After reading Computronix's c2d/q temp guide, all I can say is THANKS. I been looking for info about this for the past 2 years since I got into overclocking. Anyway, I tried calibrating speedfan (albeit w/a rather crude thermometer until I can get a good one) and am getting some odd readings. The core temps (Thermal Junction temps) in speed fan which also match the core temps in realtemp have only about a 7c delta between idle and prime95 small ft load (both before and after calibration). While before and after callibrating speedfan for idle and load, the Tcase temp shows about a 20c delta from idle to load. (incidentally, DFI's smart guardian program identifies the Tcase temp as "cpu")
So the odd thing is that when loaded (after callibrating) the core (tj) temp shows 5c higher than the tcase temp but when idle the core (tj) temp is about 15-20c higher than the tcase value. I'm not sure what to do here, b/c I really want to try overclocking this sucker (have already gotten it prime large ft stable to 4ghz for 2hrs without bumping voltages) but I don't want to keep pushing unless I can have some assurance that my temp readings are remotely accurate.
I mean, should I just ignore the core temps and rely on the tcase temps? After callibrating they are running about 28c idle and 45c under load.
You've described "stuck" Tjunction sensors (DTS), which is all too common among 45nm processors. From the Guide:
Section 9: Calibrations
Part 2: Calibration - Tjunction Load (Uses maximum cooling at Stock Vcore, Frequency and Multiplier)
Note 2:Tjunction sensors were designed to be linear at high temperatures for Throttle and Shutdown protection, so Idle could indicate too low on 65 nanometer variants, or too high on 45 nanometer variants, many of which have faulty sensors that won't Idle below certain temperatures. Sensors can be tested using Real Temp 2.7 - http://www.techpowerup.com/realtemp/
If Tjunction Idle is not ~ 3 higher than Tcase, then use Tcase Idle for accuracy.
Section 15: Troubleshooting
Note: A significant percentage of 45 nanometer processors (E7000, E8000, Q9000 and QX9000 series) are being reported with faulty DTS sensors, where one or all the Cores won't decrease to low Idle temperatures. Offsets between Cores exceeding 10c are also being reported. Sensors can be tested using Real Temp 2.7 - http://www.techpowerup.com/realtemp/
No, this means that Tjunction Idle may be useless, while Tjunction Load may be OK. Most 45nm DTS sensors begin to respond properly above 50c, which is why your Tcase temp and Tjunction temps make sense at Load. In the instance that your DTS sensors won't Idle down properly without "sticking" at a certain value, then just ignore Tjunction Idle, and use Tcase Idle for accurate Idle temperatures.
SpeedFan's default Tjunction Max value for the E8400 is 100c, which is correct, according to Intel. However, each processor is unique, and Tjunction Max values should be regarded as "up to 100c", which means that the value could actually be somewhere between 95c to 100c. I've found the most common values to average 97c to 98c. This is why Tjunction Max values are round numbers, while Tcase Max values are specified to the exact tenth of a degree.