Originally Posted by
rge
Intel has a mathematical formula (posted several times in Real temp thread) that describes the relationship of Tcase max (temp at center of cpu casing/IHS) and Tcore max (temp hottest core) at full TDP load using stock cooler and stock settings, proper testing ambients, etc. But basically at full TDP load, given hottest, ie worst case i920, Tcore should be 32.1C hotter (100C tjmax-67.9Tcase max) than TCase. There is about 5C gradient between Tcore and Tcase at idle, and 20C gradient at moderate loads and up to 32C gradient at max TDP loads, and even higher gradient between Tcore and Tcase when OCing at new full TDP. So if Tcase sensor was calibrated correctly, and located in casing of cpu it would read approximately 5C cooler at idle than core, 20C cooler at mod load, and 32C on max load, assuming stock settings, cooler, certain testing conditions, and worst case/hottest cpu for leakage at loads.
Tjunction max (temp of hottest DTS sensor in core) is by far the most accurate temp measure, and it is calibrated by intel on per part basis and measures hottest temp in core by scanning and recording multiple sensors in all potential hot spots...which is only accurate way to measure.
"Tcase" sensor is not even located in correct position (too expensive to place a sensor in cpu casing for mass production, not to mention it would still not be accurate as misses hot spots that multiple core sensors catch). This "tcase" sensor is located between the cores, where it has no chance of accurately measuring tcase or the hot spots. Nevertheless it is calibrated (via generalized average inaccurate calibration, not an accurate per part calibration like tcore directly from intel) and is calibrated much lower than actual temp of location it is at...for a poor attempt at approximating the temperature at casing (where it is not located). Most bioses to accomplish this (follow intel recommendations) and guesstimate a calibration for moderate loads for stock cooler, often being inaccurate and under best cases, in the ball park.
And look at what happens when one calibrates Tcase for moderate load, when the sensor is in the wrong position. Tcase sensor if accurately measuring its actual location (still in die substrate btwn cores), at moderate load if core measures 70C, Tcase should read ~65C (for duo, who knows with i7) for its true location between cores. But actual Tcase (casing cpu) would be say 50-55C....so bios simply calibrates/subtracts ~10C from Tcase sensor, so it accurately approximates Tcasing at mod load instead of its actual location between cores. But at idle where the gradient is less, the Tcase will now read too low, perhaps even ambient or subambient, both of which are laughable and commonly seen. And this ambient or subambient reading is not an error (though it is wrong), it is the only way to make it semi-accurate under moderate/heavy loads. Not to mention when you put a different cooler or overclock, then this inaccurate bios calibration is made even worse.
Bottom line like unclewebb said, ignore Tcase, that is what intel does. Or if using Tcase, realize it should read about 15-20C lower than tcore at moderate/heavy loads if properly basterdized to guess a location where it is not located, and that it will then be completely inaccurate at idle and at OCed loads, as it can only be accurate at one exact load setting.
Reply With Quote
