Another delid post for the 4770k

[grayson360]

Hello all. This is my first post on Anandtech but I've been following the site for years. I love this place.

So, I have an i7 4770k on an Asus Maximus VI Hero cooled by an H100i. I'm currently using stock fans but will be grabbing better ones soon.

Using the included software with my board, AI Suite 3, it over clocked my CPU to 4.4ghz with an adaptive vcore of 1.267. I get great speeds and max temps of about 70°C. The highest I've seen it go was 72°C on Core 4 when I was stressing my CPU and GPU with the heat on in my dorm. Generally though, my PC actually cools my room down. Still though, I was hoping for lower temps.

I've been looking at delid videos and seeing how people do it. Its crazy but looks so much fun. I plan on doing the razor method simply because I tend to be very good with gentle pressure. Me and hammers don't mix haha.

So question 1. Do I have to use Cool Laboratory Ultra because it's metal based or can I just use whatever?

Question 2. What is the best thermal paste to use for the top with my H100i. The pre applied paste looked great actually and was put on so cleanly.

Question 3. Best way to apply thermal paste?

Thanks everybody.

 

[Headfoot]

You don't have to use CLU and its probably wiser to not do so as I can corrode certain metals. You get at most +2C for using the second best kind. Check Idontcare's testing thread where he shows the deltas for each TIM.

 

[Yuriman]

Generally though, my PC actually cools my room down. Still though, I was hoping for lower temps.

As for the TIM, I was using CLU and it's fine so long as all of the contact surfaces are copper. It WILL stain/permeate the copper but it doesn't deteriorate it, so far as I can tell.

I'm currently using Noctua NT-H1 and I'm actually not very happy with it. Temps are great for a few weeks, but as the paste dries out, it performs more poorly. I'm not sure what I'll try next, will probably PM IDC about it.

 

[grayson360]

As for the TIM, I was using CLU and it's fine so long as all of the contact surfaces are copper. It WILL stain/permeate the copper but it doesn't deteriorate it, so far as I can tell.

I'm currently using Noctua NT-H1 and I'm actually not very happy with it. Temps are great for a few weeks, but as the paste dries out, it performs more poorly. I'm not sure what I'll try next, will probably PM IDC about it.

Yea my PC exhausts really cold air haha. Its quite nice. GTX 780 gaming edition stays under 30°C idle and 43°C load.

The only area I would use CLU is for replacing the TIM if I delid my 4770k. Would I then have issues with it drying out or would it be fine?

 

[Yuriman]

CLU doesn't dry out, but if you're looking for a long-term solution, Noctua NT-H1 is probably not the best paste, even though it delivers the best short-term temps as per IDC's testing. I hadn't reseated in close to 6 months and found that under IBT, I was getting close to 100c with the overclock in my sig. I reapplied paste and it's down to 72c - that's how much it (Noctua) deteriorated over 6 months.

 

[Lepton87]

Generally though, my PC actually cools my room down. Still though, I was hoping for lower temps.

Are you serious? How can you cool something by dumping heat into it?
BTW. I think your core temperatures are fine so unless you want to overclock further lowering core temperatures won't bring much of a benefit. Sure, you will have a few less watts consumed by the virtue of lower static leakage and in theory your CPU will last longer, but this isn't any practical advantage. 15 or 30 years lifespan? Who cares.

 

[grayson360]

Are you serious? How can you cool something by dumping heat into it?
BTW. I think your core temperatures are fine so unless you want to overclock further lowering core temperatures won't bring much of a benefit. Sure, you will have a few less watts consumed by the virtue of lower static leakage and in theory your CPU will last longer, but this isn't any practical advantage. 15 or 30 years lifespan? Who cares.

Well the vent in my dorm blows air basically in a straight line to the front of my Corsair Carbide 540, plus having two 140mm fans in the front, h100i in pull on top with a 140mm in the back. Just the air comes out cold and our AC is set kind of high, 17°C.

You make sense but I don't know. I just like the idea of colder temps or at least a higher clock.

 

[DigDog]

CLU/CLP are nowhere near as scary as people think they are. you can use *any* TIM, but the coollabs would do just fine.

for application, check any of a million youtube videos.

not everyone has had success with the vise method of delidding *cough* me for example *cough* and the razorblade is a reasonable alternative.
however, at 4.4ghz for 1.25 is really, really good. i wouldn't delid with those numbers.
realistically the most you can get is 4.6 at 1.35, with more noise too. you got lucky, don't change anything.

 

[Tristor]

I would highly suggest using the vise method over the razor blade method. It's significantly safer for your CPU as there's no risk of cutting into the traces on the PCB or removing surface mount components like the razor blade method has. If you're curious what it all looks like, here's a set of pictures and videos I did when I delidded the i5-4670K for the rig in my signature.

Album of first part of the delid
post-delid cleanup
lapping IHS
Video of lapping at 220 grit
Applying CLU and test-mount prep
Applying CLU video Part #1
Applying CLU video Part #2
Applying CLU video Part #3

The application of the CLU to the die is the scariest part on Haswell because of the surface mount VRMs. If I had it to do all over again I would use a dab of liquid electrical tape over the VRMs prior to applying the CLU. Other than that, it's no worse than working on Ivy Bridge, which was also an easy delid.

 

[BonzaiDuck]

I'm just going to throw in again over this TIM issue, weary as I am about it.

First, IDontCare's experiments with delidding were of great value here, but with some 30 tests, something bad happened to his i7-3770K. The "bad" coincided with his test of IC Diamond paste. then people started speculating that the nano-diamond particles contributed to "copper-migration," or that they "scratched" the film on the processor die. In fact, a lot of TIM products include Aluminum Oxide and other abrasives equivalent to the nano-diamond particles.

Now, industrial economic behavior aside, the Innovation Cooling folks came back with posts here and there and an e-mail to me, denying that any such thing happens. They have a lot of customers who use diamond paste for a lot of direct-die applications with no complaints. I'd personally used it without any concern for bare-die GPU cooling. So the coincidence of IDontCare's test of ICD and his dead processor is a single outlier in the distribution. Something else caused it.

Personally, I'd be at a crossroads of decision in a delidding project when it comes to choosing between liquid metal or ICD. The metal might be worth another 1C -- 2C at most. The ICD doesn't pump out. Electrically, it gives nothing to worry about.

I just think if you're going to do something as drastic as delidding and bare-die cooler installation, you should choose your TIM and other aspects carefully. And since we were discussing the relative thermal conductivity merits of copper and silver in another thread, it's worth noting that if silver is 20% better than copper, diamond is about 400% better than either. The only thing that trumps it is "Ultra/Liquid-Pro."

 

[Zardnok]

I used Liquid Ultra on the bottom of my heat spreader and PK-3 on the top. I lapped the bottom and top both to shiny copper and have had no problems or temperature fluctuations. I was able to OC up to 4.9 with the same voltage and a decrease in overall temps from my previous 4.74. I will eventually try to get 5.0 stable, but I know it will need more voltage and I am already pushing what an H100 can cool.

 

[pm]

I went with "Cool Laboratory Ultra" on my bare-die attached 4770K with an H100i. I had IC Diamond, but I went with CLU because of comments about issues with IC Diamond on bare die attaches. I had the guys in the Intel Oregon site de-lid mine since it was part of a demo system for an Intel event (but I bought my 4770K myself), so I didn't have to decide on the razor vs. vice method... but if I had had to choose, I'd have gone with the vice.

My thread on my de-lid:
http://forums.anandtech.com/showthread.php?t=2343564

Mine has been running fine - it's my main rig at home. I checked the temps a few months back and it was still running as cool as it did the day that I tested it. I've had no issues at all with it.

 

[grayson360]

I'm just going to throw in again over this TIM issue, weary as I am about it.

First, IDontCare's experiments with delidding were of great value here, but with some 30 tests, something bad happened to his i7-3770K. The "bad" coincided with his test of IC Diamond paste. then people started speculating that the nano-diamond particles contributed to "copper-migration," or that they "scratched" the film on the processor die. In fact, a lot of TIM products include Aluminum Oxide and other abrasives equivalent to the nano-diamond particles.

Now, industrial economic behavior aside, the Innovation Cooling folks came back with posts here and there and an e-mail to me, denying that any such thing happens. They have a lot of customers who use diamond paste for a lot of direct-die applications with no complaints. I'd personally used it without any concern for bare-die GPU cooling. So the coincidence of IDontCare's test of ICD and his dead processor is a single outlier in the distribution. Something else caused it.

Personally, I'd be at a crossroads of decision in a delidding project when it comes to choosing between liquid metal or ICD. The metal might be worth another 1C -- 2C at most. The ICD doesn't pump out. Electrically, it gives nothing to worry about.

I just think if you're going to do something as drastic as delidding and bare-die cooler installation, you should choose your TIM and other aspects carefully. And since we were discussing the relative thermal conductivity merits of copper and silver in another thread, it's worth noting that if silver is 20% better than copper, diamond is about 400% better than either. The only thing that trumps it is "Ultra/Liquid-Pro."

This is why I came to Anandtech for some help. You guys rock in how much you're trying to help me or just get a common question straightened out. Thank you.

I will say this though, with the Intel cooler and all of its crappy installation pins pushed in properly, I was pushing 100°C at stock speeds. So I bought the H100i and my temps dropped insanely. OC'ed to 4.4 and still awesome temps in comparison. Now I'm willing to delid and put some CLU or whatever the best replacement is for Intel's TIM and then whatever the best paste is for the H100i. I don't feel I'll have an issue with degradation of the CPU since I'll probably get a new one in 2-3 years. I do appreciate everyone's help though. Thanks everybody

 

[Tristor]

I should point out I certainly agree that IC-Diamond is safe to use, but given the cost difference there's really no point in not using CLU on the bare-die application for the slight but still measurable temperature reduction. The liquid metal also does not pump out, however it does have potential problems with "clumping" if you don't have enough mounting pressure. This is a pretty rare issue though.

I'd say on top to use IC-Diamond or NT-H1 between the lapped IHS and your lapped H100i cold plate (and yes, please lap the H100i because every one I've seen has been pretty awful from the factory). Several other TIMs are pretty decent as well, but NT-H1 seems to be the current all-around winner on price/perf and it functions perfectly fine between the lapped IHS and waterblock, although there's nothing wrong with using CLU there as well and it's technically superior.

 

[omeds]

Having done it myself and trying various thermal pastes, I will say you want to use CLU. It provided me the best temps and doesn't dry out afaik. Other pastes I tried had almost as good temps at first but soon began to lose their effectiveness with slowly increasing temps.

I would also recommend the hammer method, its much easier and safer. You don't have to smash it, just a couple gentle taps with a block of wood and it pops off very easily.

I practiced both methods on some old celeron chips I had around. I think you have a far greater chance of damaging the chip or pcb with a razor, plus its a lot harder to do and more time consuming.
The hammer method might seem more intense but its like pulling a bandaid off, you just do it in one motion and its finished. When you fiddle with the razor over a period of time, there's a good chance you're going to at least scratch the pcb. What convinced me to use the hammer method is I scratched the pcb on a practice Celeron and could see a trace wire exposed. I didn't want the same on my 4770k.

 

[Galatian]

I might add: I destroyed my Core i5 3570K with the Razor method. I will be using the vice and hammer method on my replacement soon!

Update: Well the hammer method didn't work at all...I could hammer on it all I wanted. Maybe later batches from IvyBridge had a more "firm" adhesive used on them?

 

[Tristor]

I might add: I destroyed my Core i5 3570K with the Razor method. I will be using the vice and hammer method on my replacement soon!

Update: Well the hammer method didn't work at all...I could hammer on it all I wanted. Maybe later batches from IvyBridge had a more "firm" adhesive used on them?

You're doing it wrong then. The idea behind the vice method is that the tensile strength of the adhesive is high, but it's shear strength is not. So you're using a type of force which easily will cause the bond of the adhesive to break within itself (leaving residue on both sides). It basically requires a block of wood to assist, not just to keep from damaging/marring the PCB, but also so you can appropriately center the force. You want to take a relatively hard single hammer stroke but not "follow through" (i.e. pull it at the last second like pulling a punch). It should only take one strike. A multitude of lighter taps will do absolutely nothing because you're not overcoming the shear strength of the adhesive with that level of force.

Hopefully this helps.

 

[Galatian]

Nah I had a neuro surgeon do the razor method...you know...since I failed and destroyed my first one...this time it did work...although I'm getting lower temperatures now, I'm still hitting a wall at anything above 4,6 GHz...in the end it was really not worth all the trouble if you ask me.

 

[soccerballtux]

I'm just going to throw in again over this TIM issue, weary as I am about it.

First, IDontCare's experiments with delidding were of great value here, but with some 30 tests, something bad happened to his i7-3770K. The "bad" coincided with his test of IC Diamond paste. then people started speculating that the nano-diamond particles contributed to "copper-migration," or that they "scratched" the film on the processor die. In fact, a lot of TIM products include Aluminum Oxide and other abrasives equivalent to the nano-diamond particles.

Now, industrial economic behavior aside, the Innovation Cooling folks came back with posts here and there and an e-mail to me, denying that any such thing happens. They have a lot of customers who use diamond paste for a lot of direct-die applications with no complaints. I'd personally used it without any concern for bare-die GPU cooling. So the coincidence of IDontCare's test of ICD and his dead processor is a single outlier in the distribution. Something else caused it.

Personally, I'd be at a crossroads of decision in a delidding project when it comes to choosing between liquid metal or ICD. The metal might be worth another 1C -- 2C at most. The ICD doesn't pump out. Electrically, it gives nothing to worry about.

I just think if you're going to do something as drastic as delidding and bare-die cooler installation, you should choose your TIM and other aspects carefully. And since we were discussing the relative thermal conductivity merits of copper and silver in another thread, it's worth noting that if silver is 20% better than copper, diamond is about 400% better than either. The only thing that trumps it is "Ultra/Liquid-Pro."

I had AS5 kill 2 of my 8800/9800GTs because I got it on the capacitors surrounding the die. Sure, it doesn't conduct, but it does have capacitance!