From the creator of RealTemp himself:
"We're debating on the XS forum about whether the new TjMax=100C number might have a couple of degree margin of error built into it by Intel. They mentioned in their presentation that there is a margin of error where they calibrate their CPUs but didn't bother mentioning how big that error might be. Maybe the official engineering TjMax is 97.5C +/- 2.5C. They might call that TjMax=100C and make sure that all of their processors are under the upper 100C limit.
None of this effects anything. The Distance to TjMax is still the same no matter what TjMax is truly correct. There can also be so much error in these sensors at lower values that there really is no 100% correct absolute temperature for these things. Software has to do some guessing because these sensors are not that accurate. At least Intel has finally published that fact. When I used to try to tell people this they'd never believe me but if Intel says it, then they have to believe it.
Here's a link to beta version 2.75 which includes Intel's new TjMax numbers.
http://www.fileden.com/files/2008/3/3/1794507/RealTempBeta.zip
I don't yet agree 100% but I'll be doing some more testing this week to see how this theory agrees with a few different processors. Because of the amount of error in these sensors, it can be very difficult to prove an exact TjMax. To avoid confusion, the default RealTemp TjMax needs to be exactly the same as what Intel has told us but based on real world testing, users might feel comfortable dropping that a couple of degrees or at least not accepting any temperature number as 100% accurate because they're not. The on chip digital thermal sensors were never designed for that purpose.
...
Distance to TjMax is the best thing to monitor and rely on but the problem has always been that users brains don't relate well to a sensor that decreases in value as their processor increases in temperature. It works backwards to how our brains process information. We all like to see hard temperature numbers like 60C.
Just going by Distance to TjMax was floated around for a long time but users have never accepted it. In my testing, most processors will start to thermal throttle when the Distance to TjMax reaches 2. This means the multiplier will cycle between its normal value and 6.0 which reduces the heat output and keeps the processors from getting any hotter if possible.
The calibration method would stay the same. I just need to create a new formula to better match the information that Intel has provided. I don't know yet how accurate the Intel information is but I have to follow it, come up with a new version and see how that compares.
CoreTemp has already adjusted TjMax and released a new version. There continues to be no way to adjust for sensor error which Intel now readily admits to."