[SOLVED] Core i7 8700 temperature spikes over 90°C at start of Prime95 test

[panz3r88]

Good morning I have a question related to CPU temperature testing.
I work on some wheel aligment machines that have a PC mounted inside and since one of them had a temperature issue(the CPU temperature was constantly over 90°C) we decide to implement some temperature test to detect this problem before selling the machines.
So I did some research and found here Computronix Intel temperature guide and I am trying to follow it.
After reading it I did some test on various PC in the following way: I took Prime 95 26.6 and started the small fft test, run it for 10mins and checked the temperatures with core temperatures. I also checked the ambient temperature(it was always around 19°C and 21°C).
In all the test I did I always checked that the temperature was always below 80°C. Actually I am always getting around 75°C on all cores in every situations so everything should be ok.
There is only one odd situation that occured in the test I did with one PC: the motherboard is an Asus Prime Z370, the CPU an i7 8700. What happens in this case is that in the first few seconds of the test the temperature goes up to 90°C and then returns to 70°C-75°C.
Why is this happening? It could be a sign of something not working properly on that PC?

 

[Petrossa]

The temperature most likely spikes and then drops as the CPU fan will ramp up its speed once the system detects that the CPU is running hotter. This seems to have the desired effect, and you get down to your 70-75.

 

[kanewolf]

You may want to make your fan curve more aggressive so that the fan speeds up BEFORE the temps gets to 90. The BIOS has options for changing the fan behavior.

 

[panz3r88]

The temperature most likely spikes and then drops as the CPU fan will ramp up its speed once the system detects that the CPU is running hotter. This seems to have the desired effect, and you get down to your 70-75.

I also think that the fact that I get down to 70°C-75°C should be ok, the odd thing is that if I take a different PC with the same hardware configuration this situation does not happen and when I start the test the temperatures goes from 20°C to 70°C in a few seconds without that spike. I also checked that the two motherboard have the same version of the BIOS to be sure that everything was really the same in both situation

 

[Petrossa]

That is quite an odd situation.
I'd do as Kanewolf suggests above and try to adjust my fan curve to speed the fan up before hitting 90C

 

[panz3r88]

That is quite an odd situation.
I'd do as Kanewolf suggests above and try to adjust my fan curve to speed the fan up before hitting 90C

Actually I took the two PC and I noticed one thing: I kept the task manager opened and looked at the performance tab.
For both PC I see that the speed is around 4.3Ghz-4.4GHz when there's no load. When I start the Prime95 test I see a difference in the behaviour of the two PC: in the one where temperature does not reach 90°C, as soon as I start the test the speed goes down to 3.3GHz, while for the other PC it remains 4.3GHz and then goes down when the temperature hits 90°C.

 

[Dragos Manea]

Are you doing this tests on stock intel cooler?

 

[panz3r88]

Are you doing this tests on stock intel cooler?

Yes

 

[Dragos Manea]

Intel stock cooler it is not made to support such loads, it is made for casual using (Gaming, browsing, watching a movie), for rendering video filles or constant max load you need an aftermarket cooler. This is not the best way to test if it is overheating because they will throttle, like you saw, some drop down to 3.3 some remain to 4-4.3. I repeat intel stock cooler it is not designed to run prime95 on it. Try to run blend mode, it is much easier on CPU and you can see if it overheating but i do not recomend to keep running prime95 or any other syntethic loader on stock cooler.

EDIT: Any pc i made for someone i always added an aftermarket cooler, intel stock cooler is a piece of crap. I have over 100 PCs build for peoples and i always made sure to add a cheap aftermarket cooler (30-50$), no more headaches.

 

[CompuTronix]

Dragos Manea,

Respectfully, the OP apparently knows that only Prime95 version 26.6 should be used, which is non-AVX, and is why he's not reaching Throttle temperature at 100°C on the stock cooler, given his ambient temperature. Further, while "Blend" is useful for testing memory stability, if you observe the thermal signatures on a graph over the course of an hour, you'll notice that the Blend test periodically transitions between memory intensive and CPU intensive segments, during which time the latter will produce Core temperatures equal to the Small FFT's test.

Due to the AVX issue, when mentioning Prime95, it's always important to specify which version is being utilized. As per Intel’s Datasheets, TDP and Thermal Specifications are validated “without AVX”.

See page 87, Section 5.1.1, Thermal Considerations, first paragraph.

8th and 9th Generation Intel® Core™ Processor Families Datasheet, Volume 1 - https://www.intel.com/content/www/us/en/products/docs/processors/core/8th-gen-core-family-datasheet-vol-1.html

panz3r88,

Assuming fan curves are currently identical on both computers, changing the fan curves on only one machine will introduce yet another unwanted variable which will complicate troubleshooting. For the present, ensuring that fan curves are identical on both computers is critical.

There's obviously a difference between the two platforms, most likely in BIOS, which is causing different clock speeds, and thus voltage and thermal behaviors. It's also possible there's a difference between processor Microcode (shown in BIOS). There's also an inherent difference in firmware unique to each individual processor such as VID (Voltage ID), which will cause differences in thermal behaviors as the processors react to Auto or Adaptive voltage settings relative to workload requirements.

In order to compare apples to apples, you must first verify that absolutely all firmware, BIOS and software settings and variables are identical. Otherwise, you won't be able to narrow the cause of the thermal spiking to any source(s).

While troubleshooting, there's a few items to keep in mind:

Core temperatures respond instantly to changes in load. Intel’s specification for Digital Thermal Sensor (DTS) response time is 256 milliseconds, or about 1/4th of a second. This happens much more rapidly than fan curves can respond or compensate.

Also, since Windows has dozens of Processes and Services running in the background, it’s normal to see rapid and random Core temperature “spikes” or fluctuations, especially during the first few minutes after startup. Any software activity will show some percentage of CPU Utilization in Windows Task Manager, where unnecessary Tray items, Startups, Processes and Services that contribute to excessive spiking can be disabled.

Moreover, 6th Generation processors introduced "Speed Shift" technology in Windows 10, which responds much faster to changes in workload than "SpeedStep" due to having many more Core speed and Core voltage transition levels.

Since 7th through 9th Generation Speed Shift is twice as fast as 6th Generation, some users complain of Core temperature spikes which can also cause fluctuations in fan RPM at idle. Motherboard manufacturers are implementing BIOS updates that include separate SpeedStep and Speed Shift settings with more flexible fan curves and time delay options.

In addition to SpeedStep (EIST) and Speed Shift, other BIOS settings that affect spiking are Auto or Adaptive Vcore and Race To Halt. Reducing Vcore by using manual settings can help to reduce spiking, however, differences in processor Voltage ID (VID) is unique, which is shown in the monitoring utility Core Temp. Higher VID means higher operational Core voltages, temperatures and spiking.

Regardless, providing the spikes don't approach throttle temperature at 100°C during normal operations, thus causing momentary software hesitations, spiking problems are inconsequential and can be ignored.

CT

 

[Dragos Manea]

panz3r88,

Assuming fan curves are identical on both computers, changing the fan curves on only one machine will introduce yet another unwanted variable which will complicate troubleshooting. For the present, ensuring that fan curves are identical on both computers is critical.

No 2 CPUs are identical, even if there is same microcodes and same fan curves (i updates my bios manually with all available microcodes because Asus omitted some and i had X99 OC bug in the latest windows update), you will still see differences between CPUs, only thing that should be considered is temperature to not approach throtling limit (usually under 80C core temp is decent, under 70C is great).