I have asked intel this exact question for 4th time/try now, and have yet to get an accurate or even pertinate answer, still waiting on answer for 4th try, hopefully will get escalated to get an accurate response.
But according to intel data sheets:
Core temps are die temps, there are no thermal specs for die temps.
Cpu temps are between the cores, but still in silicone die with a high thermal conductance of greater than 100W/M-K, and thus even at full load will have temps no less than ~5C from die temps.
http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf (see figure 5)
Tcase temps are at the geometric center of the IHS, and thus across TIM1 where the majority of the gradient from core to Tcase exists on full load, because TIM1 has a thermal conductance ~20x lower than die conductance.
Also according to intel specs, throttling is supposed to occurs at or near tjmax, however, that is providing a lot of parameters are strictly defined like intel heatsink, ambients, exact loading program, etc.
In other words suppose you were using intel heat sink, and an E8400 with tjmax of 95C, and used exact intel parameters for testing. When die temps were 95C, cpu temps may be 90C, Tcase may be 72.4C, and the cpu would throttle. So the cpu would be self protective right at the thermal margin.
While that gradient of ~20C between die temps and Tcase, when tested under exact intel parameters is documented by intels published and tested accurate formula...Tjmax-Tcase=J(jt)*TDP, this gradient fluxuates widely based on load, type of load ie floating point etc, ambients, volts, cooling, etc.
When intel is asked what temp to watch, they always reply cpu temps. The reason is easy to see, if cpu temps are 72.4C or less, you are clearly within specs, as Tcase has to be lower at full load as it is across TIM1 with a thermal conductance ~20x less than die and across tim1 is the major portion of the gradient.
What I am sure about is
1) Providing your cpu temp is accurate (bios support/calibration are correct), then if your cpu temp is within specs, for example less than 72.4 if E8400, then by definition your Tcase is within spec and you are ok.
(Note, the question is however, can you still be within specs if cpu temp is higher than 72.4 assuming Tcase is still less than 72.4. The gradient for E8400 is max 23C on full TDP load, given all exact intel parameters and cooling. Even if your cooling gives a 10C gradient, when cpu reads 72.4, that means Tcase 62-67. Worst case scenario of no gradient (impossible on load) and you are still just within spec. )
2) since max gradient by formula (using max specs) is 23C for E8400 from case to die, if you are >23 from Tjmax, using delta to Tjmax, (note your die temps at this point are 73C), you are clearly again within spec, especially given your die temps are at spec. In fact, on almost all cpu's if you are >25 from tjmax, by definition, even under worst case scenario ie minimal gradient, you are within specs.
However, what intel has not answered is when they say to use cpu temp, has consideration already been factored into the necessary gradient from tcase to cpu, hence cpu temps should never go over 72C, or is it that you can still be within spec if cpu temp is above 72C, because indeed Tcase is IHS temp strictly, without having fudged the gradient factor into it.
though granted, that is a less important question, given intels quote "for every 10C above thermal parameters it halves time between failures", its always probably prudent to be on the safe side and keep cpu temps below 72C.
Facebook
Twitter