copper on your die may not be all that much incremental risk.

Alright, if it Is a recipe for early chip death, for next to nothing, then i'm definitely not going to bother. I'm alright with replacing them now if I botch the delid, but I don't want to have to try and get another one until i'm ready to replace everything again in a few years.
Does this mean all those folks using pure copper watercooler blocks on the bare die are courting early chip death as well?

IRC the IHS top and HSF lapped are good for a few C's without any risk whatsoever, really, so will probably just stick to that.

Looking forward to seeing how it works with the TC14PE. It's supposed to be quite a few C's cooler then the NH-d14. Supposedly The best air-cooler when tested using identical fans.

Edit: Was just thinking, I know You've done a fair bit of testing on it, but does anyone know what the Best 'fire-and-forget' TIM is for TIM1 between the die and IHS? From what i've seen of your testing i'm going to stay away from the diamond stuff there(don't want any scratchin') You've showed alot of good things about the CL Ultra, even if it is a PITA to remove later on. Anyone have experiences with CL Pro or the Indigo stuff? Think i've read that those two tend to... fuse... stuff together, rather permanently. So if they have that pump-out effect or lose effectiveness over time in that application, you are basically stuck with it?
 
Last edited:

24HZ

Member
May 25, 2013
52
0
0
Im really looking forward to the Haswell delid data before I buy one. Bonus points for direct die.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Does this mean all those folks using pure copper watercooler blocks on the bare die are courting early chip death as well?

Yes they are.

There is a passivation layer on the bare die, it isn't just a silicon surface we are looking at when we delid, so they have some time before things get bad. But any kind of scratch means Russian roulette, and even without a scratch they still have Fick's law of diffusion working against them.

I did not mind doing it, but there is a difference between making a calculated risk versus being completely unaware of the risk one is taking with their actions.

Edit: Was just thinking, I know You've done a fair bit of testing on it, but does anyone know what the Best 'fire-and-forget' TIM is for TIM1 between the die and IHS? From what i've seen of your testing i'm going to stay away from the diamond stuff there(don't want any scratchin') You've showed alot of good things about the CL Ultra, even if it is a PITA to remove later on. Anyone have experiences with CL Pro or the Indigo stuff? Think i've read that those two tend to... fuse... stuff together, rather permanently. So if they have that pump-out effect or lose effectiveness over time in that application, you are basically stuck with it?

Tw33k made a great thread comparing Liquid Ultra versus Liquid Pro which shows Liquid Ultra is the best of the best, including email communications with the president regarding thermal conductivity values of the two leading TIMs.

To my knowledge no one has attempted to make Indigo Xtreme work for this application. You may not be aware but IX is a reflow-solder, so you have to clip the solder package to fit on top of the die, and even then it will be a big lump of solid solder until it heats up enough to melt.

That is just a recipe for cracking your die as the IHS pushes down and concentrates all that down force onto the die through the narrow bead of IX solder :(

So I'd say don't try it unless you are adventuresome, just go with Liquid Ultra and be happy.
 

Garzhad

Junior Member
Jun 1, 2013
6
0
0
Yes they are.

There is a passivation layer on the bare die, it isn't just a silicon surface we are looking at when we delid, so they have some time before things get bad. But any kind of scratch means Russian roulette, and even without a scratch they still have Fick's law of diffusion working against them.

I did not mind doing it, but there is a difference between making a calculated risk versus being completely unaware of the risk one is taking with their actions.



Tw33k made a great thread comparing Liquid Ultra versus Liquid Pro which shows Liquid Ultra is the best of the best, including email communications with the president regarding thermal conductivity values of the two leading TIMs.

To my knowledge no one has attempted to make Indigo Xtreme work for this application. You may not be aware but IX is a reflow-solder, so you have to clip the solder package to fit on top of the die, and even then it will be a big lump of solid solder until it heats up enough to melt.

That is just a recipe for cracking your die as the IHS pushes down and concentrates all that down force onto the die through the narrow bead of IX solder :(

So I'd say don't try it unless you are adventuresome, just go with Liquid Ultra and be happy.

So in other words, to 'play it safe' with bare die applications, ideally you would use a nickle-plated water block, instead? Wonder how that would have effected your temps. From what I remember from your tests, there was only a 4C or so difference between your bare die and IHS temps; wonder how much more that gap would close having to use the nickle-plated block.

Thanks for the link. You said any kind of scratch means Russian roulette... could it perhaps be that this is what killed your one 3770k during your TIM testing? Since it looked like the IC diamond put lots of micro scratches on it? At any rate, it's another possible explanation besides EDS.

I hadn't originally planned to mess with the indigo anyway, just curious if anyone else had. While the Ultra is the best for TIM1, is this also the case for TIM2? Finding decent TIM comparisons on the web is sorta hit or miss, from what i've seen.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Thanks for the link. You said any kind of scratch means Russian roulette... could it perhaps be that this is what killed your one 3770k during your TIM testing? Since it looked like the IC diamond put lots of micro scratches on it? At any rate, it's another possible explanation besides EDS.

I suspect that is what killed my chip. It was either the copper gaining access to fast-diffusion pathways provided by the scratches, or it was the gallium from the Liquid Ultra doing the same thing, or it was from ESD.

Since I have no way of knowing for sure, I'd rather chalk it up to my stupidity (static discharge) rather than create fear and suspicion regarding the TIM. But I naturally have my concerns.

I hadn't originally planned to mess with the indigo anyway, just curious if anyone else had. While the Ultra is the best for TIM1, is this also the case for TIM2? Finding decent TIM comparisons on the web is sorta hit or miss, from what i've seen.

Indigo Xtreme is bar-none the best from purely a thermal conductivity perspective. The only reason we don't try it on bare-die is because of the practical limitations that crop up with the reflow aspects.

As such, IMO the best combo you could do is Liquid Ultra under the IHS and IX on top of the IHS.

Second best combo would be Liquid Ultra on both sides of the IHS.
 

Garzhad

Junior Member
Jun 1, 2013
6
0
0
I suspect that is what killed my chip. It was either the copper gaining access to fast-diffusion pathways provided by the scratches, or it was the gallium from the Liquid Ultra doing the same thing, or it was from ESD.

Since I have no way of knowing for sure, I'd rather chalk it up to my stupidity (static discharge) rather than create fear and suspicion regarding the TIM. But I naturally have my concerns.
Whether it was the copper, the gallium or simple static, moral of the story is don't scratch the die and keep yourself grounded and you shouldn't have any problems, I suppose.
I think you said yourself that the second you saw those scratches, you basically knew the die's days were numbered. Crack propagation, or something? If anything, perhaps it serves as a cautionary tale regarding the use of mildly abrasive diamond-based TIM's on the delicate die.


Indigo Xtreme is bar-none the best from purely a thermal conductivity perspective. The only reason we don't try it on bare-die is because of the practical limitations that crop up with the reflow aspects.

As such, IMO the best combo you could do is Liquid Ultra under the IHS and IX on top of the IHS.

Second best combo would be Liquid Ultra on both sides of the IHS.

Since i'd be getting it for Under the IHS already anyway, probably just stick to the Ultra and not have to spend more just to get some indigo. It also seems easier. No need to mess around with 'heating it' or any of the other specific steps you need to follow to get the IE to work properly.
It'll be something I do gradually and in steps, though, once all the parts arrive. Going to just put everything together as is, using the paste the phanteks came with. Test that for awhile. Then lap the HSF, reapply, see how much of a dif that makes. Who knows. Might have a golden chip. Not going to lap the IHS until I delid though. Seems many times easier to lap it once it's removed from the PCB, no need for any tape or concerns about getting stuff on it.
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Does the Liquid Ultra screw up the Intel markings when used on top of the IHS?

Yes it does. If you are worried about the Intel markings on the IHS then you need to stay away from all the liquid metal TIMs as well as IC Diamond.
 

Yuriman

Diamond Member
Jun 25, 2004
5,530
141
106
Too bad there's no data.

Looks like you'll need to be a lot more careful slipping a razor under Haswell's IHS than with Ivy.

fe5mny.jpg
 

Dahak

Diamond Member
Mar 2, 2000
3,752
25
91
Seeing as it is possible to de-lid it, Idontcare do you thing you will do some similar test as you did with IB with a Haswell chip?
 

24HZ

Member
May 25, 2013
52
0
0
Indirectly from Xtremesystems and not very scientific.

MdtwLVK.jpg


Supposedly thats on an Antec Kuhler H2O 620 closed loop cooler (120mm x 1). No idea if its direct die and what TIM is being used but you can bet its top shelf and more likely direct die.

Quick take aways

1)Haswell is still hotter than IVB delidded
2)Possible Haswell needs slightly more volts for the same freq compared to IVB
3)Delidding Haswell still nets big thermal gains a la IVB
 

Garzhad

Junior Member
Jun 1, 2013
6
0
0
Yes it does. If you are worried about the Intel markings on the IHS then you need to stay away from all the liquid metal TIMs as well as IC Diamond.
If he's worried about the intel markings he probably shouldn't be in a thread dedicated to the warranty-voiding delidding process :D
Once you've popped your lid its over anyway so what does it matter if the markings are intact or not.
 

Garzhad

Junior Member
Jun 1, 2013
6
0
0
Looks like you'll need to be a lot more careful slipping a razor under Haswell's IHS than with Ivy.
Or find a vice and piece of 2x4. Considering the closeness, that's probably going to be the preferred method over razors with Haswell.