The latest beta includes a Start Minimized fix for Vista, with or without TaskBar enabled, as well as an adjustment for the B3 QX processors back to TjMax=85C.

http://www.fileden.com/files/2008/3/...alTempBeta.zip

hardtarget: Welcome to the 45nm crappy sensor club.
The Intel budget seems to have gone into processors that overclock like crazy so it's hard to be too upset with them.

Unfortunately, these sensors were not designed to report accurate temperatures. 10C of error in a 45nm sensor at idle is not unusual. If you end up with a Dual Core CPU that has a sensor on one die that reads too high and the other die reads too low then the combined error between core0 and core1 can be huge. A 9C difference between two cores isn't too bad and can be improved with a RealTemp calibration.

Start by reading the Calibration section in the RealTemp docs and try to do that. Since version 2.70 was released, Intel has told us that the official TjMax for your processor is 100C so uncalibrated, your reported idle temperatures are going to be even more unbelievable. You can either manually set TjMax=100C or update RealTemp to the latest beta version from the link above.

In version 2.75, the calibration formula was adjusted to try and compensate for 45nm sensors that are way off when using Intel's official TjMax value.

Try running Prime95 small FFTs and see if the temperature difference decreases at full load. In theory by about 90C the difference should be close to zero but most people never get their chips that hot. To get closer than 9C at lower temperatures you need to try and do the calibration.

You can also send me a log file of 1 minute idle time, 2 minutes of small FFTs and then 1 minute of idle time. Set the log interval to 1 second so there is lots of data. This will give me a better look at how your sensors respond.

Captn:

1) I might have to add some code to RealTemp so when users swap processors, their previous TjMax values get dumped.

I think most users look at the bios idle temperature and believe that it is 100% accurate. It's not. As I said, my bios temp is out by about 7C at idle. If this value is not accurate without some calibration then we shouldn't be making any comparisons to it. I think the temp guide by Computronix over at Tom's Hardware has a method to try and calibrate that if you're interested.

If your core temps are low and getting stuck then you might have to rely on the data coming from the CPU sensor instead.

I always avoid the Auto setting for voltages. My board is terrible on AUTO.

A good rule of thumb is to use enough core voltage so that your CPU is stable at whatever MHz you want to run at. If 3600 MHz is your goal, set it to 1.200 volts. If it's Prime stable, you can try a little less. If it's not Prime stable, then try some more. I tend to use very conservative memory settings when first testing / overclocking so it's not my memory holding back my CPU overclock. Once your CPU is stable then you can go back and work on optimizing your memory timings and speeds.