3770k delid + tests on stock heatsink (with photos!)

[lehtv]

My 3770K has always been a bit of a hot head, especially now that I've resorted to using Intel's stock cooler. What's more, its warranty just expired... So there's only one way forward: crack it open and put some metal on!

Equipped with Coollaboratory Liquid Pro and a vice, I was eager to continue. Problem was, the vice was rusted, made worse by the fact that my dad crafted it way back when in school. For now, I had to leave the CPU sweating in its socket.

1. Refurbishing the vice​

To melt off the rust, I soaked the vice in a vinegar salt solution for about 40 hours. The bulk of the rust dissolved, but there was still some rust visible after scrubbing with steel wool. At least the old industrial green paint came off easy. One part of the tool just removed from the vinegar bath:



Now painted black (with paint compatible with rusted surfaces):



What a beauty. Now, to the matter at hand.

2. Delidding​

The CPU, its old paste wiped off with IPA and secured between the jaws of the vice (which were covered with sticky tape):





The head came off easily when adding pressure constantly and slowly. The stock paste was very tough and in some parts dried, but IPA did its trick. Most of the glue I scraped off with an IKEA family card.

3. Applying Liquid Pro​

A droplet of liquid metal on the die (probably slightly too much, but no matter). Spreading it was easy with the supplied Q-tip.





First the CPU back in the socket, then the lid back on (about 5 mm offset towards the base of the locking mechanism).

4. Test bed and method​

My "test bench" was assembled on nothing more than the motherboard's cardboard box. In a well ventilated case, cooling would be much better. Ambient temperature was about 24°C or 75 F during testing.



The essential specs -
CPU: i7-3770K, Vcore offset -0.045 V
Cooler: Intel stock with Arctic MX-4
Motherboard: Asus P8Z77-V

Fan RPM during idle was 1000, during load it was 2000. These are the stock fan's minimum and maximum, respectively. At minimum RPM, the cooler is whisper quiet, while at maximum it's not terribly noisy but annoying if continuous.

Idle temperature was measured after 10 minutes of waiting, and the reported figure is the average of minimum core temperatures. Load was achieved with 10 minutes of Prime95 SmallFFT, on two settings: stock 3.7GHz turbo, and 3.9GHz (stock single core turbo forced an all four cores). The reported figure is the average of maximum core temperatures.

5. Temperatures​

This is the part you came here to read, am I right?

Setting ... °C before ... °C after delid
Idle ...... 40.25 ....... 36.5
3.7 GHz ... 81.5 ........ 71.5
3.9 GHz ... 93.75 ....... 80.5

More than 13 °C improvement at 3.9GHz! A clock speed that previously was too hot for 24/7 use, was now fine. I didn't quite expect such results with only the stock cooler - I have to admit, I'm very pleased. A massive difference with just a 10 euro investment and a bit of manual work.

(Original screen caps combined from HWInfo64.)

Thanks for reading!

 

[Burpo]

Informative & nicely done. Thank you

 

[Smoblikat]

Nice writeup. I used a razor blade to delid my 3770K and I wound up scratching off a little bit of the green on the edge of the CPU. My load temps at 4.5ghz (stock volts) on an H70 dropped by about 24C.

 

[lehtv]

My load temps at 4.5ghz (stock volts) on an H70 dropped by about 24C.

Whoa, nice. You must've had particularly badly applied stock paste. D:

 

[TeknoBug]

Wow was the Malay CPU that bad? I had the Costa Rica i7 3770K and never had heat problems, how are you getting it to 93C? With just a Hyper 212 at 4.5GHz (stock voltage) idle was still 28-32C and load was 55-60C. I normally ran it at 4.2Ghz with a -0.115v offset, yes undervolting while overclocking, it was a beautiful silicon lottery.

 

[BonzaiDuck]

I think the temperature improvements with delidding and CLU are consistent with what IDontCare and others posted with older delidding threads.

Using CLU for a soldered chip or without delidding between the HSF and IHS is probably only good for a few C degrees.

I was especially interested in your method using the clamp. that's a heck-a-lot easier than fiddling with a razor-blade . . .

 

[Ken g6]

The head came off easily when adding pressure constantly and slowly.

Hey, that's a nice vice method! The methods I'd seen involved using the vice just on the IHS, and then whacking the PCB with a hammer! This looks a lot safer.

 

[Yuriman]

Looks great!

I dug through IDC's delidding thread for some of my old numbers for my 3570K:

Decided to repost my results both here and on XS. I have some updated temps now that CL Ultra has had some time to "settle in". Here is approximately what I posted on XS:

To begin with, my system is as follows:

3570K
ASRock Z77 Extreme4
2xHD4870 Crossfire
WC loop - Apogee GT on my CPU + MCW60 on GPU1 + Black Ice GT 360 (3x120mm) rad + Laing DDC w/modded top (1/2" intake)
2x 4GB G.Skill Sniper DDR3 1600 1.25v
Intel X-25M G2 80GB
BFG LS 680w
+ some other stuff that isn't terribly relevant to this post.


Throughout most of my testing, room temperature has had little variation. When I say "Prime" I am generally referring to Prime small FFT testing done for at least 5-10 minutes to get my water temperature up to peak. Voltages are all the lowest drooped voltages (ie under Prime or Linpack) as reported via CPU-Z. Temperatures are all reported via CoreTemp or RealTemp and are taken after the paste has had at least 24 hours to cure.


To begin with, my chip seemed to be one of the worse Ivy Bridge i5's. It needed slightly above average voltage to hit a given clock and had well above average temperatures. After reading Idontcare's post on Anandtech I decided to delid my chip and see if I could improve my temperatures and/or lower my operating voltage.

Before delidding, I had been using Phobya HeGrease between the IHS and my waterblock, a compound that reviews generally agree is among the best non-metal pastes. I had the following temperatures loaded with Prime Small FFT at a room temperature of approximately 25c:
4.4ghz - 78c (1.192v)
4.6ghz - 89c (1.304v)


Post-pop Prime Small FFT temps, with the IHS remounted over the die (secured in place with the stock processor retention bracket) and having replaced Intel's stock paste with Phobya HeGrease, I had the following temperatures:
4.4ghz - 79c
4.6ghz - 94c

^
Temperatures actually got worse after replacing Intel's TIM with HeGrease. Seemingly, my temperature woes were not a result of either excessive gap or poor thermal compound used between the IHS and die, which were the two leading hypotheses on Anandtech for high temperatures on Ivy Bridge chips.

I decided to check the flatness of my waterblock's base and the surface of the IHS, and discovered that the IHS was more than a little concave. I thought about lapping the IHS, but lacking sandpaper I decided just to mount my waterblock directly to the die.

^ Initial temperatures right after delidding, before TIM had a chance to settle in. Due to increased temperatures I had to increase voltage to get stable.

24 hours of cure time later (after my bare die mount), my Prime temperatues dropped to around 4c below where I had been before delidding (78c -> 74c). This still seemed very high for a bare die mount, so picked up some sandpaper and went at my waterblock.

Before:

After (1000grit max):

Lapping my waterblock brought my load temperatures down to 71c at 4.4ghz, and down to 84c (from 89c before delidding) - I was again stable at 4.6ghz. IBT load was down to 91c from 98c before delidding (~1.3v), but this still wasn't anywhere near the results I wanted.

I ordered some CoolLaboratory Liquid Ultra, and applied it:

Which brings me to where I am now.

Prime loaded -
4.4ghz:
78c unmolested
71c lapped WB + bare die mount
61c using CL Ultra instead of Phobya HeGrease

4.6ghz:
89c unmolested
84c lapped WB + bare die mount
70c using CL Ultra

4.7ghz:
?? unmolested
100c lapped WB + bare die mount
78c using CL Ultra


IBT loaded -
4.6ghz:
98c unmolested
91c lapped WB + bare die mount
77c using CL Ultra

______________


I hope this is useful!

Here are my long-term results of using CLU:

http://forums.anandtech.com/showthread.php?t=2362347&highlight=clu


Recently I played with voltages a bit, to see how low I could get them with a 4.0ghz OC:

http://forums.anandtech.com/showthread.php?t=2443083&highlight=

 

[know of fence]

I always wondered about lapping, if you have it polished like an antique mirror, wouldn't that also reflect some of the IR radiation (radiated heat) back on the CPU. Or maybe distances and flatness are just more important.

 

[lehtv]

Looks great!

I dug through IDC's delidding thread for some of my old numbers for my 3570K:

Here are my long-term results of using CLU:

http://forums.anandtech.com/showthread.php?t=2362347&highlight=clu

Recently I played with voltages a bit, to see how low I could get them with a 4.0ghz OC:

http://forums.anandtech.com/showthread.php?t=2443083&highlight=

I remember's IDC's delid & lapping thread, but don't remember reading your post before. Very interesting stuff. What a massive difference between Phobya HeGrease and CLU when applied directly on die. I've read of some attempts using NT-H1 and the results have been almost as good as with Liquid Pro, as in 10+ C better than Intel's stock paste

I really hope I don't run into that crystallization effect you had with CLU, although reapplying Liquid Pro shouldn't be much of a hassle as there's no need to completely remove the previous CLU before applying another 'layer'. Here's what CLP should look like after a year of aging:

https://www.youtube.com/watch?v=5tQx1LYDKsk

It's a little clumped and crystallized, but very easy to clean

 

[Idontcare]

I always wondered about lapping, if you have it polished like an antique mirror, wouldn't that also reflect some of the IR radiation (radiated heat) back on the CPU. Or maybe distances and flatness are just more important.

The primary thermal conductivity pathway between the CPU silicon and the copper IHS is conductive, not radiative.

 

[PhIlLy ChEeSe]

Nice work, I have a 3770K, I may pop the lid on. It's been under water is why I haven't thought about it yet.

 

[Yuriman]

Nice work, I have a 3770K, I may pop the lid on. It's been under water is why I haven't thought about it yet.

Yeah, I like his method. It's probably applicable to Skylake owners, too, since they won't have any surface bits under the IHS.

 

[MongGrel]

Bravo, you went there.

Have always known about that but never have tried it myself.

 

[PhIlLy ChEeSe]

Bravo, you went there.

Have always known about that but never have tried it myself.

LOL!

@Lehtv
Your pretty ballsy to pop it in the vise and crank on it, IMHO I'd be scared it would let go and boom no more CPU.

 

[lehtv]

Heh thanks PhIlLy. Where's the fun without a little bit of risk ^^

 

[RampantAndroid]

Heh thanks PhIlLy. Where's the fun without a little bit of risk ^^

Hmm....how hard was it though? As in, did the CPU come flying off, or did it slowly give and you could just do the rest by hand? How hard did you really have to crank the vise? Would dropping some IPA around the edges of the IHS to try and soften the glue help any?

Neither shoving a razor into my CPU not hitting it with a hammer are really what I want to do to my 3770k...but I can see slow and steady force being a little safer and easier to control.

 

[lehtv]

Hmm....how hard was it though?

Not hard at all. First you just firmly secure the CPU in the vice. As soon as you start cranking it beyond what's necessary to keep it in place, the lid has no choice but to give in. It's best to go slow so that the glue has time to let go sort of on its own.

What actually happened was the lid shifted a few millimeters while still attached to the PCB, and still secured between the jaws. At that point the vice had done its job and it was easy to just loosen it, then grab the chip and lift the lid with fingers.

edit: now that i think of it I realize I should've recorded it :hmm:

 

[know of fence]

It's best to go slow so that the glue has time to let go sort of on its own.

I imagine this is a point of some contention, as the original "vice method" as seen on Youtube involved giving a hard knock to the PCB with a rubber hammer, with just the IHS secured inside the vice, which sent the delidded CPU flying through the air.

When I saw this Skylake delid fail, I thought of the slow vice method.

http://www.anandtech.com/show/9505/skylake-cpu-package-analysis

 

[DrMrLordX]

Ooof. Better get out your razors for Skylake. If you dare!!!!

 

[lehtv]

Who dared give this thread 1/5 stars!

 

[PhIlLy ChEeSe]

Heh thanks PhIlLy. Where's the fun without a little bit of risk ^^

True True!
Again still flat out Ballsy IMHO! Nice work!:thumbsup:

 

[know of fence]

So I have to ask, how long did the CPU run before the temperature readings were taken?
People like to talk about running stability tests for 24 hours, but rarely about temps. Now Realtemp or HWiNFO64 will record the time the program has been running recording Tmax. In my experience the longer the program was run the higher Tmax got eventually, for statistical reasons as well as heat buildup perhaps.
Has anyone bothered to determine the time it takes for Tmax to not increase any more, it could be hours.
Anyway I wouldn't mind if both temperatures were determined after 5 min of stress testing as long as there is consistency. But I think the temptation is high to fudge temperatures. I assume people talk about Tmax of highest core taken after the same period of time, when saying the temps were this and that, and of course similar conditions ambient temperature, humidity and pressure, but I can only guess.

 

[lehtv]

So I have to ask, how long did the CPU run before the temperature readings were taken?

10 minutes of Prime95 SmallFFT. That's version 28.5, I should clarify. Earlier versions stress the CPU a bit less.

Has anyone bothered to determine the time it takes for Tmax to not increase any more, it could be hours.

IME it stabilizes pretty fast as long as your case cooling is up to par. If heat keeps rising after a long period of load, there's not enough exhaust airflow.

 

[Piano Man]

Awesome job. I'm assuming you rubbed off the glue on the back of the IHS as well?

I'm about to try this on my i7-3770K. I've had it for awhile, but this hot summer has made me reduce my overclock, which was never that great to begin with (no A/C).

I remember Idontcare's original thread, and thought "no way in hell", but I think I'm up for the challenge now.