Quote:
|
Originariamente inviato da unclewebb
The temperature sensors do not respond linearly to changes in core temperature. The Core i7 sensors are very good and changes in actual core temperature seems to track changes in these sensors very closely but the 45nm Core 2 sensors can be horrible.
Intel calibrates these sensors somewhere around 100C and I find that core 0 is usually very accurate at that point. The further you get away from the calibration point, the more the amount of error increases. Even though TJMax is correct, the reported temperature at idle might read 5C or 10C too low or too high. Every CPU is unique so I gave users a way to correct for this error at idle by using some Idle Calibration adjustments.
If users saw errors in their reported temperature at idle, they used to increase or decrease TJMax but all this did was move the sensor error from idle up towards TJMax so now their load temperatures would be reported incorrectly. Idle calibration lets you correct the temperature curve of your sensors without having to change TJMax which will screw up your full load temperatures.
The data from these sensors might change 9C or 11C for every 10C change in core temperature. The +/- 10% accuracy number was a rough ballpark number from Intel's presentations last year at IDF. The amount of error might be worse than that. When you get 60C away from the calibration point at idle, it's easy to see that a +/- 10% error in these sensors can become significant.
Most users don't bother to calibrate their sensors because it's not an exact science. It was just nice that Intel finally admitted that these sensors are far from perfect with errors at the TJMax calibration point, errors in the slope of the temperature curve as well as sticking sensors at low temperatures.
|
in merito alla calibrazione dei sensori
