4790K overheating with NH L12

[Sylvester]

I am using a 4790K at stock with a Noctua NH-L12 in a mATX (LianLi PC V354) case with plenty of airflow. Mobo is a Gigabyte Z97MX Gaming 5, latest BIOS (F4).

Hexus' own review of the NH L12 shows it to be an adequate cooler for the ivybridge i7 3770K, getting 70.5°C under Prime95 15 mins full load (only 5°C above the Corsair H60) and its small enough to fit the case which has limited space due to the PSU placement. After installing I found idle temps in Windows7x64 ultimate were about 8°C above ambient which is about right.

But under load at stock I am getting 100°C in Prime95 with this set up within seconds and it lasts about 3 mins in OCCT before it shuts down due to one or other core overheating (>85°C). Obviously this is way too hot. Real world gaming max temps are about 75°C in Core Temp playing Elite : Dangerous beta for example (not very CPU intensive) which is way too hot and indicates spikes will be much hotter in CPU intensive applications and games which could cause thermal protection throttling, shut down or damage the CPU.

I have already checked the seating which was good and reseated with fresh thermal interface material which is Noctuas homebrew NT-H1, which is up there with AS-5 and the problem continues. Also set the HS fans to max in BIOS in cpu headers 0 & 1.

I have looked around the net and I am seeing reports of people getting very hot 4790Ks and RMAing them, so I am wondering, is this chip just way too hot or is the cooler completely inadequate ?

I am wondering if I need to raise this as an RMA with SCAN or splash out and get a Kraken x40 and put the rad in a push pull on the top 140mm extraction fan point. Hope that makes sense. Thanks for your thoughts.

[shaithis]

What's the temps like if you remove the case lid? If you prime temps drop to ~75 then you know it's case cooling.....if they still remain high, then your TIM hasn't been applied correctly, the heatsink isn't flush or the chip has an issue.

[Sylvester]

OK thanks shaithis, your points are worth considering.

I left the side panel off after reseating the HS anyway so temps as above, up to 100°C. There are 4 fans on the case, 2 in and 2 out, plenty of air, has a 770 in there with it, but better out flow than in. In any case the GPU is not a factor benching the CPU.

The contact pattern in the TIM on the CPU was symmetrical so the base plate is correctly mounted. There was TIM over all the CPU surface though a fraction shorter at one corner then the other three where the pressed steel plate curves away, but it just didn't have as much excess there is all. Under the contact surfaces it showed a kind of hourglass shape of slightly different thicknesses but it was completely symmetrical, mirrored top and bottom and left and right with slight excess at the edges from the spread of a half pea blob. So it was correctly seated IMHO but I am not entirely sure about the contours of the CPU plate that does seem to have a few microns symmetrical warp due to the manufacturing process being more appropriate to car body parts (if a £50 HS can have a machined mirror flat base why cant a £250 CPU have a machined flat lid ?) But such as it was, it was completely in contact and as thin TIM as it could go because of the pressure, both securing screws done up in alternate half turns and tightened to their full extent.

So I guess I am saying I have ruled out the seating and the airflow to my satisfaction.

I have underclocked the CPU to 3.6 and disabled turbo and manually set vcore to 1.1v and at those settings it runs prime95 at about 70°C. Still surprisingly hot. So if the Noctua NH-L12 cooler is reckoned to be up to the job I guess this chip has an issue.

[Sylvester]

I have been checking the Noctua heat pipes and they are not all the same temps. Is that normal? I did somewhat ease the heatpipes a bit due to the RAM clearance so I may have damaged the base or something. I think maybe I am going to have to try a hydro before doing an RMA just to be sure.

[Domestic_Ginger]

The chip will only be a portion of what's under the heat spreader. I imagine one part will be cooler. Could you rotate the cooler 180 to see if the colder pipes get hotter. Just wondering. The CPU sounds a bit poorly, from what you've written it wouldn't work under the stock hsf.

[Sylvester]

You may be right, I have had a little think about it and the truth is that the heat drops very rapidly, less than a second to go from 65°C to 46°C during an underclocked OCCT test, and then one minute to go to 38°C, which it would not do if the cooler was not working as intended.

Here is the OCCT graph for Core #0.



This is comparable to the NH-C14 cooling on my QX9650 which takes a minute to drop from 40°C to 32°C after stopping Prime95 which maxes out at 64°C.

Likewise when I start a test the heat builds up immediately and then it plateaus, which suggests that the cooling is ongoing. So scientifically speaking, it looks like the cooler is working even if it doesnt feel like it to my finger tips.

[watercooled]

I'm not sure what the thermal profile of this CPU is like, but a very rapid temp increase immediately upon load would suggest there could be poor contact somewhere along the path. A big cooler like the one you're using has significant thermal inertia, so it should take some time to ramp up temperature until an equilibrium is reached.

For comparison, my 1055T under a Hyper 212+ only increases by a couple of C when applying full load, then ramps slowly over a few minutes before reaching a stable temperature. The 1055T is soldered to the IHS vs using TIM, but I was under the impression the 4790k was meant to improve that massively...

[Sylvester]

Well I am not sure either watercooled.

So this is a comparison chart for the QX9650 with NH-C14 which I guess is working because the max temps are reasonable.



I am not an expert but the heating and cooling curves are similar in form, with sheer sections at activation and deactivation of the test and both have comparable heating and cooling ramps through the test, though the QX9650/C14 is only a bit slower its a bigger mass of CPU/HS, so I am not sure if I am seeing any clues there.

But there is a much bigger max temp for the 4790K and a bigger difference in the way the OCCT activity changes are affecting temps, as it has those little surges every other minute which show up more than with the QX9650, though overall it ramps up the same over the first minute there is a bigger total leap in the first few seconds.

The size of the leap suggests there is either a cooling bottleneck or an excess of heat output. I dont know which. Is it greater power output changes from the 4790K due to 8 threads? Or is there more boost technology in the later So1150/Z97 mobo even though turbo is disabled? I know LLC is on for example which I have deactivated on the QX9650.

If a bottleneck I am reasonably sure it is not the contact region, so I am wondering if it is in the cooler base to pipe connection, or the CPU thermal interface? Considering how solid those pipes are in I am inclined to wonder if what I am seeing with the assymetric heat load on the pipes could be an assymetric heat output from the CPU. I do know that Cores 0 & 3 are about 2-3°C cooler than Cores 1 & 2, if they are reporting correctly, not a lot but its something. Though each pipe is at a different angle and the bottom pipe is hottest, which drops down when moving away from the base to the fins, there is one orientation you must not use the HS so it makes a difference (capillary-wick vs gravity). The hottest pipe at the CPU is not necessarily shifting the most heat away from it.

Ideally I would test the L12 on the QX9650 but I have a liquid cooling set up on the NB threaded through the C14 so it would be more than a pain. The other thing I could do is test the stock HS/F and see how that goes.

[watercooled]

I posted that quickly on my way out the door, but something I forgot to consider, is I'm using the CPU sensor rather than the core sensors to check temperatures, as the reported core temperatures are weird on AMD CPUs. IIRC though (but don't quote me on it), they're reasonably accurate for judging temperature delta, just not absolute temperature i.e. offset is constant.

For what it's worth:



For comparison, what does the CPU sensor look like on your system?

[Sylvester]

OK my hosting service FTP server has gone AWOL but I repeated the 4970K OCCT test with the stock Intel heatsink, open chassis, (side panel laid down on the floor and hooked up, window open, nice and cool) and the CPU underclocked to 3.6GHz, no turbo, 1.1v as before.

With the NH-L12 this setting maxed out as per my first screeny at 69°C and did a 4 minute test easily.

With Intel stock, carefully seated etc the temps soared to 72°C within 20 secs and by 65 secs the test stopped because it hit 85°C.

I think this shows that the CPU is making the heat and that the NH L12 cooler was working just fine.

CPU was SR219 MALAY L418C133, had some nice closeups of the thermal compound contact pattern as well but until the FTP comes back I cant share

[watercooled]

Perhaps the CPU is faulty then, especially since you mentioned you've read about people RMAing them for similar problems - the fact the supplied cooler can't cool it adequately at stock settings doesn't sound good to me.

Maybe the die-IHS gap as mentioned here is still causing issues on some DC samples?

[Sylvester]

The CPU works perfectly well, completed 6 passes of 32Gb in Memtest86+ without error under the NH-L12, albeit at over 60°C. So I would tend to agree the lid is probably the problem and that article is very interesting thanks. (Noob question how do I give you an official thanks vote watercooled?)

But if it is the lid then its still a faulty product and I don't like the gamble of delidding a £260 CPU when it deserves an RMA, as I am pretty sure delidding would void the warranty and I have never done it before, also my Scansure has expired. But I have contacted SCAN now with a view to an RMA and gave a link to this thread.

I would only give delidding a go if all else failed. I have cooled a flip chip before now and could just shave a mm or two off the HS mounting spacers to bring the NH-L12 down to the silicon. I have a vice and a minisledge! But thinking about it the NH-L12 is only retained by two screws so it would have a tendency to rock on the silicon strip if moved. Really needs a 4 point mounting system for bare silicon.

Its just a bit disappointing that this enthusiast grade CPU is still using such a rubbish IHS system after all the hype.

[watercooled]

There's usually a thanks button at the bottom left of posts under the picture etc, but I think you have to have a certain post count before it's permitted. No worries either way.

Oh no I definitely wouldn't recommend delidding the CPU; if it's causing you problems then I'd recommend the RMA route. After all if you wanted to run at stock and/or didn't care about the better TIM the 4790k is supposed to offer, you could have gone with a cheaper model.

On the subject of hype, I agree this must be frustrating considering this is one of the issues it's actually meant to address, especially when you're paying a premium for it.

[Sylvester]

FTP has resumed service. This was the stock Intel HS underclocked and undervolted.



This was the contact pattern from the NH-L12. As you can see full coverage and no bubbles or blank spots. This thing is just too hot at stock settings. If I send it to SCAN for assessment I am pretty confident that running Prime95 small FFT will vindicate my concern.

[watercooled]

Well if you choose to RMA, I'd explain the issue in detail, maybe post in Scan's forum and link them to this thread so they know what to test for and issue an RMA based on that.

[Sylvester]

Good point, started a thread there.

I found a thread in which someone apparently has very similar problems and is treated to a fairly severe judgement from SCAN, though I sense the hand of Intel in this.

So I had it RMA'd and the Scan Engineers tested it today. Guess what they had to say:

Quote:
Thanks for your enquiry,
Your Item Intel Core i7 4790K s1150 has been tested and no fault has been found.

Please find below the testing notes by the engineer:

Tested Prime95, Core temp max 95C, clocks between 4.0 ando 4.3 GHz.

You will be charged £10+VAT for collection, £10+VAT for testing and £10+VAT for re-delivery

Please reply directly to this email with your telephone number and a suitable time window of when we can take a payment from you over the telephone.

Is it just me who is shocked that anyone can say that a stock clocked CPU operating within 5C of melting is OK. You can see from their results as well that it never even reaches intels 4.4ghz as sold.

They wouldn't even tell me what CPU cooler they achieved those results with for all I know thats with a custom watercooling loop.

And don't even get me started on the number of times I had to repeat myself when explaining the issue.

Luckily for me I was still within my DSR rights period.

So before deciding RMA I need to know how they are going to test it and what the pass criterion is.

His was a bit cooler than mine which is reaching thermal throttle temps on two cores in Prime95. Intel might argue its supposed to be like that and thermal throttle exists to prevent damage plus I have a 3 yr warranty so why worry. My consideration is what happens when the warranty runs out, this will shorten the CPU life, as Intel's own thermal spec for the CPU is 72.5°C, which it clearly exceeds by a large margin and which SCAN are apparently not using as the fault criterion.

IMHO if the CPU is reaching thermal protection limits (100°C) with adequate cooling then it is a trading standards issue as the CPU is not running as advertised and therefore also not fit for purpose.

From my point of view, some of these CPUs are like this and some run cooler. If I can get an RMA test approval I may get a cooler replacement CPU but I gamble losing £40 in the process and my hot CPU back if the test passes the CPU as not faulty.

At 100°C the delta with ambient is high and small differences in cooling efficiency could make all the difference as to whether it passes or fails the test in SCANs lab. My house ambient is warm, 27°C+ due to my personal requirements so it may produce a different result in an air conditioned 23°C test lab.

It is, as they say, a question of degree!