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

[Akantus]

Whatever they call it, it doesn't matter. It is (1) an extra thickness of nickel-plated copper between the die and the heatsink, (2) it requires two layers of thermal interface material instead of one. All of these things reduce the transfer of heat per unit of time into heatpipes or a waterblock.

Intel knows this; we know this. Whatever Intel wants to call the IHS, it's there to protect the die first, and "spread" the heat as second priority.

Yep, notebook processors don't have Heat spreaders, because it would only hamper the heat transfer, and they are assembled in factory and stays the same through the life of the notebook.

But give average Joe bare die CPU and see what happens while he tries to install his DH-14 D: That's what IHS is for.

 

[yottabit]

Yep, notebook processors don't have Heat spreaders, because it would only hamper the heat transfer, and they are assembled in factory and stays the same through the life of the notebook.

But give average Joe bare die CPU and see what happens while he tries to install his DH-14 D: That's what IHS is for.

you know, I love to defend delidding and bare die setups since it was common back in 2000s, but you bring up a good point because I failed to remember the afternarket heatsink size has probably tripled since then

One thing IVB has had me wondering is that from a thermodynamic heat rejection standpoint, couldnt raising the temerature of the CPU be a good thing? at least in a desktop application where it's not sitting on your lap, a CPU designed to operate at a higher temperature should be able to reject heat to its environment faster for a given amount of airflow, shouldnt it?

as in a hypothetical cpu running at 120 c should dissipate heat more effectively for a given size heatsink and fan speed than a cpu with the same tdp running at 60 c. So theoretically if you are ok with those higher temperatures you could use less fan speed with a hotter running cpu. Or effectively cool a CPU with a higher TDP.

 

[sm625]

I am very curious to see how much one of these chips can be overclocked using the stock cooler, aided by the process of delidding and replacing the TIM. I am looking for the absolute cheapest overclock, which seems to me would be:

1. i5-3570k
2. stock cooler
3. $7 tube of TIM
4. $2 tube of adhesive
   

If one could get 4.2 GHz out of that combination, I think it would be the best value.

 

[Yuriman]

I am very curious to see how much one of these chips can be overclocked using the stock cooler, aided by the process of delidding and replacing the TIM. I am looking for the absolute cheapest overclock, which seems to me would be:

1. i5-3570k
2. stock cooler
3. $7 tube of TIM
4. $2 tube of adhesive
   

If one could get 4.2 GHz out of that combination, I think it would be the best value.

Typically the stock cooler has no issue with 4.2 and 4.3ghz as it is. You might even be able to get away with an undervolt at those clocks.

 

[cytg111]

..So the temperature drop is real, the improvement in OC'ing is real, the question would be a matter of "is it worth it"? But that is a universally applicable question to all things OC'ing or enthusiast-like with this hobby...

For a 200mhz potential, i would not screw with anthing.
Seems like there is a good reasoning behind intels packaging here ..

 

[Haserath]

If you like modding and are very meticulous, I would say the cooler temps, lower power consumption, and lower noise would be worth it.

 

[Puppies04]

One thing IVB has had me wondering is that from a thermodynamic heat rejection standpoint, couldnt raising the temerature of the CPU be a good thing? at least in a desktop application where it's not sitting on your lap, a CPU designed to operate at a higher temperature should be able to reject heat to its environment faster for a given amount of airflow, shouldnt it?

Sure, the higher the differential between the CPU and ambient temps, the easier it will shed that heat but you forget to factor in the extra voltage a chip is going to require to run stable at 120°c which would be a major downside to running at these sort of temps

 

[Rubycon]

Good post, lots of information!

Makes you want Intel to go back to soldering IHS directly to the core again!

I wonder what temperatures would be like with a real water block mounted to it.

 

[cytoSiN]

Wow, just finished reading this thread. You, sir, are my new hero. Thanks for the amazing information!

 

[Denithor]

One thing IVB has had me wondering is that from a thermodynamic heat rejection standpoint, couldnt raising the temerature of the CPU be a good thing? at least in a desktop application where it's not sitting on your lap, a CPU designed to operate at a higher temperature should be able to reject heat to its environment faster for a given amount of airflow, shouldnt it?

as in a hypothetical cpu running at 120 c should dissipate heat more effectively for a given size heatsink and fan speed than a cpu with the same tdp running at 60 c. So theoretically if you are ok with those higher temperatures you could use less fan speed with a hotter running cpu. Or effectively cool a CPU with a higher TDP.

There are two flaws with this reasoning. First has already been mentioned above, the hotter a chip runs the more voltage it requires to remain stable at a set speed. Which leads directly into the second flaw - heat (waste energy) generation and subsequent increased temperature is a result of the power consumption in the chip.

The temperature you reach is simply a symptom of the amount of heat generated versus the efficiency of the total cooling package. So temperature will be lower if your chip consumes less power (less waste energy to eject in the form of heat) and also if your cooling system is more efficient at drawing that energy out of the cores and venting it into the atmosphere. If this process is inhibited the waste energy has a longer "dwell time" in the cores and is dissipated as heat directly from the cores instead of being ejected into the HSF. And this is exactly what TIM does - energy transfers more slowly through it than through metals (which is why the liquid metal TIMs are so effective).

The next obvious question is "Why do we use TIM then?" Easy answer - energy transfer through even the worst TIM is still faster than through air. In the interface between the cores and IHS and IHS to HSF there are gaps where you have no metal-to-metal contact. Lapping reduces these gaps but even mirror polished surfaces have microscopic gaps where they don't touch. And these gaps are horribly inefficient at transferring heat. When you apply TIM you fill these gaps with a material that, while nowhere near as effective as direct metal-to-metal contact is still many times more effective than air at transferring heat. So the net effect is a faster dump of the heat up into the cooling fins of the HSF for dissipation.

And this is why the solder Intel used to use was so effective - it maintained a direct metal-to-metal interface from the cores to the IHS. So there was only one layer of non-metallic TIM to impede the flow of heat from cores to HSF. Now, with the new TIM Intel chose to use in IB, there are two layers to slow the transfer and therefore the waste energy sticks around and heats things up.

 

[Idontcare]

I read a number of requests from members for additional tests, those requests will be honored, just give me a few days as I am heading out to go camping with the family for the weekend.

you know, I love to defend delidding and bare die setups since it was common back in 2000s, but you bring up a good point because I failed to remember the afternarket heatsink size has probably tripled since then

One thing IVB has had me wondering is that from a thermodynamic heat rejection standpoint, couldnt raising the temerature of the CPU be a good thing? at least in a desktop application where it's not sitting on your lap, a CPU designed to operate at a higher temperature should be able to reject heat to its environment faster for a given amount of airflow, shouldnt it?

as in a hypothetical cpu running at 120 c should dissipate heat more effectively for a given size heatsink and fan speed than a cpu with the same tdp running at 60 c. So theoretically if you are ok with those higher temperatures you could use less fan speed with a hotter running cpu. Or effectively cool a CPU with a higher TDP.

I can see where you are coming from with this line of thinking, but alas it doesn't work out quite that way once you factor in a specific problem relating to leakage and temperature in semiconductors. (see this thread)

The higher the temperature of the semiconductor, the more leakage occurs, the more the power-consumption rises, the more the temperature rises.

We can blame this temperature-dependent leakage on two scientists (Yakov and Horace) as it is referred to as the Poole-Frenkel effect:

In solid-state physics, the Poole–Frenkel effect (also known as Frenkel-Poole emission[1]), is a means by which an electrical insulator can conduct electricity. It is named after Yakov Frenkel, who published on it in 1938,[2] and also after H. H. Poole (Horace Hewitt Poole, 1886-1962), Ireland.

The Poole–Frenkel effect describes how, in a large electric field, the electron doesn't need as much thermal energy to get into the conduction band (since part of this energy comes from being pulled by the electric field), so it does not need as large a thermal fluctuation and will be able to move more frequently.

In practice this means that power-consumption for our CPU's rises dramatically and significantly with rising operating temperature:

^ This is a real-world example with a 2600k at a mere 2GHz, you can see the power consumption rise nearly 33% solely because of the increase in leakage as the operating temperature rises from 47°C to 96°C in this example.

Whatever benefits you may have thought were to be had in improved thermal dissipation are erased and overwhelmed by the simultaneous increase in power-consumption that comes with the higher operating temperatures ala the Poole-Frenkel effect.

 

[PlasmaBomb]

Good post, lots of information!

Makes you want Intel to go back to soldering IHS directly to the core again!

I wonder what temperatures would be like with a real water block mounted to it.

Feel free to send IDC a real water block Ruby

 

[OVerLoRDI]

Good post, lots of information!

Makes you want Intel to go back to soldering IHS directly to the core again!

I wonder what temperatures would be like with a real water block mounted to it.

I'll find out soon enough Got my 3570k on the way and I'll be following IDC's method and pairing it with a Raystorm.

 

[BonzaiDuck]

Feel free to send IDC a real water block Ruby

Well, the Corsair H100 is at least as effective as the Noctua NH-D14 heatpipe cooler.

We'd still like more data to see if there are voltage reductions with further decrease in temperatures; IDC already proved it with the "so-so" thermal paste. What did we do? I think we took a load temperature of 94C down to about 74C, and other "tests" show that improving the thermal paste option further reduces the temperature -- possibly into the 60's Celsius.

That leaves more exotic solutions. "Better" watercooling of the standard variety might get a few degrees. "Chilled" water cooling or anything that would bring down the temperatures to nearly ambient values might reveal more in terms of the noise and leakage reduction, lower voltage and the same or higher speed.

As I see it, you'd want to see how things improve with voltages in the vicinity of 1.30V. then we can split hairs over whether this comparison should be made for the severe load voltage, the idle "turbo" voltage, or both -- representing a range.

 

[yottabit]

Thanks everyone for clearing that up! I figured there was a reason CPUs weren't all running hot

 

[beginner99]

ala the Poole-Frenkel effect.

So the main reason for this craze for low CPU temps is to reduce power consumption?

I ask because I kind of did not get it why it was so cool to have 40 load temp vs 60. CPU won't break either way.

 

[Rubycon]

Feel free to send IDC a real water block Ruby

The best is the all copper EK supreme. There are others that may test better on some "shady" sites with flow vs. temperature, etc. But for the long term reliability and mounting that remains the one to beat.

Well, the Corsair H100 is at least as effective as the Noctua NH-D14 heatpipe cooler.

We'd still like more data to see if there are voltage reductions with further decrease in temperatures; IDC already proved it with the "so-so" thermal paste. What did we do? I think we took a load temperature of 94C down to about 74C, and other "tests" show that improving the thermal paste option further reduces the temperature -- possibly into the 60's Celsius.

That leaves more exotic solutions. "Better" watercooling of the standard variety might get a few degrees. "Chilled" water cooling or anything that would bring down the temperatures to nearly ambient values might reveal more in terms of the noise and leakage reduction, lower voltage and the same or higher speed.

As I see it, you'd want to see how things improve with voltages in the vicinity of 1.30V. then we can split hairs over whether this comparison should be made for the severe load voltage, the idle "turbo" voltage, or both -- representing a range.

H100 has advantages of the tower leviathans of less motherboard stress and is not position sensitive in regards to peak capacity.

However the H100 like most of the very entry level products employing liquid as the prime mover require obnoxiously loud fans to pull away from the pack. (while still firmly remaining in last place from true liquid cooling solutions!)

Exotic cooling with chilled water or refrigeration certainly is not required to make a big difference over these all in one solutions particularly when shooting for the moon on large overclocks and running at heavy load 24/7. There is simply no comparison. As long as one uses sufficient radiator area in a decently cooled ambient they will be fine. Liquid temperature should be at most 5C above ambient but preferably at ambient temperature. Very few hobbyist systems can achieve this level (approach 0 deg C rise) but it's not impossible.

As in the goal of the OP, one wants the most efficiency heat transfer between the actual core and the heatplate. The block needs to be as cool as possible and should never feel warm to the hand. When this is achieved - even as the system runs for months on end folding/crunching or running LinX - then you have GOOD cooling!

 

[WhoBeDaPlaya]

Video needs narration so we know what your voice sounds like

No! A video this epic needs a voice-over by Morgan Freeman!

 

[BonzaiDuck]

Exotic cooling with chilled water or refrigeration certainly is not required to make a big difference over these all in one solutions particularly when shooting for the moon on large overclocks and running at heavy load 24/7. There is simply no comparison. As long as one uses sufficient radiator area in a decently cooled ambient they will be fine. Liquid temperature should be at most 5C above ambient but preferably at ambient temperature. Very few hobbyist systems can achieve this level (approach 0 deg C rise) but it's not impossible.

I don't disagree with anything you say on this issue. Surely, with bigger radiator and reservoir, WC without "chilling" or other complications is much more effective certainly than the H100.

I probably should have avoided touching on those aspects of a cooling solution in my post, because the main thing in which we're interested here is the interplay of voltage, speed and temperature using mainstream or less-cumbersome cooling solutions.

Even so, cooling solutions better than "canned water" or "heatpipe" might tell us whether the die-shrink of IB offers additional advantages pertaining to that three-way relationship. Then, an individual user would have to evaluate whether more elaborate water-cooling is worth it for the additional gains -- if any.

But I'm wondering whether our focus on the TIM choices and the IHS will show a practical benefit in reducing over-clocked load temperatures to the mid-60s Celsius -- for certain clock speeds and voltage ranges. What hasn't been presented so far are results in the 4.7 to 4.9 Ghz range for MINIMUM stable voltages at temperatures lower than IDontCare has so far achieved. In other words, for 4.7 to 4.8 Ghz, how close can we come to a 1.30V guideline if temperatures can be reduced another 8 to 10C below those shown so far with IDontCare's experiment?

 

[Idontcare]

So the main reason for this craze for low CPU temps is to reduce power consumption?

I ask because I kind of did not get it why it was so cool to have 40 load temp vs 60. CPU won't break either way.

You're CPU won't break even if you run it at TJmax.

That is why TJmax is the value it is and not some value lower than what it is. If your CPU could break at TJmax then Intel would have spec'ed it to have a lower TJmax.

But...that also tells you what will happen to your CPU if you run it above TJmax. TJmax for my 3770k is NOT 110°C for a reason, bad things happen too quickly at that temperature, but the bad things do not happen too fast at 105°C, so Intel set TJmax at 105°C.

For sandy bridge, the bad things happen too quickly at 105°C, hence TJmax for my 2600k is merely 98°C.

Keeping your operating temps below TJmax is not necessary unless you wish to have your CPU have an expected operating lifetime that spans more than 10yrs.

That said, why do we care about temperatures?

3 reasons - power consumption, operating voltage, and noise.

The minimum voltage necessary to for your CPU to function stably is dependent on operating temperature and clockspeed. If your CPU's temperature increases then you must either increase your Vcc or reduce your maximum clockspeed (if overclocking or undervolting).

Temperature and power consumption also tie in directly to the noise levels of your cooling solution if you have dynamic/responsive cooling enabled (as is done with video cards).

My ASUS mobo enables this, it is pretty cool, I disable it for my tests but I leave it enabled when I'm actually using my rig for day-to-day stuff.

(note that the x-axis is different in both graphs)

In the end we have control over the temperature of our CPU by virtue of having control over three things - setting the CPU clockspeed, setting the CPU voltage, and selecting the cooling solution for the CPU.

Managing the temperature can then give us more overclocking headroom, lowering operating noise, lower power-consumption, or some combination of trade-offs in all of the above.

 

[Idontcare]

IDC i would like to ask if you could measure the Core i7 2600K at default settings and the Core i7 3770K (better TIM between the die and IHS) at default settings but with the default Heat-sink+fan in both of them .

I would like to see whats the impact of the smaller die size in operational temps if any at default settings and cooling solutions.

Hi AtenRa,

OK, I am setup to run this test now (got the stock HSF on the 3770k, etc) but I have a couple of questions - you say "default settings" in your post, but is that really what you want?

Default setting for the 2600k is going to run at 3.8GHz with a Vcc of about 1.371 V (ridiculously high, I know, but that is the VID for my 2600k when running at "stock" and so my mobo uses that VID setting) whereas the 3770k is going to run at 3.9GHz with a Vcc of 1.243V (also a ridiculously high Vcc).

I guess what I am wondering is (1) would you like these ran at reasonable Vcc's for their clockspeed (i.e. undervolt them), (2) or would like them ran at the same Vcc, and (3) would you like them ran at the same clockspeed?

I realize it is 4:30am in Greece right now, so the chances of you responding to this post within the next few hours is likely to be zero. Unless I happen to hear otherwise from you, I will plan to run the 3770k at both 3.8GHz/1.124V and 3.9GHz/1.149V (the same clocks and volts that I ran my 2600k with its stock HSF).

Regards,
IDC

 

[Ferzerp]

IDC, are you keeping the NT-H1 under there long term? As I've mentioned before, I start to have temp issues after a few months. I pulled it off earlier today, and the paste looked rather odd. Instead of being uniform, it appeared to have separated in to a liquidy phase and then a more solid, normal looking phase. Sort of as if there is a solvent in it that never really evaperated, but had separated from the solid.

I'd be curious if you see something similar over time.

In the mean time, I'm trying AS5 under the IHS to see if I have better results after a few months, or if I also start to have high temps on the topmost core with that as well.

 

[Idontcare]

IDC, are you keeping the NT-H1 under there long term? As I've mentioned before, I start to have temp issues after a few months. I pulled it off earlier today, and the paste looked rather odd. Instead of being uniform, it appeared to have separated in to a liquidy phase and then a more solid, normal looking phase. Sort of as if there is a solvent in it that never really evaperated, but had separated from the solid.

I'd be curious if you see something similar over time.

In the mean time, I'm trying AS5 under the IHS to see if I have better results after a few months, or if I also start to have high temps on the topmost core with that as well.

D: that's not good

I have not seen NT-H1 do that so far (act like it separated) in any of my applications thus far, but I take your words of caution very serious.

To answer your question - I have no long term plans to use NT-H1 as the CPU TIM.

If I had to choose today, barring the collection of any more data with which to make an informed decision, I would toss a coin and either use Ceramique or IC diamond. Both of those TIM's employ very different matrixes and neither would be expected to yield identical results to that of NT-H1, for better or worse.

And since it sounds like NT-H1 tends towards the worse, I would opt for something that would at least have a non-zero chance of being for the worse as well.

My exclusion of the metal TIM's is intentional within this hypothetical conjecture, having used Indigo Xtreme in the past, I would not want to put any kind of TIM on my IB that would be "permanent" perchance it too should turn out to need routine maintenance as you have observed to be the case with NT-H1.

That said, the permanent nature of the metal TIM's probably makes them ideal candidates if one is seeking to avoid the routine maintenance angle. Nothing is separating or moving around on you once you've let those metal TIM's fully set. Personally I have every expectation of using a metal TIM on my 3770k once I am all done playing around with it. But by then I am hoping for that to be an informed decision and not just a coin toss result

 

[Rvenger]

D: that's not good

I have not seen NT-H1 do that so far (act like it separated) in any of my applications thus far, but I take your words of caution very serious.

To answer your question - I have no long term plans to use NT-H1 as the CPU TIM.

If I had to choose today, barring the collection of any more data with which to make an informed decision, I would toss a coin and either use Ceramique or IC diamond. Both of those TIM's employ very different matrixes and neither would be expected to yield identical results to that of NT-H1, for better or worse.

And since it sounds like NT-H1 tends towards the worse, I would opt for something that would at least have a non-zero chance of being for the worse as well.

My exclusion of the metal TIM's is intentional within this hypothetical conjecture, having used Indigo Xtreme in the past, I would not want to put any kind of TIM on my IB that would be "permanent" perchance it too should turn out to need routine maintenance as you have observed to be the case with NT-H1.

That said, the permanent nature of the metal TIM's probably makes them ideal candidates if one is seeking to avoid the routine maintenance angle. Nothing is separating or moving around on you once you've let those metal TIM's fully set. Personally I have every expectation of using a metal TIM on my 3770k once I am all done playing around with it. But by then I am hoping for that to be an informed decision and not just a coin toss result

I bet ceramique would solidify nicely.

 

[Ferzerp]

I've never seen a paste do that before. Have you?