Changing Thermal Paste Incorrectly?

[Kirbyarm]

Hello everyone, I'll try to explain what is going on in full detail.

The CPU is an Intel i7 4770k that has been in use for approximately 3 years now.
This CPU seems to have come out of the box with some sort of stock applied turbo mode because in CPU-Z it seems to dynamically go anywhere from 3.5 Ghz - 3.9 Ghz on its own.

Basically, about 6 - 8 months ago roughly, my friend and I wanted to attempt overclocking my CPU. I've never done it before, but he has countless times for many of his friends and hasn't run into any issues with them. So we did some things in the bios (I can't remember what exactly, he walked me through it over the laptop, while the desktop was undergoing these tests). Anyway, we go it up to I believe 4.2 Ghz but during the stress tests with Prime95 he decided to abort because he absolutely did not like the temperatures he saw at all, being in the high 90s and even hitting 100C and 101C if I remember correctly (this was a while ago so I don't remember precisely). Anyway we reset the settings back to what they were. Mission aborted.He said re-apply the thermal paste and we'll try again. He strongly believed that is the most likely culprit to the problem.

Fast-forward to yesterday, I've been having problems with a new game running 40-45 fps despite packing a GTX 1080 Ti, 32 GB of RAM and this processor. Other users of the game with far inferior parts, CPU included, are running the game 60 fps smooth constantly, or so they claim. Now the game is in EA so obviously I don't think it's actually my CPU, but nevertheless I wanted to finally get around to getting that 4.2 Ghz overclock working with my friend again.

Here are a couple of pictures: first is of the temperatures under normal load (with the game open and other applications on normal day to day use). Second is temps during I believe was a stress test in Prime95. Third is just a CPU-Z readout.

https://i.imgur.com/4844kfA.png (pre re-apply temps - load)
https://i.imgur.com/OUPYdY7.png (pre re-apply temps - stress)
https://i.imgur.com/ZH5TZys.png (cpu-z)

Okay, so he gave me the go ahead to open my machine up and re-apply the thermal paste. I, however, took extra steps by taking photos of the cooler bottom and CPU top prior to wiping them clean with windex, cotton swabs and paper towel (best I have available unfortunately). These are those pictures:

https://i.imgur.com/2YNIiS6.jpg (paste pics before wipe)
https://i.imgur.com/gbmhBNH.jpg
https://i.imgur.com/joPgZsK.jpg
https://i.imgur.com/CmT2Dxd.jpg
https://i.imgur.com/7maFNoi.jpg

Wiped it sterile as I possible could, put the "line" of paste exactly the shape and approximate size I saw in the linus tech tip video, and took a quick picture of it before seating the cooler on top. Here is that picture:

https://i.imgur.com/8KvlmBD.jpg

Put the cooler block ontop of the CPU, and as friend said, put gentle pressure on it for about 20 seconds, then started screwing it into place. The first screwed felt good and had a good number of turns before it got tight (but not too tight obviously). The second screw felt different, like it only had 33% as much turning before it got tight.. not sure if there was an issue with that but it was worrying. Anyway, cooler back on, GPU back in place, I booted up the system. Friend set to let it run under normal load for 8 hours before we do more tests again. So I went to sleep. Here are the result pictures I got as soon as I woke up:

https://i.imgur.com/NxqecZT.png (64 / 74) (after sleeping with game load on, not torture test)
https://i.imgur.com/CuVy2hs.png (99 / 100) (after sleeping with game load on, torture test)

Obviously this is shocking and alarming for me and I have virtually next to no experience operating on computers (I've only built this one machine, one time, and removed the cooler a total of twice, once about 6-8 months ago to add in 32 gigs o f RAM, and once again early this morning) so I get really nervous around the components and my hands aren't steady at all in constant fear of breaking something or getting particles or something where they ought not to be.. etc.

This is when I decided to post this here and hopefully some experts on Tom's Hardware can help get some ideas churning about what the problem could be, or problems in general about this whole situation over the course of the last year and if you think there might be a problem with my chip? or something else?

Other hardware monitors seem to detect my motherboard (MSI Z87-G45) and SSD (samsung evo 840) and GPU (GTX 1080 Ti) stay very cool. My tower case is a Cooler Master Cosmos SE with 1 top fan (push), 1 rear fan (push) and 2 front fans (pull). The cooler (Noctua NH-D14) fans point the airflow toward the rear fan. When I stop stress tests and load, the temps cool within 5 - 10 seconds back down to 40~ range. So cooling down doesn't appear to be the problem in my eyes, but again I'm no expert here at all.

What do the CPU and overclocking experts here think of this situation?

Thanks for reading and thanks for your time in advance.

 

[Karadjgne]

What cpu cooler?

You have a Haswell cpu. You ran Prime95. That can be trixy. With Prime95 version 26.6 or prior, the torture test didn't include much, if any, AVX instructions as cpus of Ivy-Bridge and prior didn't have AVX capability, that was reserved more for Xeons and other office/server type cpus. Haswell and newer does have AVX and any version of Prime95 after 26.6 uses high amounts of AVX. Normally this wouldn't be an issue, except AVX instructions go well beyond a normal 100% gaming load, much closer to equitably pushing a Haswell cpu at 130%. This is a totally unrealistic expectation from a gaming or normal app, household pc as AVX instructions are only used in certain specialist professional programs.

http://www.mersenneforum.org/showthread.php?t=15504

You'll find 26.6 here for windows 64bit.

Run small fft torture, it'll give a 100% gaming equitable cpu load and appropriate temps.

Generally, Intel cpu safety line is 70°C for that torture test, anything under is golden, 70-80°C and its time to fiddle with fan curves or other possible airflow issues, 80°C+ and your cooling solution is inadequate.

 

[geofelt]

At idle, I would expect to see a temperature of 10-15c. over ambient.

From your photo, I think you have put on too much paste.
The purpose of paste is to fill in microscopic pits in the mating surfaces and drive out air which is a bad conductor of heat.
The paste, itself is better, but not nearly as good as metal to metal contact.
I find that a rice sized grain in the middle will spread out under heat and pressure.
It is hard to use too little.

In the mounting process, you need to tighten down the two screws gradually, and evenly.

When overclocking, realize that your results will differ, depending on your luck in getting a good chip.
You hear about the 4.7 results, but those with dog chips remain silent.

The heat generated will be determined by how high your vcore needs to be to sustain the desired multiplier.
Monitor vcore with cpu-Z. Up to 1.3v should be ok.

 

[Kirbyarm]

@Karadjgne the cooler is Noctua NH-D14

I have downloaded 26.6 and it's ready to try.

@geofelt 10 - 15 C when you're not doing anything really on the PC? Is that what you mean by idle?

Like right now, nothing open but steam and firefox for this page and it's floating around 6.5% CPU usage and 27 - 29 C. I have never seen it below 20 in its lifetime.

Also, I just finished re-applying the thermal paste again.. Tried the peadot method this time. I could be remembering wrong but I don't remember ever seeing the temperature below 30 C either. And this 27 - 30 I don't think I've had it that low before.

Going to try running the game for a few minutes to test the temps then I will try 26.6 prime95 small fft torture test. I'll reply shortly.

 

[Kirbyarm]

These are the temperature readings under light to normal load (in relative to what is normal going on with the PC at all times): https://i.imgur.com/ajyfyzv.png

Now trying the stress test. The results after 5 minutes are: https://i.imgur.com/aY2xcFP.png

I don't know how reliable these tests are because I've never pushed my CPU this hard to make it so hot during normal use.

This is using the 26.6 version provided above.

 

[Kirbyarm]

I should mention that after this second re-apply, the temps were better than prior re-application that was done last night/early this morning. By a good 10 degrees I'd say. So we're confident it was correctly re-applied this time. Again though, I'm not an expert. I don't know much about this stuff.

 

[Karadjgne]

96 on a D14 at stock values, even on that i7, is quite high. Actually very high. I'm wondering if the fans are actually working as intended. That cooler is good enough that you should be in the 60's-70°C at best on that torture test. It's almost like the fans are locked at specific rpm, or are you using the ultra-low adapter?

 

[Kirbyarm]

What is the ultra low adapter? How can I unlock specific rpms and have them dynamically spin as needed?

 

[Kirbyarm]

We are testing an overclock right now. The RAM was apparently on [Disabled] for XMP and he thinks that might be why it froze on test one with settings attempt 1 (oc settings) and "your pc has run into a problem and needs to restart" on test 2 with settings attempt 2 after a 10 minute blend test.

Now it's been running the blend on settings attempt 2 but with the XMP set to [Profile 1] and my friend seems confident that was the problem.. blend has been going about 15 minutes now and this is what I see: https://i.imgur.com/rLEudbB.png

 

[Kirbyarm]

Got a stable clock speed of 4.2 Ghz on the CPU. We had to tone the DRAM frequency down to 1333 Mhz, and set the timings to 9-9-9-24. Survived a 30 minutes AIDA test, after quite a few computer freezes and blue screens. Finally stable. Tested opening 2 games, 2 servers, a couple dozen or so applications all at the same time and it seemed just fine.

The max temp it reached during AIDA test after the 30 minutes was core #1 reaching 95 C. The temperature floated around 88 - 90 C most of the time for the cores at package sensors.

So I guess my initial concerns about my CPU are still in play here and relevant to the discussion. The temperature as Karadjgne pointed out, with my cooler at stock settings was running very hot. It isn't the thermal paste, so why could it be so much hotter than the average 4770k? IHS? Airflow? Cooler fan speed locked as suggested?

Any suggestions on how to test for some of these things? I'd love to invest in a better cooling solution, but I need to know the problem first.

 

[PdxPetmonster]

What thermal paste are you using? If it's generic, it could be that adding a few C to the overall temps. Also, you talked about over-clocks, but never mentioned at what voltages those over-clocks were at. High voltage is what generally leads to high temps, combined with a cooler not seated 100%, and you can end up with those temps. Make sure you've got any fan profiles in the bios set to max, as the fans on the cooler may not be running optimally. You could also get a couple molex fan adapters and plug the CPU fan(s) into those, as it would put the fans at their max rpm. You mentioned one of the screws felt different when you attempted to screw it in, can you describe that some more? I'm wondering if you've got pressure on 3 points and not 4, leading to a quadrant of the CPU not having solid contact for dissipation.

 

[Karadjgne]

Since you have new paste, there's no possibility of it being dried out. So, I'd try a fitness test. Literally grab the entire cooler 1 handed and see if it'll rotate at all. If it's seriously solid, I'm doubting the mount is an issue. If it moves at all easily, theres an issue. I've seen many cooler mounts come with 2x sets of standoffs, one for Intel (shorter) and one for amd (slightly taller). Sometimes these standoffs are colored, on my cooler the Intel are chrome and the amd are black chrome, but using the taller amd standoffs would have been a serious mistake as the cooler will now not be snug, but loose on the cpu, no matter how tight the screws.

A LN or ULN (low/ultra low noise adapter) is generally included by Noctua for use with their fans. Essentially what it does is cut down the 12v constant to 9v or 5v constant, so instead of your fans running at full speed, they'll max out at @60% or 40% of available rpm. With Noctua they look like 3 inch long fan extender wires, nicely braided with connectors on each end. The NH-D14 uses 3pin fans, so unless the motherboard allows analog voltage changes on the cpu_fan header (most boards are cpu_fan PWM only) you'll have 0 control of fan speeds, so the adapters were to slow the rpm down, and reduce any noise, for those who had no use for max rpm fans.

So, if you did use the adapters to slow down rpm (even by accident) then nominal loads would be ok, but extreme loads like torture testing would get shortchanged on cpu airflow, and temps would skyrocket.

MSI has Control/Command Center software, which does include a somewhat simple fan curve setting for the different headers. If that is enabled, it'll override bios settings or other software. If it's not set up right, it's entirely possible your fans are not reaching potential, basically running at idle speeds for any testing.

I7-4770k maximum vcore is 1.3v. And you should only be getting that kind of setting at @4.7GHz or above. For 4.2GHz, unless you really, really got unlucky in the cpu lottery, you should be far closer to @1.1x volts. But vcore isn't the only thing to OC, with Haswell there's also ring voltage, sa agent, xmp, power phases, LLC, current max/min times etc.

There's enough leeway at factory default optimized settings that 4.2GHz should require nothing more than locking the cores and bumping the multiplier.

My i7-3770K will run 4.9GHz stable at 1.32v and under p95 26.6 small fft not break 72°C on my 280mm aio. I5-3570k at 4.3GHz stable at 1.108v and 70°C with a hyper212 equivalent 120mm aio.

1.3v for that cpu is very high for 4.2GHz.

 

[Kirbyarm]

I'm using the NT-H1 "premium thermal compound." I believe it was provided with my CPU cooler, the NH-D14.

There are max RPM adapters too? Hmm. Well I wouldn't want them forcibly blasting maximum RPM all the time either. I want them to dynamically rotate the fan blades at speeds as needed under load. Could this be accomplished in the bios or with some trusted lightweight software?

I re-applied the thermal compound a second time the very next morning, because of the temperatures skyrocketing after the first re-apply as well as the screwing in issue. As explained above, this time I screwed both screws in gradually and together. It was overall a much smoother process and helped my confidence with it a bit more. So that problem isn't a problem anymore since the paste and re-apply process was successful the second time.

That said however, as said above by a CPU expert, my temps are "very high" for my chip when it was at stock settings under those stressful Prime95 tests. Almost made it sound unusually high.. this I don't know what the culprit is. It's said to be either the cooling solution is inadequate, which I don't really see likely with a NH-D14 and a big Cooler Master tower case with lots of space for airflow.. or RPM related of the fans according to the user above. I didn't see any special adapters between the fan connectors and the thing you connect them to (molex it is called?). I don't know how or where to check for RPM or RPM profiles. Please CPU masters help me if you know!

 

[Kirbyarm]

Thanks for the reply, Karadjgne!

I will try the physically turn the cooler test later tonight, but I am quite a physically weak person. It felt pretty snug, looked flush and parallel with the CPU socket but I'll give it that extra umph this evening.

99% sure that I am not using any such low-noise adapters. I described to my friend what I saw when inspecting them yesterday, and it seems like the standard connectors.

What exactly is this 'fan curve' that keeps getting mentioned? It really doesn't sound like something it optimally cooling properly based on your analysis of my stock temperatures..

As for the overclocking to 4.2 Ghz, I apologize but I can't remember the exact core voltage and ring voltage I set it too. I'll see if CPU-Z can tell me, otherwise I'll mark it down to check when I do that cooler test. I'm pretty sure it was something like 1.215V core voltage and 1.075 ring voltage. The DRAM I remember we set the frequency down to 1333 Mhz and changed the timings to 9-9-9-24 @ 1.5V voltage.

According to HWMonitor, which has been on overnight while I slept and hasn't been reset yet (with a game on as well, for further stability testing) this is what I'm seeing for voltages and temperatures: https://i.imgur.com/Xo2eeDX.png

I think the core voltage is reaching around in the 1.080V - 1.228V neighbourhood~ according to CPU-Z when I watch it, but it fluctuates between these values dynamically while I operate the computer in various tasks.