My original theory when I first started out was that 45nm and 65nm sensors were very similar with similar amounts of error. Intel's release of TjMax for 45nm showed me that I completely underestimated the amount of error in the 45nm sensors.
I couldn't imagine that Intel would be using a sensor that could be off by as much as +/- 10C at 50C but that is their official spec for the 45nm sensors used on the Intel Atom CPUs. Based on the data I've seen and user feedback, that amount of error seems to be what the 45nm Desktop E8000 and Q9000 sensors also have. Some sort of calibration is not an option for 45nm.
With the correct TjMax for 65nm, users should have very accurate temperatures from idle to TjMax. The 65nm sensors are far more accurate and rarely if ever suffer from saturation or sticking in the temperature range that most users operate at. The amount of slope error in these sensors is probably one quarter of the amount of error in the 45nm sensors. I think the latest beta of RealTemp is using the correct TjMax for most 65nm processors but it will be good to finally get this information from the manufacturer so all temp programs can finally agree at least to one part of the equation.
There are still a few minor things to do to RealTemp but it works OK so I haven't been too motivated lately. I've been having too much fun playing with my new Reapers! Not too bad for the old P5B.
Reply With Quote
Bookmarks