Fixing Ivy Bridge high CPU temps: IHS removal video

WhiteFireDragon

Limp Gawd
Joined
Jul 16, 2009
Messages
495
Hi guys, I just removed the IHS off my IB processor and thought I'd share this "tutorial" video and results.

I got a huge 40°C lower on one of the cores . If I average temps from all cores before and after the mod, then it's 79.25°C before IHS removal, and 59°C after IHS removal. That's still 20°C difference, which is quite amazing. I just got a really bad chip out of the box. It couldn't even run stock voltages because one of the cores hits over 100°C. After undervolting to 1.07v @ 4ghz OC, one of the cores still hit 98°C, so this mod was pretty much required for me to get good temps.

http://www.youtube.com/watch?v=XXs0I5kuoX4
 
Thanks for sharing. 40 degree lower on a core... I can't believe Intel didn't see something like that coming with the cheapo thermal crap they used. And given how cheap the stuff they used before is, I can't see how they saved any meaningful amount of money by using it.
 
Well I'm sure the cost of cheap paste vs fluxless soldering saved some money in terms of fabrication process and material, multiplied by hundreds of thousands of chips. But you're right though, for the cost of the whole processor, a few bucks spent extra on better heat transfer would have been appreciated.

I'm sure intel justified it by saying most people will not notice the difference between good solder and cheap thermal paste. Remember that most of their chips go to OEM for prebuilt computers, where they can't overclock or overvolt the chips. This is where they can get by with cheapo paste. But for enthusiasts that buy the unlocked K models, this is where it makes a huge difference.
 
I think it has less to do with the paste itself, now that I think about it, and more to do with the process of mating the IHS to the die. You could put cheese on the die and it still wouldn't be 40c, or even 15c difference. Thermal paste reviews stopped being interesting like 10 years ago because they were all within like 1-2c of each other, with few exceptions. Some roundups even tested against toothpaste and like Easy Cheese and the differences weren't huge.

Something about how they are attaching the IHS this time around is causing people to occasionally end up with problems like yours. I wonder if it's the glue? The glue could have been applied unevenly as well. Which could cause the IHS to lean to one side and leave one core in poor contact.
 
Thank you, excellent video. My stock i5-3570K with original TIM runs 4GHz at exactly the same temp as your modded one -- max 59C. Could it be that Intel realized they goofed and improved their production process in more recent batches? BTW mine is batch 3208C123.
 
If you were able to clean off the epoxy with your finger nail then it can't be anywhere near as hard as I had thought. I may have to give that a shot.
 
This makes me want to go out and buy an Ivy just to try this. I bet I could mount my heatkiller straight to the die.
 
I think it has less to do with the paste itself, now that I think about it, and more to do with the process of mating the IHS to the die. You could put cheese on the die and it still wouldn't be 40c, or even 15c difference. Thermal paste reviews stopped being interesting like 10 years ago because they were all within like 1-2c of each other, with few exceptions. Some roundups even tested against toothpaste and like Easy Cheese and the differences weren't huge.

Something about how they are attaching the IHS this time around is causing people to occasionally end up with problems like yours. I wonder if it's the glue? The glue could have been applied unevenly as well. Which could cause the IHS to lean to one side and leave one core in poor contact.

I agree. No way thermal paste is accounting for a 40C temp drop. Something else was wrong, and removing the IHS just happened to fix it.
 
Something about how they are attaching the IHS this time around is causing people to occasionally end up with problems like yours. I wonder if it's the glue? The glue could have been applied unevenly as well. Which could cause the IHS to lean to one side and leave one core in poor contact.
I agree. No way thermal paste is accounting for a 40C temp drop. Something else was wrong, and removing the IHS just happened to fix it.
In this case, I'm almost sure that it's either an uneven TIM application, or that the epoxy glue that they used was higher on one side than the other. This can be the only reason why one core was 30C higher than the rest, because as you can see after I took off the expoxy glue and reapplied the TIM, all the cores now run at an even consistent temp.


Thank you, excellent video. My stock i5-3570K with original TIM runs 4GHz at exactly the same temp as your modded one -- max 59C. Could it be that Intel realized they goofed and improved their production process in more recent batches? BTW mine is batch 3208C123.
Looks like you got lucky and got a nice chip out of the box with an even TIM application, and I got a terrible one. I'm sure you'll see still around a 5-10C drop in temps if you put better TIM on it. I doubt they changed their production process though.


If you were able to clean off the epoxy with your finger nail then it can't be anywhere near as hard as I had thought. I may have to give that a shot.
It feels more like rubber than some kind of super glue. Pretty easy to take off. I mostly used my finger nails because it's softer than all the metal blades at had, and I didn't want to risk damaging the PCB. Doesn't matter what you use to get the glue off the heat spreader though.
 
This makes me want to go out and buy an Ivy just to try this. I bet I could mount my heatkiller straight to the die.

Actually, there are two or three people here and on AAT that are thinking about trying this, however the general consensus is that the only way to do this without cracking the die is to use a liquid metal solution, which is... permanent.
 
I don't think liquid metal will permanently bind to the bare die. The reason why it bind to a CPU cooler and the IHS is because these two are both metal, but the die is polished silicone.
 
Nice video, Thanks. You know it occured to me in watching that when you remove the 'Glue' you effectively reduce the distance/gap between the top of the die and the inside surface of the ihs.
This may result in a bit more clamping pressure from the bracket and a tighter/better fit.
 
Great video and fantastic results.

I wish you would have used the IC Diamond though.


My 3570k has one core that is over 10C hotter than the others under load and I think the problem is the same that you had.

With results like yours it makes sense to delid and fix the problem.

Thanks again.
 
Thank you for the highly informative and interesting video :). I'm going to wait it out with my i5-750 until Haswell or Broadwell but if I ever meet someone who owns an IB system I'll show them this and coax them into doing this.

For thermal paste, here is some useful information. http://skinneelabs.com/2011-thermal-paste-review-comparison/

If you look at the graph that says "Great Contact" you will see that Prolimatech PK-1 is at the top of the chart, only beaten by Indigo Xtreme (only 0.2* which does not matter). I think that because of the way the mounting system secures the IHS there will be high pressure and PK-1 will give the best practical performance. So you can put PK-1 in between the processor die and the IHS.

It only costs $3.50 to try (Newegg on eBay) http://www.ebay.com/itm/Prolimatech...hermal_Compounds_Supplies&hash=item1c2a253ffd
 
As an eBay Associate, HardForum may earn from qualifying purchases.
Actually, there are two or three people here and on AAT that are thinking about trying this, however the general consensus is that the only way to do this without cracking the die is to use a liquid metal solution, which is... permanent.

Wrong..I have an Enzotech Sapphire (best block on the market still when considering performance vs cost) mounted directly on my delidded 3770K using Arctic Cooling MX-2 paste..Was very easy to set up using the Enzotech Socket 1155 adapter, and using 4 metal washers stacked on each peg in place of the stock Enzotech spacers...

I tightened the block's screws down in an "X" pattern until finger tight, then booted to make sure the cpu was getting enough pressure to remain in the socket..

*NOTE*, this does require the removal of the stock Intel retaining bracket..
 
I think it has less to do with the paste itself, now that I think about it, and more to do with the process of mating the IHS to the die. You could put cheese on the die and it still wouldn't be 40c, or even 15c difference. Thermal paste reviews stopped being interesting like 10 years ago because they were all within like 1-2c of each other, with few exceptions. Some roundups even tested against toothpaste and like Easy Cheese and the differences weren't huge.

Something about how they are attaching the IHS this time around is causing people to occasionally end up with problems like yours. I wonder if it's the glue? The glue could have been applied unevenly as well. Which could cause the IHS to lean to one side and leave one core in poor contact.

+1. I don't think this is something within Intel's margin of error, because if it is then that is pathetic engineering and QC.

My i5-750 when it was at 3.36 GHz on a Hyper 212 Plus, it would hit about 64* C in LinX on the hottest core. Your processor is much more power efficient than mine so when the overall result is 59* at 4 GHz that makes sense, but when it is 79* at 4 GHz... :eek: :mad: :(
 
This debate is what gots me worried about going with a Z77 setup with a 3770k over a core i7 3820 although I much rather have the beutiful maximus v formula board over any other board under 350 for the x79 platform :( what to do what to do???
 
I delidded mine and replaced with IC Diamond. 4-6 C across the different cores. Not bad and I wasn't expecting too much since my chip wasn't running nearly as hot as yours was.
 
My brother picked up an Ivy today, bookmarked the youtube link just incase.

If it appears to be cooler running Ivy (better IHS) I'll come back and say. Maybe they did correct some things.

Very nice change in temps, Warranty is voided now, so might as well lap it lol, especially with how easy it would be to do so while not connected to the die.
 
This debate is what gots me worried about going with a Z77 setup with a 3770k over a core i7 3820 although I much rather have the beutiful maximus v formula board over any other board under 350 for the x79 platform :( what to do what to do???

If your build is just for gaming, then I'd recommend you get a 3570K as there's nothing that it can't do which a 3770K can. The x79 chipset is basically for those people who wanna do Tri-SLi or 4-way Sli setups as it gives out more PCIe lanes, but you'd still be better more overall features on the Z77 platform.

Maximus 5 formula is a beautiful board but it's only more worth it for those LN2 users as you can get a much cheaper mobo with almost the same features and OC it just the same on air. In other words, you'd be paying a premium for a little of it's other features but not it's OC'ing potential as Ivy anyways can't be OC'ed that much due to the heat.

The rest is up to you man :)
 
I delidded mine and replaced with IC Diamond. 4-6 C across the different cores. Not bad and I wasn't expecting too much since my chip wasn't running nearly as hot as yours was.

I'm kinda double minded now whether to delib my 3570K or not. What do you think man, was it really worth it?
 
I'm kinda double minded now whether to delib my 3570K or not. What do you think man, was it really worth it?

I just like doing this stuff, so it was worth it for me.

That being said, I doubt it will give you much headroom for OCing, unless you are having serious temperature issues.

If I had temps like the OP was getting I would have been on the phone with Intel and getting a new processor. You shouldn't have to pop the IHS off to get functional temperatures. Seems like poor QC in this case.

The only chance it has at giving you more OC headroom is where you have high temps, but not extreme enough to get a RMA out of it. Also depends on your confidence in popping the IHS off without damaging the CPU. Expensive mistake if you brick the chip.
 
Oh man, this is a tough choice! :(

I'm hitting from 73 C to 82 C depending on the core # when doing full load and overclocking 3770K @ 4,2 GHz. Overclocking to 4,5 GHz and they hit 100 C! Note that this is measured with Core Temp not Asus Probe which reports much lower temps.

I want to stay @ 4,2 GHz and I will seldom push a full load, so it is not a big issue, but if I was sure that I could get that IHS off without an incident, I would go for it! It is just so pesky knowing that your brand new 3770K isn't allow to roam freely temp wise because of this Intel TIM nonsense.

What would you do?
 
Oh man, this is a tough choice! :(

I'm hitting from 73 C to 82 C depending on the core # when doing full load and overclocking 3770K @ 4,2 GHz. Overclocking to 4,5 GHz and they hit 100 C! Note that this is measured with Core Temp not Asus Probe which reports much lower temps.

I want to stay @ 4,2 GHz and I will seldom push a full load, so it is not a big issue, but if I was sure that I could get that IHS off without an incident, I would go for it! It is just so pesky knowing that your brand new 3770K isn't allow to roam freely temp wise because of this Intel TIM nonsense.

What would you do?

What cpu cooler are you using?
 
I'm using a noctua NH-D14

I would check your Vcore and make sure its not pumping mass volts into the chip. That cooler and the OC you have, shouldnt reach 100. It might even be uneven on the cpu/not seated right.
 
I would check your Vcore and make sure its not pumping mass volts into the chip. That cooler and the OC you have, shouldnt reach 100. It might even be uneven on the cpu/not seated right.

The Vcore is at 1,27 - 1,28 when doing full load @ 4,2 GHz. The idle temps are 26-31 C depending on the core #. I tried to do a -0.100 V in CPU Offset Mode, but that ended with a crash. Annoying. :mad:
 
The Vcore is at 1,27 - 1,28 when doing full load @ 4,2 GHz. The idle temps are 26-31 C depending on the core #. I tried to do a -0.100 V in CPU Offset Mode, but that ended with a crash. Annoying. :mad:

Way too high. For comparison I use 1.28vcore to get 4.6 out of my 3770k, and the temps you are seeing are not out of line with what's normal at that voltage range..

What do you have the voltage at when you're trying to run it at 4.5?
 
Way too high. For comparison I use 1.28vcore to get 4.6 out of my 3770k, and the temps you are seeing are not out of line with what's normal at that voltage range..

What do you have the voltage at when you're trying to run it at 4.5?

But the Vcore 1,27 - 1,28 is what the mobo / CPU automatically choose. I haven't tweaked the CPU voltage myself, so how do I get it down?

I can't remember what it was at when trying the 4,5 GHz, sorry. It was just a quick test.
 
OK, I just tried to set the CPU Voltage down from the default 1,2 V to 1,1 V in the Asus TurboV EVO.

Temps with IntelBurnTest got down to 66-72 C. Now I'm running Prime95 and the temp are 62-69 range depending on Core #.

Much better, but why is the CPU Voltage at 1,2 as ? default. Is it because I upped the CPU Frequency by 20 percent and then it add 20 percent to the Vcore? :confused:

BTW: The reported Vcore is 1,162 V in Probe II.
 
If you leave the Vcore on Auto, the motherboard will automatically increase the voltage with an increased overclock. The problem is that the motherboard is too aggresive with the voltage it uses (to ensure stability), so you can normally use a lower manual voltage to achieve similar results.
 
If you leave the Vcore on Auto, the motherboard will automatically increase the voltage with an increased overclock. The problem is that the motherboard is too aggresive with the voltage it uses (to ensure stability), so you can normally use a lower manual voltage to achieve similar results.

Or you can use a negative offset if your CPU is requesting more voltage than it needs. Core: I have the same CPU/heatsink/mobo as you (non-TB version). Here is how I achieved my 4.5GHz @ 1.2V via offset.

1) Used Asus TurboV Evo "extreme" tuning to get a general idea of my CPU could do, ended up at 4.5GHz at ~1.3V (too high/too hot).
2) Went into the UEFI (all settings still default), changed to XMP mode (to run my RAM at spec) and then changed voltage to manual mode, started at 1.3V and kept dropping it by 0.01V and doing 30min of LinX stress testing. Kept dropping it until I got a crash (around 1.19V) and kept all other voltages stock.
3) Once I figured it needed around 1.2V at load to remain stable, switched to offset mode Kept LLC on lowest setting, rebooted with default offset (0) and ran LinX to see what vcore would result, it was too low (1.175) and crashed, so I started bumping offset 0.01/0.005 at a time to get it stable.
4) Played a bit with LLC to optimize the idle/medium/max load voltages.

I'm at work so I don't have my exact settings, but I think it's XMP mode, offset +0.03V, LLC low (25%), everything else default settings. I have my NH-D14 fans spin down to 500rpm at idle and max speed when CPU hits 60C. My CPU cores are 60-67C in gaming load (BF3) and 70-79C in LinX AVX/Prime95 27.7 after a few hours.
 
Or you can use a negative offset if your CPU is requesting more voltage than it needs. Core: I have the same CPU/heatsink/mobo as you (non-TB version). Here is how I achieved my 4.5GHz @ 1.2V via offset.
.

Agreed, offset is the best way to go, if you take the time to set it up. I don't think it is actuall the chip requesting more voltage than it needs though. Auto is basically just Vid (maybe not technically called that now, but the voltage the chip is requesting) plus some arbitrary offset chosen by the motherboard which is usually too high. So picking your own lower offset works fine. Negative offset would actually take you below Vid so it normally only works for very good chips. A small positive offset usually does the trick, as it did for you.
 
If your CPU couldn't run at stock clocks with stock voltages without overheating, then it would have been grounds for an RMA since it didn't work properly out of the box.
 
Makes me feel old. I remember when CPUS didn't come with heat spreaders. I even killed a CPU while working at a shop by chipping the core.
 
Back
Top