CPU seen at 91ºC once. Thermal paste first times? SpeedFan (or MB-CPU sensors)?

I fear, for either complex or uncommon reasons, that this could happen again and with worse consequences. Basically, I have had one session at abnormally high CPU temp readings (I don’t know yet if they were real), I did 2 fixes before starting next session (fix SpeedFan startup with Windows and, kind of inadvertently, slide the heatsink over the CPU, minimally but noticeably), and CPUtemps/readings are normal now. The best possible news would be that it was SpeedFan, but if it was something about the paste or the heatsink settlement, possibly I have to do something about this (like laying the case down on its right side when shutting down until the paste “cures”, or leaving it on for several days, I’m anything but expert in this really).

Relevant specs:

  • processor: AMD FX8350 4GHz (Vishera, 125W TDP)
  • motherboard: Gigabyte GA-970A-DS3P (chipset AMD 970 + SB950)
  • case: CoolerMaster Centurion 5 (midi tower, 43.5 cm high x 48 cm long x 20 cm wide, front made of mesh, side racks for processor and cards, 120 mm exhaust fan)
  • CPU cooler: Tacens Gelus Extreme (w/o ordinal; horizontal, full copper, wide fins, somewhat heavy, 3 heatpipes w/o direct contact) with upgraded fan
  • CPU fan: Enermax Magma UCMA12 (120 mm, 1600 rpm, 69 cfm)
  • case fan: Scythe Slipstream 1900 rpm (120 mm, 1900 rpm, 110 cfm)

Rest:

  • 16 GB RAM in 2 modules
  • GeForce GTX 750Ti
  • EVGA 500W PSU
  • 2 DVD drives
  • SDD and HDD
  • Windows 10 pro 64bits

Normal situation at idle:

SpeedFan and the BIOS agree in this (room temp is 16ºC):

GPU: 25ºC
Chipset: 25ºC
Processor: 26ºC
SSD: 39ºC
HDD: 30ºC

CPU fan: 1680 rpm (not regulated)
Case fan: 1070 rpm (regulated by the motherboard)

Problem: processor seen at up to 91ºC with SpeedFan during a Windows update load.

Today I’ve switched on the computer after 48 hours off (first time this long) and I’ve seen the processor at up to 91ºC (SpeedFan reporting, rest as above) while loaded at 80-100% (w/o signs of throttling) likely for the calculations of a Windows update. I fixed a problem with SpeedFan startup with Windows, I rebooted to the BIOS and it reported 75ºC. After shutting down and switching on again, values returned to normal.

When seeing those temps, I opened the case to see the fans and touch the CPU heatsink, that was only slightly warm. With the rig off and unplugged, I did more concious checks and was able to slide a bit the CPU heatsink.

Afterwards I tried compressing a big folder (3 GB) with WinRAR and CPU temp did peaks at 65ºC only (the rest of temps did rise 1ºC if at all, what kept my exhaust fan at those 1100 rpm although it can do 1900, but I think that gaming should rise chipset temp and case fan speeds; I could also place a Slipstream 1600 instead and leave it w/o regulation).

Suspects:

  • SpeedFan: I use it only to see temperatures, rpm etc. These first days of the build I’m configuring SpeedFan startup with Windows Task Scheduler and w/o doing additional reboots. This time it started (unlike in my previous attempt 2 days ago) but complained about some drivers not being loaded and didn’t load. I know the fix for this is setting a delay (task → triggers → edit trigger → advanced → delay for 30 seconds). Now it’s already working well, but it wasn’t in the problematic startup some hours ago. I added the setting to the task and launched it successfully (to test the task and to launch SpeedFan).

¿Could a badly launched SpeedFan misconfigure the CPU temp reading until next shut down?

  • Thermal paste first times / Heatsink attachment: might the thermal paste have turned too solid or whatever for conducting well, during the 48 hours the rig was off before the problematic session? This flat doesn’t have heating and temps in the night in this room drop to 10-12ºC.

If it matters, the paste is Arctic MX4 bought in 2020 and never used before. I cleaned both surfaces with cotton and pharmacy ethanol, and I applied a small quantity in the center of the processor.

While inspecting the heatsink attachment with the computer off between the two sessions this morning, I was able to slide the heatsink over the processor a bit. This also happened while unmounting the cooler from its previous attachment, a Brisbane (ancient AMD dual core with 65W TDP, for which this cooler was overkill and never was hot; btw everything was different in this 2007 build: I used the stock paste, the flat had heating (in winter) and it was summer, and I spreaded a layer of paste instead of putting a bit in the center, what has made me work to clean the excess some days ago after the unmount) that I used just before the unmount (a common advice for softening the paste and easing the heatsink removal).

So the heatsink attachment has clearance and its contact surface is totally even, like processor’s. If the paste is fluid the heatsink can slide a bit. Is it critical? Might the paste flow on its own? Does it depend on if the paste has “cured” yet?

Amount of pressure? Moderate, not excessive. I have done mounts applying more force, but I wouldn’t say “weak”.

I have been using the same thermal paste (Arctic Silver 5) on my computers since my original tube, which I bought in March 2006, and it still works good. I last used it in 2020 when I upgraded my primary PC’s dual-core i3-2120 CPU to my current quad-core i5-3550 CPU (which appears to be 2012 technology like your CPU) for only $20 on Ebay. I use my heatsink/fan combo from the i3-2120 on the used i5-3550 CPU and then under-volted the CPU by -0.130v (through my ASUS motherboards BIOS settings) which dropped CPU temps by about 13c under full load, which is the most I can lower the voltage to the CPU and keep a stable system as if I drop down to -0.140v, while it will boot up and appear normal for a bit, it will usually hard-lock within about 24 hours or so tops. I tried dropping to I think it was -0.150v (or maybe it was -0.160v) and it starts having trouble booting etc.

but basically with under-volting… it’s a nice way to lower peak CPU temp with no real penalty as the worst that can happen is your system will freeze/become unstable, but won’t damage anything. as for how much you can undervolt and still keep a stable system, it’s hard to say until you play with it.

you can use Prime95’s torture test (i.e. GIMPS - Free Prime95 software downloads - PrimeNet ; at the moment you probably want “p95v307b9.win64.zip” (I am using Linux Mint v20.3-Xfce so I use the 64bit Linux version myself)) if your really trying to see what temps you will get under a high load. on my i5-3550 CPU if you disable ‘AVX’ in Prime95 that’s more of a real world worst-case-scenario for CPU temps (since programs generally don’t use AVX instructions). although with AVX instructions on, the CPU temp will go even higher. but, as a ball park, at least on my CPU you can usually run the test for about 20 minutes as by then CPU temp tends to level off etc.

p.s. I pretty much applied my thermal paste like you did…cleaned both the CPU and heatsink surface with Isopropyl Alcohol 91% and then applied a small amount (they say ‘pea sized’, but probably no more than BB size tops) in center of CPU and then sit heatsink on top and clipped it into place with say the top left pin and bottom right, which really stabilizes it and then clip into place the other two so it’s nice and secure and does not slide around.

I think it’s something buggy.

This has happened again. The max temp I saw this time was 89ºC (very close to first occurrence’s 91ºC) and the same 75ºC after rebooting to the BIOS (although the room was 6ºC warmer, 22ºC instead of 16ºC) and I think the cause of this is software.

It could still be a firmware bug. I have even considered ºF but the BIOS doesn’t have such option, 90ºF (32ºC) is low for the processor if loaded (in normal situation it goes quickly to 40s), and 75ºF (24ºC) idle (in the BIOS) was possible in the first occurrence with the room at 16ºC but impossible at 22ºC.

This time the firewall detected ransomware, so it could be for malicious software, at least this second occurrence, but I think some bug is involved: neither SpeedFan (in the first occurrence), nor the firewall nor the virus (if any) “try” to simulate that my processor is that hot, not even in error, but their actions call some buggy routine, whose effect is permanent until disconnecting power (either physically or just a shut down, I haven’t tested this).

This time I didn’t move any hardware, I just shut down the computer, switched off the power strip, waited like one minute and switched on the computer again.

I’m aiming to install 3dMark Vantage as stress test, possibly 3dMark06 too (the vcard is a 750Ti), it’s synthetic yet but not as much as Prime95. The creators of Furmark (am I recalling this well?) demonstrated that any hardware could be tortured much more than in any productive situation, but I prefer a hard productive situation as test. I’ve already got to install RivaTuner v2.24 to control brightness gamma etc with hotkeys.

Did you ever consider updating your motherboards BIOS?

who knows, maybe a faulty sensor?

Yeah, that’s why I suggested turning off AVX since that’s more of a real world worst case scenario for temps, although even this, might still be a little higher than just about everything else you would throw at it. but the way I see it, if your system is stable with Prme95’s torture test (I just use the test that hammers the CPU and not much in the RAM), then you can pretty much guarantee any normal use won’t have problems.

I did use Furmark in the past as it’s a GPU stress testing program. but yeah, it stresses the GPU more than any game would. but, like I mentioned with Prime95, if your system can handle that without issue then you can be confident it will be stable under any gaming situation.

If your not even sure if your general Windows install is in good running order. I always tend to opt for the nuke option… wipe the hard drive and install Windows from scratch as then you can be sure all of your software is in good running order.