Skylake OC not going so well.

[pantsaregood]

Upgrading from an i5-2500K running at 1.33v/4.4 GHz for 4 years stable using a Hyper 212+. Runs below 85C unless I throw it under the bus with AVX-capable LinPack.

My i7-6700K on an EVGA Z170 FTW isn't doing too well in comparison. 4.5 GHz at 1.34v is stable, but my temperatures are pushing 90C. I'm using a Hyper 212 Evo, which isn't high-end by any means, but it is a functional cooler.

An odd aspect of this is that the heatpipes don't really appear to get warm. I've repasted the thing three times under the assumption I had done a poor job, but I'm still having no luck. Is it possible that the thermal material between the die and heatspreader is so bad that it is bottlenecking the cooler's ability to function? Even after 15 minutes of running at 90C, the copper heatpipes aren't really warm.

I have read a few reports that using adaptive voltage can make stability tests flip out a bit. I am using adaptive voltage right now, so if necessary I can avoid doing so while stability testing.

I'd like to try to push at least 4.8 GHz out of this thing, but I can't think of what else I could possibly do to make it run cooler.

 

[lehtv]

That's because Sandy Bridge has a soldered on IHS, while Skylake has poor quality thermal paste under the IHS. In other words, heat transfers much more easily from 2500K cores to the cooler's heat pipes, allowing it to run cooler even though it consumes more power.

The only ways to substatially improve temps in your case:
1) buy a better cooler
2) lower voltages
3) delid the CPU and replace the thermal paste with some better quality stuff like Coollaboratory Liquid Pro

Option 1 is unlikely to acheive the results you're hoping for. Option 2 is probably the wisest thing to do, read below. Option 3 will have by far the best results but involves voiding warranty and of course, when modding your CPU, there's risk involved.

I would avoid high voltages on Skylake until we know for sure where the safe limits are. 1.3V is going to be my absolute max for the first year, then maybe I'll delid and re-evaluate the situation. This guy apparently killed some part of his 6700K (possibly the integrated memory controller) by stress testing Prime95 at high volts and temps.

 

[pantsaregood]

I understand and am aware of that, but the heat transfer seems exceptionally bad here. The heatpipes failing to even get warm strongly implies to me that the TIM is pretty much failing to transfer heat from the die to the IHS. I expected higher temperatures, but this is absurd.

Any recourse short of delidding?

Also, the crazy thing is that I'm almost positive option 1 would be pretty ineffective. The Hyper 212 Evo isn't even getting warm, which implies that it isn't particularly close to its thermal capacity.

Not really worried about accidentally frying the CPU. I got Intel's performance tuning plan just in case.

 

[lehtv]

Any recourse short of delidding?

Apart from options 1 & 2 above, no.

 

[Bearmann]

I'm no expert, but maybe you're not cleaning off the old paste good enough from the CPU and/or heatsink before applying the new.

 

[Shmee]

If the cooler is not getting warm, maybe reseating it is worth a try.

 

[frozentundra123456]

I am not really an overclocker, but just intuitively from what you said, it doesnt seem that bad. About the same overclock, same voltage, and nearly the same temps. What is the problem, if it if fact stays below 90?

 

[Yuriman]

You could lap one or both surfaces. 90c is about what you'd expect with Haswell running an AVX load with those clocks and voltages though. I recommend using a non-AVX stress test to get more realistic results.

 

[Burpo]

New BIOS available 4 days ago..
http://forums.evga.com/EVGA-Z170-ClassifiedFTWStinger-BIOS-104-Released-m2387637.aspx

 

[ehume]

@pantsaregood: I had the same issue with my i7 4770k. The chip runs fine as long as it is not producing too much heat for the IHS/CPU TIM to handle. If you get really granular about it, you will hit a point where the core temps will shoot up. At that point, no amount of cooling will help you. I tried an NH-D14, at the time on of the best heatsinks. It did not help. I had to back down on the OC. I thought, Heck, and put in an Armageddon.

This is the reason I have decided to wait for Kaby Lake. I had the impression that Skylake is like the pre-Devil's Canyon Haswell: a lousy overclocker.

BTW -- I did get a DC for testing heatsinks. It has a much better TIM. I had reason to buy a new DC yesterday. It is running cooler still.

So your heatsink is fine. You will simply have to moderate your OC hopes. Trim your OC back until you get good temps. At least you will know that the temps are what is holding you back. You will be rock stable. This will become your production machine, with no e-peen.

 

[pantsaregood]

Anyone have any experience using Intel Extreme Tuning Utility or x264 for stability testing? I've seen some places advocating them because they stress CPUs sufficiently without causing absurd amounts of heat. Apparently XTU even makes use of AVX/AVX2/FMA3 without the nasty overheating issue.

 

[YBS1]

Seeing that you've tried multiple remounts with no improvement and the heat pipes aren't even getting warm it would appear you're assumption of poor IHS/TIM mount is likely. Delidding is the best option but not likely one you'll use with a new cpu seeing as you've purchased the tuning plan. I would maybe try adding a shim of some kind to increase mount pressure.

 

[pantsaregood]

After playing around with it a bit, I believe I can confirm that the TIM between the die and IHS is the issue.

At stock speeds and voltage, benchmarks known to generate loads of heat (Linpack and Prime95 28.5, for example) peak out at about 80C. Pushing up to 4.4 GHz with an increase of 0.01v peaks out at around 84C.

After passing this point, temperatures begin to rapidly go out of control. Attempting 4.5 GHz with an additional 0.02v (that's +0.01v compared to stable 4.4 GHz) shoots temperatures to about 97C with Prime95 28.5 within a few seconds. The CPU isn't stable at that voltage, so it crashes within a minute or so. Regardless, this makes it look like the TIM/IHS combo is simply hitting its thermal capacity.

I've kept up with Ivy Bridge and Haswell temperature issues, but I didn't realize they were this bad. Even when not under load, the heatpipes on my Hyper 212+ on my 2500K feel warm to the touch.

I suppose I'll be delidding soon.

 

[MongGrel]

After playing around with it a bit, I believe I can confirm that the TIM between the die and IHS is the issue.

At stock speeds and voltage, benchmarks known to generate loads of heat (Linpack and Prime95 28.5, for example) peak out at about 80C. Pushing up to 4.4 GHz with an increase of 0.01v peaks out at around 84C.

After passing this point, temperatures begin to rapidly go out of control. Attempting 4.5 GHz with an additional 0.02v (that's +0.01v compared to stable 4.4 GHz) shoots temperatures to about 97C with Prime95 28.5 within a few seconds. The CPU isn't stable at that voltage, so it crashes within a minute or so. Regardless, this makes it look like the TIM/IHS combo is simply hitting its thermal capacity.

I've kept up with Ivy Bridge and Haswell temperature issues, but I didn't realize they were this bad. Even when not under load, the heatpipes on my Hyper 212+ on my 2500K feel warm to the touch.

I suppose I'll be delidding soon.

Wow, really.

Interesting information.

Curious to see what happens in the future.

http://www.overclock.net/t/1571038/confirmed-my-6700k-is-dead

Yeah, ouch.

 

[LTC8K6]

Is the 212Evo a new cooler?

Is it possibly defective?

 

[pantsaregood]

Is the 212Evo a new cooler?

Is it possibly defective?

Doubtful. It's a set of four copper pipes pushed through some aluminum fins. Aside from a bad mounting surface (it's pretty flat), there's not a lot that could go wrong. It has to be pulling some heat from the CPU, as without it the CPU would overheat very quickly due to lack of surface area on the IHS.

There is heat being transferred to the heatsink, but it can only transfer as fast as the TIM transfers from the die to the IHS. The heatsink isn't getting warm to the touch because it isn't even being pushed close to its thermal rating.

A liquid cooler wouldn't be any more effective in this case - liquid coolers have a higher capacity for heat dissipation than air coolers in general, but the current cooler isn't at its limit. I suppose using some type of sub-ambient cooling could potentially help with the bottleneck, but I'm not interested in taping a large diamond that's chilled to -200C to my CPU every few minutes.

Edit: I am aware heatpipes are more than just slabs of copper, but if the fluid/vapor mechanism was failing to function, the copper in contact with the IHS would still heat up.

 

[LTC8K6]

Doubtful. It's a set of four copper pipes pushed through some aluminum fins. Aside from a bad mounting surface (it's pretty flat), there's not a lot that could go wrong. It has to be pulling some heat from the CPU, as without it the CPU would overheat very quickly due to lack of surface area on the IHS.

There is heat being transferred to the heatsink, but it can only transfer as fast as the TIM transfers from the die to the IHS. The heatsink isn't getting warm to the touch because it isn't even being pushed close to its thermal rating.

I know what a 212evo is, I own one.
Along with a TX3.

I meant, have you used it before? Did it work okay before?
I have heard of 212+ that were defective.

If they forgot to put the fluid in the heatpipes, it wouldn't work too well, but it would work.

 

[MongGrel]

I know what a 212evo is, I own one.
Along with a TX3.

I meant, have you used it before? Did it work okay before?
I have heard of 212+ that were defective.

If they forgot to put the fluid in the heatpipes, it wouldn't work too well, but it would work.

That's actually a very good question.

Have you used that cooler on another build or is it a new one ?

It could be a defective 212.

But he is probably right about the base not heating up.

 

[Ranulf]

I've kept up with Ivy Bridge and Haswell temperature issues, but I didn't realize they were this bad. Even when not under load, the heatpipes on my Hyper 212+ on my 2500K feel warm to the touch.

I suppose I'll be delidding soon.

This would indicate it is not a faulty heatsink.

 

[cytg111]

well sandy die was like 200-ish mm2 and skylake is 120-ish .. generating about the same amount of heat pr tdp as indication. Obvious fact is obvious that it's getting harder to dissapate that heat.. Im surprised they're still able to keep it (aboutish) under 100 as with the 22nm sku's... one way or the other they must have gotten better at moving those joules around 22nm->14nm.

 

[LTC8K6]

This would indicate it is not a faulty heatsink.

The 212+ is not the heatsink in question, though.

Putting the 212+ on the Skylake chip would be a good test, though.

 

[CakeMonster]

4.5 is pretty decent but I can understand that you want a higher numerical value given that earlier generations oc'ed to similar frequencies. But you still got a very nice IPC increase with your upgrade. The limits seem to be in the 4.5-4.8 range for SL for pretty much everyone.

The IHS is what it is unfortunately. I'm not happy with it either but I considered everything and found that getting a 6700K would provide me value and I think it did. I'm running at 4.5GHz too and I haven't really played with it but I seriously consider undervolting because the default mobo settings run amok with voltage even at stock speed.

 

[MongGrel]

4.5 on a SL is probably pretty rocking I'd imagine.

Considering the instructions on the chip itself is a lot more advanced I imagine.

 

[Deders]

Ok, a few things I noticed when overclocking.

1: My old cooler (Thermaltake Contac 29) that worked very well on my oc'd i5-750 (60c max and silent with Noctua fan) was no longer enough, partly due to the added effect of the bad tim between the IHS and CPU. Now a Noctua DH-15 keeps temps lowish (77c max@4.7GHz)

2: how does your adaptive voltage work on the EVGA? on my Asus it is different to Offset voltage, instead of offsetting, it specifies the max voltage that is applied when cores are running full speed. I found this to be much cooler than offsetting as it lowered the core voltages as well as the VID. Before I used this the core voltage was much higher than the VID which was unnecessary and just added to the heat.

3: What are you LLC options? you should definitely use some at 4.6GHz and above, again it will affect the Vcore and Vid voltages.

 

[MongGrel]

Ok, a few things I noticed when overclocking.

1: My old cooler (Thermaltake Contac 29) that worked very well on my oc'd i5-750 (60c max and silent with Noctua fan) was no longer enough, partly due to the added effect of the bad tim between the IHS and CPU. Now a Noctua DH-15 keeps temps lowish (77c max@4.7GHz)

2: how does your adaptive voltage work on the EVGA? on my Asus it is different to Offset voltage, instead of offsetting, it specifies the max voltage that is applied when cores are running full speed. I found this to be much cooler than offsetting as it lowered the core voltages as well as the VID. Before I used this the core voltage was much higher than the VID which was unnecessary and just added to the heat.

3: What are you LLC options? you should definitely use some at 4.6GHz and above, again it will affect the Vcore and Vid voltages.

Your rig is making me drool, looks a bit like an upgraded version of what I have running on the CPU and a few things.

Nice setup.

:thumbsup::thumbsup: