The highest representable temperature is the processor's Maximum Junction Temperature (Tjmax). You can determine your processor's Tjmax by reading the IA32 Temperature Target Model-Specific Register (MSR). The Tjmax temperature (in degrees Celsius) is found in bits 16-23 of the value returned.
If the temperature of the processor goes above Tjmax, the reading will continue to be Tjmax. That is, if your processor's Tjmax is, for example, 100c and the processor's current temperature is actually 105c, then the readings returned will still be 100c. At any temperature above Tjmax, the performance of the processor will be throttled to protect the processor from thermal overrun. If the temperature should continue to rise, however, it will eventually hit the ThermTrip threshold and the processor will be powered off. Intel does not publish what the ThermTrip threshold actually is, however, so I cannot give you a number.
In Desktop and Server processors, there is an additional threshold, called the Control Temperature (Tcontrol), that is also provided in the IA32 Temperature Target MSR data (in bits 8-15, as an offset below Tjmax). This is the temperature level that, if maintained (kept at or below), guarantees that your processor will never suffer any thermals-related silicon degradation. By definition, in the absence of full knowledge of your processor's Thermal Load Line, your fan speed control programming is required to ensure that the processor cooling fan (or blower or pump) is running at 100% duty cycle (i.e. full speed) if the temperature is above the Tcontrol threshold. If your processor's Tcontrol was, say, 85c, then you could run the processor at 85c constantly, for its entire warranted lifetime, without suffering any thermals-related silicon degradation.
In most Mobile and Embedded processors, no Tcontrol value is provided (i.e. the offset in the MSR is set to zero). This is unfortunate. It means that the only value you have to work with is Tjmax. I suggest that you use a non-zero Tcontrol value. In the 4th generation WY NUCs, which also used an Embedded processor (Core i5-4250u), we did a lot of testing and eventually concluded that we would use a Tcontrol offset of 17 in our default fan speed control programming. That is, if your processor's Tjmax was 100c, we used a Tcontrol of 83c. You don't have to be this conservative, but I don't recommend that you use zero (i.e. allow the temperature to rise to Tjmax all the time).
Hope this helps,
Hi, I have the same processor and as I've already mentioned here my CPU downclocks to base frequency 2.49 GHz when it reaches 80°C while gaming. I wonder if your processor does the same? You could test it with Intel XTU and we could compare the results.
Otherwise the temperature on my i5-7200U is around 45°C when idle.
What a competent, well formulated and exact answer....a true pleasure :-)
Then we only need Intel to specify this Tjmax and Tcontrol just as clearly on the ark.intel.com spec. page.
How hard can it be?
It's a nightmare to find it in the datasheet for the CPU generations datasheet, if it can be found at all.
I will add a few additional comments,
- In Desktop and Server processors, the Tjmax and Tcontrol values are specific to the individual processor. That is, they vary from one individual processor to another. During the validation of the silicon, the needed values for Tjmax and Tcontrol are determined and fuses are blown in the silicon to lock in the values necessary for that silicon. Bottom line, you cannot list Tjmax or Tcontrol values in the datasheet or on the ARK spec. pages as the value could vary from one individual processor to another.
- In Mobile and Embedded processors, they used to bin the processors so that the Tjmax was always a specific value - and indeed they listed this value (typically 100c) in the datasheets. I do not know if this is still the practice (I have been retired and out of the loop for a couple of years). If it is, then yes, I suppose they could list this value in the datasheet or on ARK. If it is now being handled like Desktop and Server processors, however, they would not be able to list Tjmax values in the datasheet or on the ARK spec. pages as the values could vary from one individual processor to another.
- In the past, datasheets were a lot more specific than they are today. Listing the Tjmax value in the datasheet was thus something easily done (though digging them out was, I agree, a pain in the ****). With how the datasheets are currently being coalesced together, this would be much more difficult. I suppose you could list them in the Spec. Update documents (via a table) or individually on the ARK spec. pages.
Bottom line, not so cut and dried.
Hope this helps,
Ahh...that make sense! I understand now why they are not specified.
Tried to lookup those tjmax values in Linux but was not successful. My IPMI/BMC do indicate the High-None-Critical to 95C and High-None-Recoverable to 100C.
SuperMicro claims that they do read those numbers from the CPU (Target Model-Specific Register, whatever that is :-P)
I guess I just have to trust them.