I've been looking at this a lot, and I believe the reading is a false value. It's proportional to the temperature, but not calibrated against any real number.
The temp sensor was designed using information provided by Intel in a preliminary version of the datasheet. When I went back to verify the circuitry, I found they've totally removed any reference to the thermal sensor from the datasheet. They don't even admit that it has a thermal sensor anymore.
Even when it's reading a very high number, I can comfortably lay my finger on the chip. If that were a real reading, my finger should be burned.
As for differences between units, I suspect it has to do with whether you're plugging it into a 100 megabit, or gigabit switch. There's also been a push for power-saving versions of ethernet. If your switch supports this feature it can also affect the reading.
I've been looking at this a lot, and I believe the reading is a false value. It's proportional to the temperature, but not calibrated against any real number.
The temp sensor was designed using information provided by Intel in a preliminary version of the datasheet. When I went back to verify the circuitry, I found they've totally removed any reference to the thermal sensor from the datasheet. They don't even admit that it has a thermal sensor anymore.
Even when it's reading a very high number, I can comfortably lay my finger on the chip. If that were a real reading, my finger should be burned.
As for differences between units, I suspect it has to do with whether you're plugging it into a 100 megabit, or gigabit switch. There's also been a push for power-saving versions of ethernet. If your switch supports this feature it can also affect the reading.
