Jump to 100+ degrees Celsius on 100% load

[WhatsAHardware]

Hardware specifics in the bottom for convenience (I hope).
I recently installed a new motherboard, CPU, RAM, and PSU in my PC, keeping the CPU cooler, case, and graphics card. A few seconds after being turned on, however, the PC would shut down. After two tries, I dismounted the cooler and reattached it more thoroughly, and after another one or two cases of shutting down soon after being turned on, the PC ran smoothly.
Running Prime95, however, I found that running 100% load on the cores brings the internal temperature to 100+ degrees Celsius, and I quickly ended the tests, retried to confirm, and again had to end the tests immediately. I checked with Core Temp 1.3 and SpeedFan 4.52, and both agreed on temperatures in the 95-105 range. That can't be right!
In idle, I get temperatures around 25 degrees, which is slightly lower than I had before (as I recall). Running World of Warcraft, I reached only a fair range of 35-40 degrees.
My first guess would be that the cooler is still not properly mounted as I did have some trouble removing the back-plate from the old motherboard (the double-sticking material was pretty stubborn) and screwing the cooler itself into the back-plate through the new motherboard (without the tape it was a bit hard to get the position right). Does that sound right? What else could be wrong?
Bonus question: shouldn't the CPU shut down when reaching 95/100 degrees to avoid burning itself out?
Much thanks and many appreciation!!

• 
  Cooler: ZALMAN CNPS9900MAX-B
  CPU: Intel Core i7-6700K
  Motherboard: GIGABYTE GA-Z170X-Gaming 7

 

[Decends]

Which version of Prime 95 are you using? There's a version that uses a instruction set inside Intel processors that make them shoot to unrealistically high temperatures.

 

[techgeek]

The instruction set Decends was referring to is AVX (1/2) and they do indeed cause your CPU to run very hot. It is recommended for stress testing that you don't use any version of Prime 95 newer than 26.6, this was the last version of Prime 95 to not have AVX (Advanced Vectoring Extensions). A quick Gizoogling will find you a link for a download.

As to your issues, I am quite certain that you must be having some sort of issue with the retention mechanism on your cooler. I just verified that it does indeed support the 115(X) socket. You mentioned the double sided sticky foam tape, was this completely removed from the backplate as a result of being removed from your old motherboard. This compresses, but it's likely still stands the backplate off of the motherboard by about 1mm to 1.5mm compressed. This may not sound like a lot, but it could make a big difference to the amount of force that is being exerted on the IHS of your CPU.

Just for future reference and others that may use this, if you use a hair dryer on a moderate heat setting, and heat this tape, it will soften the adhesive and it makes peeling it away easier. Another option is to not remove the paper from the tape when you install it. This can make attaching the cooler trickier, but it completely avoids these types of issues when you need to replace motherboards or transfer the cooler to a new system.

As for shutting down, the CPU will throttle the clock to attempt to keep it below the 105°C threshold. If it's able to keep it under that temp, it will run at that temp. If it shoots above that, it will most definitely shut off.

 

[techgeek]

I just had a look at an installation video of this cooler and a look at the installation manual. It looks like its important to use the correct screws for the platform you are on. I don't know what you came from, but on this system you need to be using the Silver Bolts B (3mm). Looking at the instructions it is possible if you were using the wrong screws that you could indeed end up with less pressure between the IHS and the heatsink.

 

[WhatsAHardware]

Thanks folks!

I used Prime95 version 28.9 (p95v289.win64.zip) so that could be (part of) the problem.

Regarding the foam tape and screws: I did not consider that the width of the tape could have an impact, and I just grabbed the screws I had used for the previous motherboard. Reading through the manual now, it says to "insert four Silver Bolts B (for socket LGA 1156/1155/1150/1151/775) or Gold Bolts (for LGA 1366) through the Intel Clip holes", and since my old CPU was LGA 1366, I did probably use the "Gold Bolts" for the new CPU when I should have used the "Silver Bolts". Funny thing is, though, that as I installed it, the screws pulled the "wings" of the cooler down towards the board by quite a bit. I actually loosened them slightly as the bending looked excessive. One of the screws, however, gave major resistance before it was screwed in as far as the others, and I had to leave it slightly further out.

I'll try grabbing an older version of Prime95 when I get off work and check out the results. Thanks for the help so far!!

 

[techgeek]

Do you still have the hardware for the 115(X) platform? If so, I would use these.

As for the one hard screw, it's pretty important that you have even pressure on the IHS. Otherwise you may have less area making good contact. You especially want good contact over the center of the IHS. The die lies directly below the center of the IHS.

 

[techgeek]

Also from the look of it, it's possible to install the Intel clips the wrong way around. The bend in the clips should be so that the ends of the clips are closer to the motherboard, not raised away from it.

Having the screws tightened properly is important. If the pressure is uneven, the heatsink isn't likely to have proper contact with the IHS.

 

[WhatsAHardware]

Running Prime95 26.6, the temperatures increased to 90-100 within a minute, and I shut it off. The temps did seem to have stabilized there, though.

"Also from the look of it, it's possible to install the Intel clips the wrong way around. The bend in the clips should be so that the ends of the clips are closer to the motherboard, not raised away from it."
Yep, I noticed this when reading the manual yesterday as well, and I seem to have oriented it correctly.

I'll try replacing the screws with the correct ones for the 1151 now if I can do that without taking off the cooler entirely.

 

[WhatsAHardware]

After replacing the screws, the temps go to 85-95 which is of course (a little bit) better but still way higher than I would expect and have liked. The difference between the gold and silver screws seems to be that the actual "screwable" portion of the silver ones (which I now correctly installed) is shorter than on the gold ones, meaning there is a hard limit to how tightly the cooler can be fastened. Maybe it's because I'm missing the tape, but I actually managed to screw all four of them as far in as they'd go. I guess that means the pressure is now even (which is great) but also that it may be insufficient (which is not so great). Any ideas?

I'm tempted to switch to the Corsair H100i I ordered when I initially thought the cooler didn't fit the new socket, but then I'd have to buy a new case as well (since I later found out it's too big to fit my current Fractal Design Define R5), and I'd honestly rather get money back from the water cooler than spend more money on a compatible case

 

[techgeek]

Would you take a look at Vcore while running Prime 95 (26.6) and tell me what it is. I've built a few Skylake systems and I've found that the motherboard on AUTO Vcore was setting the voltage north of 1.3V which is quite high. I was able to lower them to below 1.2V which drastically dropped the temperatures.

 

[WhatsAHardware]

Which program should I use to check the Vcore? Core Temp 1.3 fluctuates in the 0.8-1.3 v interval on idle and stays around 1.3 v on 100% load (assuming the VID field is the Vcore). Speed Fan doesn't list any voltages (nor does it list any fan speeds) so it probably has trouble reading the value...?

 

[techgeek]

That is consistent with what I have seen. The fluctuating is normal, it's part of Intel SpeedStep and it will change the voltage quickly to respond to small changes in CPU load. Even "idle" isn't totally "idle". Most software has a polling interval of 1 second or more, so quick things that the OS is doing doesn't show under CPU usage. However since the CPU doesn't adjust Vcore quite as quickly, it is common for you to see the voltage fluctuation.

Core Temp is fine for monitoring voltage. CPU-Z has given erroneous readings in the past, so I tend to steer clear of it. HWInfo is good, but maybe a bit overkill for this process. HWMonitor is also good and less complicated than HWInfo.

As I said what you are seeing is pretty consistent with my findings of Skylake. That Vcore is much too high in my opinion for a CPU at stock. I would consider using a negative offset in the BIOS to tame that down a bit. Well I said offset, but Adaptive mode is probably the best if your motherboard supports it. If you need specific instructions on how to do this, please ask and I"ll download your motherboard manual to give you the exact settings and process. If you go this route start with an offset voltage of .05V. This should lower your load voltage to 1.25V. Test this for stability. If it's stable, then you can try lowering it some more, maybe in .01V increments until you find where it becomes unstable. Once you've determined your last stable voltage, set it up so that its .01V higher than that. So lets say you find that 1.20V is your last stable voltage (meaning you are using an offset value of .1V), then you would set it for 1.21V (which means you would be using an offset voltage of .09V).

Assuming 1.25V is stable, you should see a fairly significant improvement in your temperatures. Ultimately when you've dialled in your final stable voltage you should definitely be a lot cooler than what you have been.

 

[WhatsAHardware]

The BIOS was set to auto voltage, and I changed it to 1.25 v. Funny thing is, after doing that, Core Temp reported VID 1.31-1.33 on load. However, the temperatures were stuck in the 60-70 interval, and the system seemed stable, so that's of course a huge improvement. I'll try lovering it to 1.24 v now.

Edit: 1.24 v was stable, yielding temps in the same interval, although generally 2-3 degrees higher. Core Temp reported VIP of 1.32-1.35 on load. It would seem the value I'm setting in the BIOS affects the voltage inversely... I'll set it to 1.23 v and confirm.

Edit 2: I got similar results for temp and VID with 1.23, 1.22, and 1.21 v according to Core Temp. With 1.20 v, I got similar results, but then the PC bluescreened. I put the voltage back to 1.21 v which seemed and still seems to be stable.

I'm not sure how to interpret the above, but I reckon I can leave the voltage at 1.21 (or should I increase it to 1.22 to be sure?) and enjoy my 60-70 degrees. Thoughts?

EDIT 3: I just got a bluescreen with 1.21 v so I've increased it to 1.22.

 

[techgeek]

So are you using a Static voltage or Offset / Adaptive Modes to set you voltage. It sounds like Static Mode.

There must have been a corresponding voltage drop if you have a sudden drop in temps? Did you use the same software to view Vcore each time? CoreTemp?

If you are using Static Mode to control Vcore, if you are certain that you are stable, you might want to use Adaptive (preferable if your motherboard has this option) or Offset to achieve the same thing. The downside to Static Mode is that Intel SpeedStep won't be able to lower Vcore while idle, so while idle you'll still be at the same 1.22V that you are at load. It's not a deal killer, but you'll have slightly more heat and power consumption while idle.

 

[WhatsAHardware]

Yes, I used Core Temp to read the temperature and voltage every time.

I tried setting an offset of -0.08 v instead of the the static 1.22 v which resulted in a boot error. Changing the offset to -0.01 v, it seemed to work. When I get back home, I'll see if the PC is still running (I left it on when I went to work) and put it through a stress test. I expect I might be able to decrease the offset further, but it would seem that -0.08, although theoretically equivalent to a static value of 1.22 (based on the initial value of 1.3), is too far.

 

[techgeek]

That is odd. You may need to look in CoreTemp to determine how that offset affects Vcore. Are you sure it's not 0.1 and not .01? Either way what you are most interested in is that Vcore is the same when loaded as your stable static voltage.

 

[WhatsAHardware]

Are you sure it's not 0.1 and not .01?

What is, exactly?

Edit: When I came home, the temps on idle with offset -0.01 v were 25-30 degrees Celsius with VID 0.8 v, and on load the temps are 85-95 with VID climbing to 1.30-1.32 v.

 

[techgeek]

Well OK, that is quite literally a terrible motherboard manual. All it has is CPU Core Voltage Control in the manual and doesn't show any of the settings beneath that setting. I am an ASUS user and their manuals have every setting, what each setting is and whether or not changing a setting makes more settings available. Each setting is usually explained as well.

So I can't see what options you have under the above setting, so it'll be difficult to direct you. There is something obviously not right as your Vcore seems too high, and your temps are almost back up to when you had Auto set. While I guess Static is better than that, I'm sure I could help you tune Adaptive in if I knew what settings were listed under CPU Core Voltage Control. If you'd like more assistance with this, I'd need screenshots or pictures of this configuration screen. If any of the settings when enabled reveal additional settings, I'd need to know those as well. It's common for settings to be hidden until a setting is enabled. Such as if you switch to Adaptive, then it exposes a setting for Offset Sign (+/-) and Offset Value. Well that's an example, it may be different in your BIOS.

 

[WhatsAHardware]

Sorry about not getting back to this sooner. My -0.01 v setting caused another boot failure the other day, and I decided to just go back to default settings and accept the high temps on max load since that seems to at least be stable (as opposed to dynamic voltage with a negative offset) and not waste wear and power during normal usage (as opposed to static voltage). Thanks a lot for the help and clarifications!!
On the subject of manual quality and manufacturer, I'm going back to Asus the next time. Another caveat of this motherboard I was not aware of when I bought it was it coming with a Killer network solution which entails a mandatory manager program and, according to the internet, some questionable performance. Fortunately, the Intel driver worked as an alternative after installing the other drivers and fully updating Windows, but in retrospect I would sure have preferred Intel hardware as well.

 

[Suztera]

Unfortunately, Gigabyte tied adaptive mode with Auto so there is no box to select adaptive mode. Their UEFI layout need improvements.
If you have temperature problems with a manual voltage of 1.25 or lower at stock speeds, then either you have not installed the cooler correctly, not applied the thermal paste correctly, or the cpu cooler isn't good enough.

Running prime 28.9 small FFT tests should be fine, I get the same temps as prim 26.6
At around 1.25v with prime 28.9, you could expect 60C depending what cpu cooler you have. I not sure how well your cpu cooler fares.

Also, VID is VID and vcore is vcore. Ignore VID as that is what the processor request and vcore is what the actual voltage is given. And ambient temperatures matters to an extent.

 

[AppleGwava]

Maybe we are overthinking this,did you remove the plastic film on the contact surfaces?

 

[techgeek]

Maybe we are overthinking this,did you remove the plastic film on the contact surfaces?

The cooler was used on his previous system (1366) without this problem.