Going topless or not?

With my cooling woes on one sample of 6700k where there's a large differential between cores, I've decided to delid it and have ordered the Rockit tool. Depending on how that goes I'm tempted to do the other K processors, but I was also thinking of using an i3 as a practice run... before anyone questions if I need to do this, I do. I've observed temperature dependant stability problems once things start going above 80C or so, and know they shouldn't get that hot with the cooling solution and settings used.

Anyway, where I'm undecided is to run topless or not. The safer option seems to be a re-TIM and lid back on, where I hope I will do a better job than Intel did. The other higher risk option is to go topless. I note Aqua Computer make a dedicated shim and wonder if that is sufficient. I further note that due to removing the thickness of the IHS there may be problems with getting a good fitting depending on the cooler.

my game rig contains a 4790K that much prefers to party neked, it's a bit of a cross dresser too, there is no mount kit for them so it also wears a neked ivy kit.
on the whole, many, many sexy rig.

What is the temp differential between hottest and coldest cores at around 80c package temp? I delidded my 4790k but it's not naked. I replaced the stock paste with Lquid Ultra metal TIM. Ran much cooler but still get a temp differential between hottest and coldest cores of around 6-7c at 80c+. More than I'd like.

Be very, very careful when going naked. Make sure you have some shims to distribute the pressure on the socket from the cooler clamp. If you don't have something to spread the pressure out to the edges of the CPU PCB the pressure will fall more on the center of the socket and cause pin distortion and damage. You can also break the CPU die itself. I borked a 4790k trying to go naked without taking these precautions.

Have you looked at tests comparing temps improvement going naked? From what I have seen, it doesn't really help much. The real gain comes from replacing the factory TIM with something more thermally conductive, with our without the lid in place. Much safer too, to leave the lid in place, that is.

There is a company in Germany making dedicated shims which I would get if I went topless. Looking at notes it was 69 and 77C for coolest and hottest core similar to the current running configuration, bare in mind it is now cooling down for winter and it would be higher in summer. The hotter temps and bigger differentials were in a different config earlier but there was a suspected mounting problem there.

Relidding would be the lower cost and arguably safer route. As an example, I find it ironic that even when under load, my 250W rated 980Ti is far cooler than my under-volted 6700k, both on the same water loop, probably due to the poor TIM under IHS. I'm not going liquid metal, and will stick to conventional non-conductive pastes. I wonder how much isn't the TIM, but the tolerances on the IHS. If there is a bigger gap, that would not help. That would be an area for further research somehow. I was wondering if sanding the edge that contacts the substrate might help by increasing contact to the core, but will only do this if I can prove it. I have ideas on how to measure the space between core and IHS after delidding.

mackerel said:

There is a company in Germany making dedicated shims which I would get if I went topless. Looking at notes it was 69 and 77C for coolest and hottest core similar to the current running configuration, bare in mind it is now cooling down for winter and it would be higher in summer. The hotter temps and bigger differentials were in a different config earlier but there was a suspected mounting problem there.

Relidding would be the lower cost and arguably safer route. As an example, I find it ironic that even when under load, my 250W rated 980Ti is far cooler than my under-volted 6700k, both on the same water loop, probably due to the poor TIM under IHS. I'm not going liquid metal, and will stick to conventional non-conductive pastes. I wonder how much isn't the TIM, but the tolerances on the IHS. If there is a bigger gap, that would not help. That would be an area for further research somehow. I was wondering if sanding the edge that contacts the substrate might help by increasing contact to the core, but will only do this if I can prove it. I have ideas on how to measure the space between core and IHS after delidding.

Click to expand...

Excellent question. I've wondered that too. I have done a little sanding on the edges of my lid but not sure if I did enough to make an impact. I was afraid of doing too much and causing the die to be crushed by mounting pressure.

Can you give a link for the German company you reference that makes dedicated shims?

Probably Caseking.

Mr.Scott said:

Probably Caseking.

Click to expand...

Didn't see anything like that there.

trents said:

Can you give a link for the German company you reference that makes dedicated shims?

Click to expand...

http://shop.aquacomputer.de/product_info.php?products_id=3378
http://www.watercoolinguk.co.uk/p/Aquacomputer-Spacer-for-Intel-Skylake-CPUs_56150.html

2nd link is a UK reseller where I would go if I were to get it.

I would re-tim and use the original IHS. I tried the shim and ended up cracking my die. I had spotty issues when trying to post and it seemed more pressure was needed because it worked. It posted then BSOD'd I pulled it out and the PCB was bent and the die was cracked. 6700k down the tubes . I had good luck just re-applying the TIM though.

This last paragraph in their description of how the Aquacomputer spacer works concerns me:


Important note:
The usage of the CPU without the heatspreader makes it necessary to pay attention to the contact surface of heat sink. If its contact surface is too large it can rest on the plastic corners of the CPU socket. Since these corners are slightly higher than the CPU die, an incompatible heat sink could have no contact at all!

Aren't most aftermarket heat sink bases larger than the socket so that they would rest on the plastic corners?

Also, if Intel used PGA instead of LGA this wouldn't be quite so tricky. As it is, any distortion of the PCB will result in poor contact with the socket pins which is a problem in addition to cracking the die.

Last night I order the EK naked mount kit for Ivy/Haswell and the corresponding water block (EK Supremacy Evo). This will be for my delidded 4790k which gives considerably higher temps under load (and excessive variation between cores) than it should. I'm pretty sure the problem is the IHS is not making good contact with the die.

The problem with making a shim is you need to match the core thickness without going over. Too tall and you have bad contact. Too short, the core takes the pressure. Their solution as I understand it is to be ever so slightly thinner than the core, and a mild bending will equalise the pressure. I don't know how brittle the CPU substrate is, but typical PCBs can bend a fair amount the components on them will break long before the board does.

Ok, for now I'll plan on a relid after fresh TIM and will give the shims a miss.

Yes, but what Aquacomputer seems to be warning about is using larger heat sinks that rest on the edges of the plastic box of the socket itself that will hold the heat sink up and prevent it from creating enough deflection of the PCB to make contact with all the socket pins. What I'm saying is most aftermarket heat sinks are big like that.

I got that point, but I'm not sure about "most". I think popular ones like CM212 (at least when orientated so heatpipes leave to the side) or most Noctuas would probably be ok. Have to consider each one on a case by case basis.

Thinking more, I'd also worry about what the top part of the socket retention mechanism would actually grip onto... and if the heatsink would conflict with that.

I remember MSI made the MSI Die-guard for their selected boards. Can't recall anyone here use one tho and it wasn't available unless you purchased a specific board from MSI.

I wanted to try that die guard.

If you havnt done it already, I would suggest training on these 2$ old pentium CPUs. I know i will train on them before delliding my Q8400.

I've delidded several processors, a couple of 4790ks and a couple G3258s (just for practice). Never had any problems with it. Used the vice and tap method. I didn't even know until reading this post that someone has come out with a delidding tool. Kind of nice.

I just hope the tool is as good and safe as claimed. Various videos make it look simple enough. I don't want to do the manual knife method for sure.

Looking at tracking. Since ordering it has only made its way from Houston to Chicago. When do they stick it on a plane to me?

Chicago is an International hub. It's probably on the plane now.

Just used the tool and I now have a topless 6700k in front of me. I never realised before the IHS was so thick! I thought it would be a thinner cap and the die would be taller.

Removing the old glue now...

Here is the CPU shortly before I put the lid back on. I found using a fingernail to scrape off most of the old glue, then isopropyl alcohol to clean off most of the rest worked well. Using Noctua compound both inside and out of the CPU. Tried using superglue to put the lid back on and that didn't go great. It was still soft when I took it out of the fixture after a couple hours. I decided to just put in back in the PC anyway.

Reassembly wasn't a significant problem. I plugged in the power cable and network, and hit the power button. Things happened, but I couldn't reach it over the network. Hmm... I had to plug in a monitor, and it was stuck on "new CPU detected" even though it was the same one as before. It also did the "keyboard not found" thing and asked me to press F1. Anyway, a quick trip in and out of the bios and I was back in Windows.

I took temperatures before the operation. I'm using PrimeGrid PPSE units as the load. This is equivalent to Prime95 FFT size between 96k and 120k. HT was turned off as it doesn't benefit this task. After allowing to run for a long time, hottest core was 74C average, 77C max. Coolest core was 67C average, 69C max. Ambient was 22C.

Now I might not have left it long enough to stabilise yet as it has been running for 25 minutes so far, but the chart in AIDA64 looks pretty flat and I'm about to go to bed. After over 20 minutes running, I'm seeing hottest core 64C average, 66C max, and coolest core 59C average, 61C max. Ambient has reduced slightly to 21C. Allowing for that, this is an improvement of 6C on the coolest core and 9C on the hottest. Not bad at all.

I'll leave it running overnight anyway but this does appear to be a nice improvement.