SPCR

Dear SPCR forumites,
I find myself in need of your wisdom once more.

CPU: Intel i7-6700K
Cooler: NOCTUA NH-D9L with two fans (front and mid)
MB: ASUS Z170I PRO GAMING
Open case (Ncase M1 with no panels)

The behavior I am seeing is CPU temps exceeding 90°C on load. Even stranger is the temperature dynamic: Basically from 40°C idle the temps go to 84°C within the 3 seconds resolution of the graph. I have seen jumps to 95+ and using different SW with 1 second resolution I can see jumps from 40 to 80 and back within 2 seconds. The air coming from the cooler is quite cool.


My first thought was I botched the thermal paste application but the imprint does seem OK to me: Not great but also not terrible. The results are the same after second paste application.

Any ideas what might be wrong? I really don't understand the extremely fast rise and drop in temperature. Thanks in advance.


Best regards,
Regis

Disable HT in BIOS and post the new results: same 85-95°C peaks?

Hi,
Assuming HT is Hyper-Threading then it does not seem to make much of a difference: Just to be entirely clear the temp peaks are not random neither they are only at the start of a CPU load. The moment the load goes up the temp goes up. The moment the load goes down the temp goes down. Basically the same second. The graph shows roughly 30s run of Prime95 and the high temp area directly coresponds to the times I started and stopped the Prime95 run... Delta is some 40°C.

Yes, HT was hyper threading, I wanted the CPU behaved like a more frugal Core i5. How much is the actual clock? How much low is the CPU fans pace? May you test the combo in an open air scenario?

ATM it looks like there might be an issue with the CPU/cooler interface (TIM, mounting mechanism, and so on), but nothing definitive, as it might even be that your NH-D9L can't handle your 6700K (did you checked the relevant ASUS bloatware? Usually it overvolts by default the cores, heating them up pointlessly).

Idle at 380 RPM@~45C.
Load at 2000RPM, 85-95C
4100 MHz
The case is open (no panels attached) but the GPU is close by. I just rested a 15cm fan against the GPU/CPU gap and it dropped the maximum temp to 79 so maybe I do have an airflow issue even in such a open enclosure. I'll try different cooler orientation to see if it makes a difference.

In the SPCR test the NH-D9L managed 33°C rise at 95W TDP so it should be just fine for 6700K unless I am missing something.

Obviously I suspected the CPU/cooler interface as well due to the very sharp temp rise/fall but the thermal compound imprint looks OK, doesn't it? Has anyone seen such sharp changes in CPU temp with any CPU/cooler? Especially the drop seems ridiculous to me... 40°C instantly.

As Luca pointed out, check for unintentional overvolting by the ASUS tools. Try running everything at stock speeds/voltages and see if the problem lessens. If it does, but still shows relatively high temps, try underclocking the base frequency by a fair bit (25% or so) and see what happens to temps.

The steep temp transition up and down plus the relatively small "good" cooler interface area surrounded by the fractal pattern for a "bad" interface lead me to believe that the cooler/CPU interface "as is" can't handle the wattage and the interface needs to be improved. Since you show a clean center area, it implies the cooler and or cpu could use lapping. While the D9L isn't the biggest/best cooler out there, it should be able to handle a stock i7 set up if careful. An ideal interface would show a steep rise to a stable temp with load and if the cooler was somewhat 'underpowered', there would be a slowly increasing thermal tail over time.*


*not to be confused with a slowly increasing thermal tail by ALL components -> that implies the case is restricting cooling.

Regis wrote:In the SPCR test the NH-D9L managed 33°C rise at 95W TDP so it should be just fine for 6700K unless I am missing something.

I guess you missed the 22°C ambient temp, the open air scenario, the absence of a large, discrete GPU and the different IHS TIM used by the Sandy Bridge generation.

However, that one



is an enough good TIM footprint: it looks like a bit different from your one.

Regis wrote:Obviously I suspected the CPU/cooler interface as well due to the very sharp temp rise/fall but the thermal compound imprint looks OK, doesn't it? Has anyone seen such sharp changes in CPU temp with any CPU/cooler? Especially the drop seems ridiculous to me... 40°C instantly.

That's how a modern CPU works, and that may be quickly exacerbated by a poor contact (it might even be a defective base, drop a mail to the Noctua customer support to have a more informed opinion).

Hi,

I experienced somewhat similar problems with my 4790K: everything was fine, the CPU was at max frequency and acceptable temperatures (for a mITX build), but 5-10-20 seconds into a prime run, the temperatures would simly go haywire, without any warning.

The source of the problem was that the AsRock motherboard defaults were actually quite heavy in terms of overclocking: the power targets were set to unlimited, instead of the official values. I downloaded the intel extreme tuning software, (or you can set it in the BIOS, if you know what you are searching for) and set all power targets to be the official 85W, and the problem was solved. Note that I used a similar Noctua cooler, one that came with two 9cm fans.

So yeah, +1 for checking ALL the relevant BIOS settings and where applicable, setting them not to ASUS default, but to Intel defaults.

Thank you all very much for your suggestions, I think I now understand the issues I am seeing.

After my initial removal of the heatsink the second imprint was much better in my opinion: I believe I may have overtightened the Noctua mounting system, bottoming out the spring loaded screws. I guess this might have caused the sub-optimal imprint. However even with this better imprint the cooler was not performing as I would have expected (see previous posts).

I have now rotated the cooler 90 degrees, drawing air from outside the case. This immediately dropped the idle CPU fan RPM from 380 to 280 and the peak temperatures are much better: So even though the GPU is pretty cool to the touch the few degrees still make a big difference. Obviously I should have realised that but people were recommending vertical alignment for better airflow...
You can still see the temp jump which really scared me is present, just a bit lower (80 max). I looks to me like the fan dynamic is slower than the CPU thermal dynamic. Once the fans kick in the temperatrue stabilises nicely at below 70.

The final piece of the puzzle which might explain why the temperature flies up so fast lies in the automatic overclocking: Thanks Nagi for pointing me in the right direction. The CPU cools just fine even at 120W (reported) but obviously it pumps out a lot of heat in very short time. It might go even higher at the very beginning when the temp is low.

So to sum up: high air inlet temperature, overtightened CPU cooler mouting and automatic overclock & overvolt seems to be the culprit.

Thanks again everyone. Now I need to figure out why the high voltage and low clock frequency and yet high watts...

Have you updated the BIOS of the motherboard yet? (And make sure you reset to defaults - sometimes a BIOS update can fubar the settings)

I actually just ordered the same motherboard, so I'm interesting in seeing if I run into the same issues you have. It should arrive sometime next week hopefully.

flinx wrote:Have you updated the BIOS of the motherboard yet? (And make sure you reset to defaults - sometimes a BIOS update can fubar the settings)

With some Asus boards the auto overclock features might be enabled, or be set to auto with the same results. Make sure to disable those features.

Since I now have my system all set up - I haven't experienced any of the issues here. My max CPU temp is 62C, and that's only during extended Prime95 load. I had to update the BIOS to get it to recognize the DDR4 2400 speed of my RAM, but there's no overclocking applied by default. Max speed for the 6700K is 4.2Ghz btw when using one core, 4Ghz when all four cores are loaded. Your core voltage looks fine, but if you like, you can try setting it manually to something lower.