[jqp]

Anyone building a Haswell machine?

Rant alert:

All I wanted was a new machine with the new Intel Haswell CPU, reusing a really nice case and power supply. I did not research ALL the details of the Haswell chip, successor to Sandy Bridge and Ivy Bridge. It has better onboard graphics and can handle 6 Sata III drives instead of just 2, and plenty of USB 3 ports without add on chipsets.

This machine would be a server for virtual machines and maybe some gaming, video editing, photo processing, etc.

Then, I read that this CPU runs HOTTER, not cooler than the previous Intel CPUs. SO iread some more and ordered a super-duper Noctua after-market air cooler.

This Cooler was out of stock for my 2-day free shipping deal on Amazon, so I ordered from an alternate on Amazon fulfiller at the same price and with free shipping. Then I discovered that it would tale 16 days to get to me! It's in stock!  I consider that unacceptable from Amazon or anyone really.

Meanwhile I am waiting for this massive beast of a cooler to arrive today, looking at Youtube videos from PC enthusiasts who are building hot-rod gaming machines and just overclocking all their CPUs for sport.

What do I discover? This CPU (in this case the top of the line i7-4770K) has a fatal design/manufacturing flaw! It runs hot, hot, hotter because Intel is using cheap thermal paste under the metal heat spreader (IHS) instead of using heat conducting solder. To the tune of 20 degrees C hotter than it could/should be!!!

What does this mean? Very limited overclocking for the average overclocker - a chip that will throttle itself when doing high performance tasks - a chip that could even go into thermal shutdown  just going for a run. One symptom that I saw, even on my own machine with limited testing, is that one core can be 10-20 degrees hotter than the other cores!

This all goes away when you mod the chip by prying off (delidding) the IHS, carefully scraping and cleaning the glue and the lousy thermal paste Intel used, reapplying a better thermal paste or higher end "liquid metal" paste, and reattaching the IHS.

What a bunch of manure!

So I have to wait another week for the high end metal paste so all this work I am doing will be worth it. THe average user may not notice or care, but anyone who is an enthusiast of any kind will want to do this with the Haswell chip, and many are doing it with the Sandy Bridge chips which have the same problem, only with fewer symptoms!

Here is what most are doing:

 http://www.youtube.com/watch?v=XXs0I5kuoX4

but I will probably do it this way - much quicker and probably safer all around:

http://www.youtube.com/watch?v=n3dMgRSEi2Y

[jqp]

Und ziss iss ze vay ze Germans do it:

http://www.youtube.com/watch?v=6_CNoEnRjG0

[dragoonxp20]

That is so stupid.....I can't believe Intel would do something like this. Is the thermal paste bad enough to make that dramatic of a difference in temperatures?

Kind of glad I got in on sandy bridge now.

[jqp]

Apparently not all CPUs are as bad as many of them are shown to be. But that seems to be a matter of luck. The engineering decision to use cheap paste and silicon glue seems to be a poor decision. If you want your chip to run as cool as it can, to have a good amount of thermal overhead, you will want to do this mod. Some chips have a core or cores that run up to 90 degrees C under a significant load, and I believe thermal shutdown is at 100 degrees.

[TF1216]

Any thoughts about lapping the processor?

[pjchappy]

I have a 3770K (Ivy Bridge) overclocked to 4.4Mhz with a Noctua cooler.  This chip also has paste instead of solder on the IHS, which many people were mad about.  I don't have any overheating issues.


Paul

[jqp]

Any thoughts about lapping the processor?

The way a CPU is cooled, whether by Intel's stock cooler that comes with the CPU or with a third party cooler, is via a heat sink effect: metal surface to metal surface with a thermal paste in between to fill in air gaps, and then a method of cooling off the heatsink either with air or with water or even liquid nitrogen. You can see this approach  in any amp with a heat sink, where the heat sink is cooled with ambient air or maybe a fan.

These gaps between the metal surfaces that conduct heat can be macro or microscopic, and the thermal paste that fills these voids can be more or less conductive depending on the nature and quality of the thermal paste. This is a materials science issue as well as a more general mechanical issue.

Lapping is making the outside of the IHS covering the intel chip as flat as possible (and it is not truly flat when it comes from Intel), so that there is more surface area contacting your cpu cooler's heatsink interface plate, which is usuallu extremely flat. Thermal paste is required regardless of how flat the IHS is, but flatter surfaces should make thermal paste more effective.

Here, the bigger problem in this heat transfer equation is inside the IHS between the chip die and the IHS. I am going to use modern high-end thermal interface material to replace Intel's cheap stuff, which I also suspect is not applied very well to the die/HIS interface.

And from some of the reading I have done, lapping on the outside has a much lower temperature benefit than de-lidding and reapplication of quality thermal material . I plan to use the Noctua supplied paste (which I suspect is higher quality than what Intel used inside the IHS) on the outside of the IHS between it and the Noctua cooler. Inside under the IHS I will use this stuff called Coolaboratory Ultra, which is a metal paste.  http://www.coollaboratory.com/en/products/liquid-ultra/

Based on my reading and videos I have seen, I expect to get 15-20 degrees C benefit from de-lidding, but would only get 2-3 degrees C from also lapping the IHS.

I will try to be methodical and "scientific" about my process and try to get some benchmarking. After seeing what happens with delidding and then using the Noctua cooler with the Noctua supplied paste, I will probably switch to the Coolaboratory Ultra for the Noctua cooler as well.

[jqp]

I have a 3770K (Ivy Bridge) overclocked to 4.4Mhz with a Noctua cooler.  This chip also has paste instead of solder on the IHS, which many people were mad about.  I don't have any overheating issues.


Paul

I understand. But, I believe your 3770 could be much cooler if you de-lidded. You could also theoretically overclock more , maybe 4.5, 4.6, 3.7. 4,8? Is it worth it to you? Probably not, if you have what you wanted.

Honestly I don't really care so much about overclocking at this point - I have never done it yet! But for me at this point, when I see my CPU at 65 degrees just playing a little Civilization V while testing the machine, and one core significantly hotter than the others, I scratched my head and looked into it. I knew these chips ran hotter than the Ivy Bridge, but this really bugs me. Then I read up on it and now I am unhappy! My FIRST aftermarket cooler ever, and I see that it will not really help that much relatively speaking!

I want this machine to be a very robust and reliable server for me, and I want plenty of thermal overhead, so I am a little obsessed about this project. People seem to have some chips that run cooler than others. I just want to make this chip as cool as I can with the cooler that I bought, and since I am doing this project I am going to take this extra step. We'll see how it goes

[jqp]

 OK I have found some more important info


There are some very serious folks working on this issue - they are professionals and complete amateurs both in forums discussing and comparing notes, just they we have done here for audio hardware on AC.

1. I have learned that no one knows whether there is a valid reason that Intel made the engineering decisions they did except Intel. As in audio, there can be different reasons to do different things by vendors, some may be questionable, some may be valid based on info that the consumer just does not have (like testing data).

2. Modifying the chip to be like the Sandy Bridge generation where the IHS was soldered to the CPU die does definitely improve the temperatures of the cores assuming it is performed in a successful way.

3. Delidding this chip or lapping the IHS not only voids your Intel warranty (obviously) but is also way beyond normal mods a typical home enthusiast would do, This is just a disclaimer for those here on AC. So for the purposes of this topic, I am a crazy modder (even though I haven't quite modded yet)

4. Here are some pics to help you see what we are talking about in addition to those videos




This is a picture of what you have when you delid one of these - you have some careful work to do



This is the CPU put back on the motherboard without the IHS, with some Thermal Interface Material being appplied. Yes some people are placing their coolers heatsink right on top of the Intel cpu die! Not just willy nilly but with careful measirements and support. Others are merely placing the IHS back on the pcb with a shim to keep the gaps appropriate, and claming it they way it would be without delidding. This is probably what I will do.




And here is Diffusion Welding, what I am considering. The Coollaboratory Ultra that I have ordered will literally fuse two appropriate pieces of metal together at 50-70 degrees C, temperatures that the CPU will provide. This is pretty much a permanent solution - you can pull the metals apart, but it will not be pretty.

5. One of the guys I am following goes by Idontcare - he is a moderator on Anantech. He has worked at TI and knows a lot about this technology and is willing to share it. He has actually led the way on many of these mods and does pretty thorough testing. Better him than me.

6. The Intel cpu die you can see is upside down mounted to the pcb. The electronics are on the pcb side, not even near the back of the die. The surface of the die is silicon nitride, which is tough but it can be fractured.

7. You can easily fry your cpu if you are not careful. Idontcare started with an i7-3770 and after many different tests he did fry it. Again, better him than me, but there is always the possibility as we in audio know.

[jqp]

Well I have successfully delided my 4770K and applied CLU liquid metal to do a kind of fusion solder betwee nthe Intel cpu die and the Intel IHS. The I attached a new Noctua cooler. I am getting about 40 degrees C lower temperatures! Probably at least 20 degrees lower if I kept the Intel stock cooler.



My delidded 4770k - you can see the thermal paste Intel used. I had to clean it and the silicone glue off of the cpu pcb as well as the IHS and the paste off of the cpu die.



Here are the cleaned chip die on the pcb and the IHS with an application of CLU. As you can see it is a thin layer only.

The IHS was not glued back on, but placed on top of the die only, in a position where the cpu retention mechanism on the motherboard would clamp it in place. Care must be taken as there are capacitor pads next to the die. Once I had the chip clamped down like a normal installation, I installed the Noctua cooler with the NT-H1 paste that comes with the Noctua cooler.

[pjchappy]

Nice! 

This is encouraging me to do the same with my 3770K.


Paul