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

[Idontcare]

My findings mirror yours, IDC. My gap must have been very small, because delidding and remounting the IHS with aftermarket paste resulted in a 2-3c rise in temps at high overclocks.

4.6ghz, ~1.304v, LinX loaded:
Unmolested CPU: ~98c
TIM replaced under lid: ~100-101c
Bare die: ~95c

Surprisingly, the IHS itself wasn't hurting temps as much as I suspected either. The paste has had time to settle in for about 18 hours after I lapped my WB and I'm only 3-4c below where I was at stock.

Exactly what I was thinking too.

Awesome update IDC. While I don't think I'll ever de-lid myself, reading your very thorough work is very interesting and educating.

Do we get to look forward to that H100 mounted on the die?

Edit:
Hmmm...De-lid myself could be read a couple of different ways

Yep, got the bolts already for my H100 direct-die tests! Next on my list. Try not to delid yourself in the meantime

 

[BonzaiDuck]

My findings mirror yours, IDC. My gap must have been very small, because delidding and remounting the IHS with aftermarket paste resulted in a 2-3c rise in temps at high overclocks.

4.6ghz, ~1.304v, LinX loaded:
Unmolested CPU: ~98c
TIM replaced under lid: ~100-101c
Bare die: ~95c

Surprisingly, the IHS itself wasn't hurting temps as much as I suspected either. The paste has had time to settle in for about 18 hours after I lapped my WB and I'm only 3-4c below where I was at stock.

So . . . it's fundamentally a question of the gap, then? An addition of a certain "tedium" to a project "method" . . . And one of us had tried direct-die contact, and still had disappointing results?

 

[Ferzerp]

And one of us had tried direct-die contact, and still had disappointing results

While the plural of anecdote is not data, this was also reported elsewhere very early on.

 

[mrob27]

OK I just couldn't let this go, it bothered me that we had not definitively determined whether or not the benefits observed in delidding these Ivy Bridge CPU's came from replacing the Intel stock CPU TIM with something superior or if it came from reducing the gap between the IHS and the CPU (reducing the thickness of the CPU TIM). [...]

Thank you for your persistence and following through with these experiments.

By the way, two photos both read (in yellow lettering) "Entire CPU package (including IHS adhesive)". The second of these should instead read "Entire CPU package (with paper shim)" or something like that.

 

[Idontcare]

Thank you for your persistence and following through with these experiments.

By the way, two photos both read (in yellow lettering) "Entire CPU package (including IHS adhesive)". The second of these should instead read "Entire CPU package (with paper shim)" or something like that.

Doh! You are quite right. I have now fixed it but it may take photobucket a while before they update their cached images.

 

[Ferzerp]

Doh! You are quite right. I have now fixed it but it may take photobucket a while before they update their cached images.

Looks corrected here. Maybe they don't have so many static caches.

 

[Yuriman]

Update:

My bare die mount looks to have finally settled in. 4.4ghz is 64/69/71/63 under Prime Small FTT, whereas before delidding my hottest core was around 78c. 4.6ghz @ ~1.3v I'm getting 73/81/84/76 Prime'd, down from 89c on the hottest core. IBT at 4.6ghz is at 91c, down from 98c.

 

[Idontcare]

Turns out doing a direct-die mount of your HSF onto an Ivy Bridge chip is rather easy to do.

What isn't easy to do is getting the right amount of downforce on the CPU such that everything is copacetic.

Here's the socket bracket, simple star-key type bolts, took all of about 15 seconds to remove the entire socket bracket assembly.

Next item on the agenda was setting up the H100 with mounting bolts that don't have the standoffs that are too high given that the IHS won't be there anymore.

This turned out to be rather easy to solve, a trip to Home Depot for some #4-40 bolts was all it took. I threaded them up through the H100's support bracket from behind the mobo (silvery bolts in the picture below).

You can tell in the pic above that I really slathered on the NT-H1 TIM for this test in hopes of having the TIM serve a dual-purpose of thermal conductor as well as being a springy-spongy mass of sorts to dampen the inexperienced hands that were about to slam an H100 waterblock down onto the that bare silicon die :whiste:

Once the H100 was placed on top of the die I proceeded to add washers and nuts (inner diameter of the nuts was > the #4 bolt diameter, they are there truly as nothing more than standoff shims) until the remaining bolt length was just right for me to use the H100's stock mounting thumb-nuts (is that even a word? it feels wrong to write it ).

Now then, getting this to boot was a real trick. And keeping it from rebooting was yet another trick. I intended to put a spring on the bolt to provide some resistance to the thumb-nut torque but I got lazy and just went straight for the kill.

The result is that the operating temperatures were very sensitive to how much I tightened those thumb-nuts (as one would expect), but it was also very easy to overtighten them and cause the system to reboot (and promptly hang, BIOS gave an error code that ASUS claims means "no memory installed").

How sensitive are we talking? At 3.5GHz with the initial tightness my peak temperatures were 67°C. I tightened the thumb-nuts by a mere 180° (half a turn) and the peak temps reduced to 61°C. So I turned them another 180° only to have that cause the computer to spontaneously reboot. I backed the nuts off by 90° and the system booted and the peak operating temps were a mere 55°C.

That is a huge range in operating temperature for what amounted to less than a single full turn of the H100 mounting nuts, and going just a smidgen too far and the system reboots. Very touchy.

That said, what kind of results were I able to arrive at with my new found finicky friend?

^ compare cases "c" to that of cases "d" to see the improvements I observed in removing the IHS entirely from my system. In general I only benefited by a few degrees.

Now I realize throwing a big'ole table up like that will basically make people's eyes glaze over, so here is a graph of the same results:

The take home message there is replacing the CPU TIM doesn't really provide any benefit. Reducing the gap between the CPU and the IHS does. And removing the IHS entirely doesn't really provide much benefit either.

And that stands to reason. The heat is going to flow through the copper of the IHS the same as it does through the copper of the H100 waterblock.

But if there is a thick pad of thermal paste in the way, as was the case with my 3770k at time of purchase, then it doesn't really matter how good the TIM itself is (unless it too is made of metal) because the mere presence of that thick pad of thermal paste becomes the weakest link in the thermal conductivity equation.

Once you eliminate the bottleneck that is the gap between the IHS and the CPU, or if you happen to end up with an Ivy Bridge CPU which doesn't have much of a gap to begin with (Yuriman ), then you have pretty much optimized the system at that point regardless which CPU TIM of choice you employ and regardless whether or not you leave the IHS in place.

Now the choice of CPU TIM still plays a role in terms of the robustness in maintaining those nice low temperatures. If the so-called "pump out" effect is real then we can expect it to bite us unless we choose a substitute CPU TIM that is designed to avoid such thermo-mechanical effects.

I haven't really got into testing that part out yet, but I expect IC Diamond and the metal TIMs like Liquid Ultra to be key there.

 

[Ferzerp]

Nice...

I'll stick with my current setup though.

That, and my home depot did not have the right bolts the time I went looking.

 

[Yuriman]

IDC, I think you should invest in some springs. I feel it made mounting a lot safer (and easier). I believe your reboots are caused by the CPU tipping up in the socket and some of the pins on the underside not making contact. I never had this using springs.

My method is to tighten each bolt down until the springs just barely start to compress, and then alternate tightening each spring 360° until they bottom out and then back off a half turn.

You have to choose the right spring of course. I don't know how many pounds my springs exert when fully compressed but I believe 30-40lbs is a safe mounting pressure for a bare die, so 8 to 10lb springs would be a good place to start.

 

[rge2]

awesome info IDC, really appreciate all the testing!

 

[Ed1]

IDC, I think you should invest in some springs. I feel it made mounting a lot safer (and easier). I believe your reboots are caused by the CPU tipping up in the socket and some of the pins on the underside not making contact. I never had this using springs.

My method is to tighten each bolt down until the springs just barely start to compress, and then alternate tightening each spring 360° until they bottom out and then back off a half turn.

You have to choose the right spring of course. I don't know how many pounds my springs exert when fully compressed but I believe 30-40lbs is a safe mounting pressure for a bare die, so 8 to 10lb springs would be a good place to start.

Right , I was wonder to why the reboot and could only think of either the chip itself is flexing as there no IHS to stop it or MB flexing as there no housing on the chip to keep it in place on the pins .

Interesting results none the less .

 

[mrob27]

IDC, I think you should invest in some springs. I feel it made mounting a lot safer (and easier). I believe your reboots are caused by the CPU tipping up in the socket and some of the pins on the underside not making contact. I never had this using springs.

I kinda think IDC was counting on the LGA-1155 socket's pins to provide the needed springiness. Since there are 1155 of them (duh ) and they're not really designed to be used as springs per se, I imagine getting the right contact pressure at all four corners is pretty hard.

You have to choose the right spring of course. I don't know how many pounds my springs exert when fully compressed but I believe 30-40lbs is a safe mounting pressure for a bare die, so 8 to 10lb springs would be a good place to start.

Intel says it should be between 70 and 135 pounds for the entire socket, i.e. about 18-38 pounds per spring if you have 4 springs. See Intel's Thermal Mechanical Specifications and Design Guidelines for LGA-1155, table 503 on page 38.

They give the "ILM static compressive load on processor IHS" as 311 N [70 lbf] minimum, 600 N [135 lbf] maximum, with footnotes explaining that "This minimum limit defines the static compressive force required to electrically seat the processor onto the socket contacts." 70 lpf is the downward force exerted by something that weighs 70 pounds in normal Earth gravity.

 

[know of fence]

Now I realize throwing a big'ole table up like that will basically make people's eyes glaze over...

By golly! It's the prettiest table I've ever seen.

If I had the kind of budget, balls and bunny suit. I'd probably try to "can open" the IHS with a Dremel cutting close to the taped PCB, just to keep the retention bracket.
Better yet, I'd cut the top metal edge off, in a 45° angle so there are no sparks flying towards the die, then file down the edges. Then I'd take the flat upper lid punch a hole though and wear it as a dogtag around my neck.

Will the bracket still get in the way, once the IHS is off?

 

[pwoz]

Intel says it should be between 70 and 135 pounds for the entire socket, i.e. about 18-38 pounds per spring if you have 4 springs. See Intel's Thermal Mechanical Specifications and Design Guidelines for LGA-1155, table 503 on page 38.

They give the "ILM static compressive load on processor IHS" as 311 N [70 lbf] minimum, 600 N [135 lbf] maximum, with footnotes explaining that "This minimum limit defines the static compressive force required to electrically seat the processor onto the socket contacts." 70 lpf is the downward force exerted by something that weighs 70 pounds in normal Earth gravity.

Remember of course that the IHS is removed here. I doubt you want to use the same mounting pressure, especially given the IHS doesn't even seem to sit on the die.

 

[Yuriman]

Intel says it should be between 70 and 135 pounds for the entire socket, i.e. about 18-38 pounds per spring if you have 4 springs. See Intel's Thermal Mechanical Specifications and Design Guidelines for LGA-1155, table 503 on page 38.

They give the "ILM static compressive load on processor IHS" as 311 N [70 lbf] minimum, 600 N [135 lbf] maximum, with footnotes explaining that "This minimum limit defines the static compressive force required to electrically seat the processor onto the socket contacts." 70 lpf is the downward force exerted by something that weighs 70 pounds in normal Earth gravity.

Just curious, can you pull up the specs for the Pentium 3 or Athlon XP?

 

[BonzaiDuck]

So . . . the tricky part of what seems otherwise easy . . . derives from removal of the retention mechanism, which means that the HSF bolts (and springs, if any) are the only thing that provides good contact and "retention" for the CPU pins in the socket-1155. Funny that I hadn't even thought about that, but it should have been obvious from the beginning. . . .

 

[Ferzerp]

Remember of course that the IHS is removed here. I doubt you want to use the same mounting pressure, especially given the IHS doesn't even seem to sit on the die.

The IHS sits on the die after you get rid of the thick black adhesive.

 

[Ed1]

This brings up the question of how flexible is that black adhesive , maybe running slightly higher mounting pressures might improve with stock IHS on chip .
You might even be able to back off a bit once its burned in a bit .

It might be one reason we see some get better results than other with delidding .

 

[mrob27]

Remember of course that the IHS is removed here. I doubt you want to use the same mounting pressure, especially given the IHS doesn't even seem to sit on the die.

Oh, darn Sorry, I wasn't paying attention somehow.

Yes, this raises an important point, which is: how can we get all 1,155 pins to contact sufficiently well when all the pressure to make those pins contact has to be delivered through the die!

 

[mrob27]

Just curious, can you pull up the specs for the Pentium 3 or Athlon XP?

Here ya go

 

[Yuriman]

Looks like for the Pentium 3, Intel recommended a max of 689kPa which works out to about 12lbs on an 80mm² die (which the P3 had). If we use the same 689kPA for Ivy Bridge which has about twice the die area, that's 24lbs.

It obviously isnt as simple as that, but I wouldn't want 100lbs+ on a bare die.

 

[Klunt Bumskrint]

So is delidding worth the risk?

 

[Yuriman]

So is delidding worth the risk?

Delidding and mounting bare die lowered my temps enough that I can safely run 4.6ghz where my temps were high enough before that 4.5 was as high as I wanted to go. I'm already well into the realm of diminishing returns though.

I wouldn't recommend it to anyone who isn't trying to push clocks like you could on Sandy Bridge. With Sandy, and if you had a good chip, you could keep adding volts and scale the clocks up to the stratosphere. I believe there's a guy on this forum who has a SB that he's taken to 5.6ghz without extreme cooling, and you simply can't do that on Ivy Bridge, lid or not, because you'll hit tjmax.

 

[Idontcare]

So is delidding worth the risk?

Answer will obviously depend on why the chip was purchased in the first place - and that will be different person-to-person.

I already owned a 2600k when I bought the 3770k. I bought the 3770k 100% purely for hobby/entertainment purposes and planned to pretty much just mess around with it and either kill it in the process or eventually repurpose it to replace an existing Q6600 box that I use to harvest market data at this time.

So for me the answer would be yes, the entertainment value I have derived from delidding my 3770k has been thoroughly worth the risk of killing the chip in the process. It has been far and away more enjoyable than the fun I had with my 2600k which at most all I could do was lap the IHS.

But if I were purely looking at this from the perspective of "did it improve the performance of the 3770k enough to justify the risk of possible destroying a $300 investment just to get an extra 100MHz or 200MHz?" then I would have to say the answer is absolutely not, the paltry performance gains of adding on another 100-200 MHz is probably not worth the risk of permanently damaging a $300 CPU.

If I were in that much need for the performance, a situation where 4.4GHz was not acceptable and I really had to have 4.7GHz then I would just go buy a 2600k or 2700k, lap it, and push the volts through it and hit 5GHz.

If I were building an IvyBridge based system for anyone else I would not delid it. I would leave it "as is". If cooling was a concern because the person didn't want to listen to a noisy HSF then I'd spend money on quieter fans versus adding in the risk of destroying a $300 chip.

But if you are looking at wanting a cooler CPU and you enjoy tinkering under the hood then delidding is a rewarding process to put yourself through from a hobby perspective.

Just my opinion, I suspect everyone has a different take on it.