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

[Denithor]

Three things.

First, I'm a chemist who works with silicone emulsions every day at work. And I don't think these TIMs are emulsions by the typical definition. An emulsion is defined as one liquid dispersed in another. In my work this is usually a silicone fluid of some kind that we disperse into water by various methods. Most TIMs on the other hand are a solid of some kind (silver, diamond particles, etc) dispersed into a liquid (silicone for the most part).

Second, have we confirmed that the liquid-metal TIMs actually bond/adhere to the metal surfaces they contact? And how about the silicon die? Because if they adhere to the metals but not to the silicon the hot/cold cycle mentioned above could do bad things to them (whole thing is fixed to IHS, swells under load, cracks the die it's not bonded to).

Third, guess you (IDC) haven't had time to start evaluating other TIMs over/under the IHS? As a corollary, if I could get some 'interesting' compounds from work would you be interested in evaluating for thermal transfer properties? I know we have magnesium stearate, titanium dioxide and several other things milled to very low particle size in silicone fluids...?

 

[BonzaiDuck]

Three things.

First, I'm a chemist who works with silicone emulsions every day at work. And I don't think these TIMs are emulsions by the typical definition. An emulsion is defined as one liquid dispersed in another. In my work this is usually a silicone fluid of some kind that we disperse into water by various methods. Most TIMs on the other hand are a solid of some kind (silver, diamond particles, etc) dispersed into a liquid (silicone for the most part).

Second, have we confirmed that the liquid-metal TIMs actually bond/adhere to the metal surfaces they contact? And how about the silicon die? Because if they adhere to the metals but not to the silicon the hot/cold cycle mentioned above could do bad things to them (whole thing is fixed to IHS, swells under load, cracks the die it's not bonded to).

Third, guess you (IDC) haven't had time to start evaluating other TIMs over/under the IHS? As a corollary, if I could get some 'interesting' compounds from work would you be interested in evaluating for thermal transfer properties? I know we have magnesium stearate, titanium dioxide and several other things milled to very low particle size in silicone fluids...?

Can't remember if I heard it directly from Innovative Cooling techs or from their web-page. Can't be sure if ICD doesn't use silicone grease, but it was described as "oils which dry out" a bit over time.

 

[Rvenger]

Seems that many over at OCN are replacing the stock TIM with Liquid Metal Pro. I haven't heard of any issues with it yet.

 

[Idontcare]

I figured that if you spent the $25-bucks, you're going to try the diamond stuff? One wonders if that would leave a stable result . . .

Yeah I have the IC Diamond stuff here. Not having any luck getting the shim to work out though. Going to have to ditch the 0.20mm metal shim and pursue something else. Still brainstorming. If anyone has any ideas then lets hear'em. I'm stumped at the moment.

Three things.

First, I'm a chemist who works with silicone emulsions every day at work. And I don't think these TIMs are emulsions by the typical definition. An emulsion is defined as one liquid dispersed in another. In my work this is usually a silicone fluid of some kind that we disperse into water by various methods. Most TIMs on the other hand are a solid of some kind (silver, diamond particles, etc) dispersed into a liquid (silicone for the most part).

Yeah I was using "emulsion" when really I should have been saying "suspension". Given the solid/liquid mixtures involved, I suppose we can specify they are colloids.

Second, have we confirmed that the liquid-metal TIMs actually bond/adhere to the metal surfaces they contact? And how about the silicon die? Because if they adhere to the metals but not to the silicon the hot/cold cycle mentioned above could do bad things to them (whole thing is fixed to IHS, swells under load, cracks the die it's not bonded to).

Regarding the liquid-metal TIMs...they bond all too well to metal. When I used Indigo Xtreme I had to relap my IHS and HSF to remove all of it.

I don't know how well it adheres to silicon nitride, but my suspicion is that it will adhere as well as the metal solder Intel uses on Sandy Bridge. Whether or not that is considered to be good adhesion is a good question.

This would all be a lot easier to understand if Intel would just tell us enthusiasts the straight answer on why they opted to not do solder on IB and why they opted for the specific TIM they are currently using. Then we'd know what demons we are dancing with, what risks we are unknowingly accepting in these tests.

Of course that could also be said of overvolting as well. Intel has all that data, they know exactly how quickly their chips fall apart when overvolted (it is standard data generated during burn-in for a product qual). If only they'd share some of that data with the community then we'd all be better informed overclockers, taking less risk by being less ignorant of the specific risks we are taking when overvolting to any given level.

Third, guess you (IDC) haven't had time to start evaluating other TIMs over/under the IHS? As a corollary, if I could get some 'interesting' compounds from work would you be interested in evaluating for thermal transfer properties? I know we have magnesium stearate, titanium dioxide and several other things milled to very low particle size in silicone fluids...?

I haven't started any of those tests yet as I am still tinkering with the situation as it relates to attempting to control the height of the gap between the CPU and the IHS.

Not that that conundrum must be resolved before I can move on, its just that it would have been nice to get that data point generated so we could definitely say something about the quality of the TIM Intel is using under the CPU.

For all I know right now, the stock CPU TIM could be the best stuff on earth but the gap is so large (0.06mm) that the thermal conductivity is compromised and the gap alone is the reason for the poor temperatures of IB.

I'm intrigued by your offer to test some fun lab stuff. Are those materials expected to yield good thermal conductivity though? I get why they use diamond in the IC Diamond stuff, diamond itself has a high thermal conductivity.

But magnesium stearate? (common binder used in medication tablets) I would not have considered that to be a good thermal conductor. But I really don't know, that's just a guess on my part. What are your expectations?

 

[Idontcare]

Seems that many over at OCN are replacing the stock TIM with Liquid Metal Pro. I haven't heard of any issues with it yet.

Any particular reason for using the Pro versus the Ultra?

I was going to order some and it looked like the Ultra was the way to go (2x higher thermal conductivity). But all I ever hear about is people using the Pro for these delidded Ivy Bridges.

 

[BonzaiDuck]

Any particular reason for using the Pro versus the Ultra?

I was going to order some and it looked like the Ultra was the way to go (2x higher thermal conductivity). But all I ever hear about is people using the Pro for these delidded Ivy Bridges.

I think I'd seen reference to the Ultra product in forum posts elsewhere picked up in web-searches I made -- quite possibly in regard to the IB TIM replacement. From whatever I read of product descriptions, I came away with the notion that Ultra was supposed to be a thicker application. Somehow, I made the tentative decision -- in case I ever build an Ivy Bridge system -- to use the Ultra for giving the IHS and TIM a makeover.

Also, I recall enough about the Indigo product to imagine that it softens or partially liquifies with temperature, although that temperature must reach about 80C. My earliest interest in this feature arose for the need to heat the HSF assembly with a hair-dryer, or make it operate at a level generating the appropriate temperature. So you'd expect that it would be easier to get the die itself to reach those same temperatures, or that any partial change of state would occur more easily at the die.

How would such things affect our concerns about a scenario such as Denithor described?

I have to say -- I'm not spending money at the moment for practical reasons. But I'm interested in this project. It has me continually checking back to this thread. I already have plans to disassemble an older machine, use some parts to upgrade my brother's computer and the remainder to complement an IB, Z77 mobo and RAM.

Like you said, and no less for Intel voltage specs, there are things we'd like to know. A good assumption can be that Intel already knows them, but can we be absolutely sure? We already agree that they didn't get at least two things straight about the old Q6600 C2Q: the TjMax and the voltage specs which (I think) you said had been revised long after we bought our chips.

 

[Rvenger]

Any particular reason for using the Pro versus the Ultra?

I was going to order some and it looked like the Ultra was the way to go (2x higher thermal conductivity). But all I ever hear about is people using the Pro for these delidded Ivy Bridges.

Pro is cheaper and probably doesn't make much of a difference in thermals I guess. Not really sure why. I would get the ultra for $2 more. It may get you 2c better thermals .

 

[BonzaiDuck]

Pro is cheaper and probably doesn't make much of a difference in thermals I guess. Not really sure why. I would get the ultra for $2 more. It may get you 2c better thermals .

Did you find any benchmark comparisons of the two? I haven't looked at these things for a while. One forum post elsewhere showing a demonstration of Pro on the IB de-lidding was boasting a 28C+ improvement in temperatures, or so I recall. First, the poster cited a 14C+ improvement with AS5, and then replaced it with the Pro. the quality and therefore credibility of those posts is a stark contrast with what IDontCare has done so far.

 

[Ferzerp]

I don't discount it, I believe it. What I take from this is that NT-H1 is simply incompatible with the thermomechanical environment that exists in direct-die applications.

For example we know the mismatch in coefficients of thermal expansion will be much larger between that of the silicon Die and the metal IHS versus that which exists between a metal IHS and the metal base of an attached HSF/water-block.

Additionally we know the wettability of the silicon die is markedly different than that of the metal surfaces for which NT-H1 was developed to adhere to. This makes issues of viscosity and surface adhesion a factor when matters of thermomechanical forces are a concern, which they are.

So what does this boil down to in terms of why NT-H1 could be giving rise to these results? Two things can be happening, and possibly in combination.

One is the push-pull pumping effect that comes with the mechanical expansion and contraction of the silicon, the TIM, and the overhead IHS. Eventually, given time, the TIM itself will be squeezed out of the gap that it occupies between the die and the IHS and an air gap will be left in its place. The air gap most likely forms when the CPU cools down.

Then when the CPU heats back up, the air gap prevents the IHS from heating up in concert with the die and the TIM, resulting in an IHS that is not expanding to fill in the air gap, leaving the CPU much hotter than before.

The second point of concern is that of emulsion separation - the molecular components that comprise NT-H1 may experience separation at the behest of the silicon die interaction (it will actually be a silicon nitride at the surface). NT-H1 was probably not designed to remain an emulsion when in contact with silicon as it was designed to remain an emulsion when in contact with copper, aluminum, or nickle (more specifically, the metal oxides of those elements).

So a chemical separation may be the culprit as well, and likely it is a combination of both.

If it is just chemical separation then choosing a different TIM will be the solution.

If it is is thermomechanical push-pump that is the culprit then going with the more permanent metal-TIMs (Indigo Xtreme or Liquid Ultra) would be the solution.

In either case, going with the more permanent metal-TIMs would appear to be the one-size-fits-all solution provided the thermomechanical stresses that will then arise from having the mismatch in coefficients thermal expansion do not cause issues for the IB packaging itself (ala bumpgate dejavu).

Right now I have AS5 on, and I'll leave it long enough to see if the same thing happens with it. Otherwise, if others have good luck with the liquid metals, I'll try it. I was an early de-lidder, but I have so far shied away from something so runny and conductive. I'm not the best at applying things..

 

[Idontcare]

Pro is cheaper and probably doesn't make much of a difference in thermals I guess. Not really sure why. I would get the ultra for $2 more. It may get you 2c better thermals .

That is what I am thinking.

For all the effort and risk one goes to in delidding their $340 CPU, why look to save a couple dollars on the replacement CPU TIM if there is a readily available alternative that is designed to provide superior results?

I'm no expert on TIMs, so I asume there must be reason why people are avoiding the Ultra? If it is just lack of supply in-hand then I'd understand. For example there is a Ceramique 2 out but I have a bunch of the original Ceramique in-hand, so it would be natural for me to test Ceramique versus going to the expense of buying Ceramique 2.

In other words maybe people are using the Pro simply because that is what they already have?

Also, I recall enough about the Indigo product to imagine that it softens or partially liquifies with temperature, although that temperature must reach about 80C. My earliest interest in this feature arose for the need to heat the HSF assembly with a hair-dryer, or make it operate at a level generating the appropriate temperature. So you'd expect that it would be easier to get the die itself to reach those same temperatures, or that any partial change of state would occur more easily at the die.

How would such things affect our concerns about a scenario such as Denithor described?

It would be easier to get IX to melt on the die. The challenge is dealing with the placement and location of the packet prior to melting.

They are designed with a self-contained plastic enclosure that dissolves into the metal during the curing process.

I've got to imagine that getting the IX to setup correctly on the IB die will take some finesse. I'd have to think about it for a bit before I attempted it.

But I'm interested in this project. It has me continually checking back to this thread.

Just a heads-up then to not waste your time checking this thread for the next week, I'm headed off to Ocean City MD for a week at the beach with the family

A good assumption can be that Intel already knows them, but can we be absolutely sure? We already agree that they didn't get at least two things straight about the old Q6600 C2Q: the TjMax and the voltage specs which (I think) you said had been revised long after we bought our chips.

Ah, this is where I have the unfair advanage of having worked on the other side of the fence in this industry, so it is not a matter of guessing or assuming

It is industry-wide standard practice to generate what is called "burn-in" data prior to releasing a new node to production and prior to releasing new products to market.

Burn-in entails accelerated lifetime testing, which is done by way of intentionally operating the IC's at elevated temperatures and elevated voltages.

This aspect of burn-in is not to be confused with the form of burn-in that people think of when you assemble a system and leave it running in a standard operating environment at say DELL or HP or in the home of the DIYer.

The data which comes from burn-in is what is then used to define maximum operating specs as well as to confirm that the minimum lifetime characteristics of the new products and node conform to internal expectations. Weibull plots and the like.

So there is no question that Intel has the data, the data is generated for every IC produced at every fab in the world as standard practice.

But why would Intel revise their numbers? Because the data themselves are not taken from a static parent population. The qualities of the node improves in time, as do the confidence limits on the sampled data themselves. In time this leads to a more quantitative idea of exactly what the weibull plot looks like (the whiskers get smaller) while at the same time the entire line moves to the right (longer intrinsic lifetimes at any given temperature and voltage).

Most companies don't bother to revise their external specs as this internally generated data comes to be analyzed, they just keep it to themselves and leave the specs be what they are. Intel raising their published specs just means they felt they had so overwhelmingly sandbagged their initial specs that it was prudent and good business for their customers (not us, the OEMs and so forth) to revise those published specs on the basis of the newfound confidence in the quality of the parent distribution.

 

[Idontcare]

Right now I have AS5 on, and I'll leave it long enough to see if the same thing happens with it. Otherwise, if others have good luck with the liquid metals, I'll try it. I was an early de-lidder, but I have so far shied away from something so runny and conductive. I'm not the best at applying things..

Yeah you definitely want to go with the "a little bit goes a long ways" approach. Also realize that once the liquid metal sets up, you won't be removing the IHS ever again without serious risk of pulling the CPU silicon off of the PCB package.

But do not be so concerned with the electrical conductivity aspects. The entire die itself is electrically sealed. What you don't want to do is short out the 3 rows of pads that are present on the top of the PCB. You must take care to not have the IHS resting on them as well.

And obviously you don't want so much liquid metal TIM as to have it spill over the side and down into the socket.

But I think it would be quite easy to limit the volume of liquid metal TIM to avoid spillage to both areas. I spend far and away more time making sure the IHS doesn't sit on top of the landing pads than I spend worrying about anything else when I reseat the IHS into the socket.

 

[Rvenger]

I have a hot running 3770k so I may try the liquid metal too. I guess if I fail I can just purchase a 3570k and call it a day.

 

[Idontcare]

I have a hot running 3770k so I may try the liquid metal too. I guess if I fail I can just purchase a 3570k and call it a day.

Heh, I'm kinda in a similar mindset. If I blow up my 3770k then I'll just go back to using my faster, albeit higher power-consuming, 5GHz 2600k.

Barring some miracle at the hands of liquid metal TIMs and direct-die contact with the H100 water block, 5GHz is not in the cards for my 3770k.

I'm not entirely happy with the Vcc needed for 4.9GHz stability either :\

If it weren't for my 2600k being the fallback plan then I probably wouldn't be willing to push my 3770k so hard.

It is really mind blowing, to be frank, that Ivy Bridge is so limited in upper clockspeed. I am referring to the electrical parametrics in this case.

NMOS and PMOS drive currents were substantially improved with 22nm over 32nm, and yet the clockspeed max (fmax) just sucks, didn't move at all.

The lower power consumption is great, that is the benefit of smaller xtors which have less capacitance and the benefit of 3D xtors giving better leakage per micron results. But clockspeed should have improved commensurate with drive currents improving and yet we don't see that in practice

From a device-physics position, I'm at a loss at the moment to satisfactorily explain the clock-limited aspects of my 3770k. I should have realized at least a 10% increase in clocks over my 2600k (5.5GHz operation at same Vcc) and yet that isn't the reality, for anyone.

Since that hasn't happened, it begs the question why Intel bothered to boost their Idrives at all. Ivy Bridge responds as if the 22nm Idrive is no more than that delivered with 32nm.

I suppose their could be some other clock-limiting speedpath issue in play with Ivy Bridge, and if bulldozer was giving Intel a run for its money then Intel would be more interested in releasing a new stepping to address the speedpath issue (akin to the B3 vs G0 for Q6600 on the eve of Phenom's release). That possibility can't be discounted.

 

[Rubycon]

What about mobile versions of IB?

 

[sefsefsefsef]

What about mobile versions of IB?

Mobile CPUs don't ever have an IHS. You know the form factor where temperatures matter the most? It doesn't use an IHS. This should tell us a bit about how effective IHSes are at managing temperatures.

 

[BonzaiDuck]

But why would Intel revise their numbers? Because the data themselves are not taken from a static parent population. The qualities of the node improves in time, as do the confidence limits on the sampled data themselves. In time this leads to a more quantitative idea of exactly what the weibull plot looks like (the whiskers get smaller) while at the same time the entire line moves to the right (longer intrinsic lifetimes at any given temperature and voltage).

Forgot about that. Well, you'd imagine whatever limits they post will be based on the burn-ins and the accelerated lifetime tests. But they're not posting as much as they used to in a processor spec.

Enjoy the beach. As much as I miss the Midatlantic, I wouldn't miss the heat you folks had this summer. But I am belatedly getting it now here after a very mild pass through middle of July.

 

[Idontcare]

Mobile CPUs don't ever have an IHS. You know the form factor where temperatures matter the most? It doesn't use an IHS. This should tell us a bit about how effective IHSes are at managing temperatures.

Its not just Intel either. When I delidded my Nvidia GTX460 I saw a 10°C drop in temps with the same cooler and same TIM.

The IHS is there on desktop CPU's to protect Intel's bottom line because they know they have no control over the physical environment that the CPU's will be subjected to at the hands of end-users.

Unlike the case where un-lidded mobile CPU's are solely purchased by qualified OEM's and are unlikely to ever see the light of day once the laptop ships from the factory. Controlled environment, limited risk to Intel.

 

[Rubycon]

IHS is for protection only. Performance was never really an issue until IB however. Wonder why they stopped soldering them.

 

[Haserath]

If solder has more thermal conductivity than paste, shouldn't the IHS actually help thermal conductivity by spreading out heat for the paste?

 

[Ferzerp]

If solder has more thermal conductivity than paste, shouldn't the IHS actually help thermal conductivity by spreading out heat for the paste?

Not sure if you're asking if a soldered IHS *does* help cooling.

In a way, yes it does. However, if you could solder the heatsink to the die, that would work even better.

Solder was effectively (not exactly, but close enough) an extension of the die itself as it very thermally conductive. Paste on the other hand, is like a thick molasses that the heat must travel through (if heat were like a swimmer), and an extra layer of paste is always going to be detrimental compared to one layer and direct contact.

 

[Haserath]

Not sure if you're asking if a soldered IHS *does* help cooling.

In a way, yes it does. However, if you could solder the heatsink to the die, that would work even better.

Solder was effectively (not exactly, but close enough) an extension of the die itself as it very thermally conductive. Paste on the other hand, is like a thick molasses that the heat must travel through (if heat were like a swimmer), and an extra layer of paste is always going to be detrimental compared to one layer and direct contact.

That's just it though. Even most enthusiasts wouldn't solder the die to the heatsink.

The solder method basically allows a high thermally conductive material(solder) to transfer the heat from the small die to a much wider area for a lower conductive material to transfer the heat into the heatsink.

On a non-solder die, then, yes, one layer is better than two, but there is only one layer of paste on soldered dies.

I brought this up without knowing if everyone was referring to the IHS as just protection(for ivy) or useful for heat transfer.

Ivy might have the IHS due to socket compatibility as well. It had to use the same socket as Sandy, which used the retention bracket for pressure onto the socket. Why design something new when the old method works well enough?

 

[Homeles]

That's just it though. Even most enthusiasts wouldn't solder the die to the heatsink.

I'd totally do that, if I had the money. Soldering my stock heatsink would be hilarious though...

 

[Ferzerp]

I'd totally do that, if I had the money. Soldering my stock heatsink would be hilarious though...

I'd do it just to say I did it, but I would imagine it takes some specialized equipment and solder to do it...

You would only want to do it with a custom water block or the like. Doing it with something like an H100 would be stupid. Imagine if the pump went out. They aren't servicable easily, and you'd be stuck with a processor glued to a broken unit.

 

[rickon66]

As a long time computer and somewhat trained electronics hobbiest, I am puzzled by what folks keep refuring to as soldering the die to the lid. My idea of solder is using a metalic intermediate, lead, silver, brass or some combo of these to join two other metallic pieces together using heat to melt the intermediate material. How do you solder the non-metallic die and what is the solder made of?

 

[Ferzerp]

We're calling it solder because that's exactly what it is.


http://www.intel.com/technology/itj/2008/v12i1/1-materials/5-solder.htm