I have Sandra, Aida, ET4 and MBM5. The program has to find where to read the built-in sensor, which on my board is a Winbond W83627HF at an ISA address of 290H. The readings will then be directly from the source in or on the board.
If the program has found it correctly, the readings will be identical to the BIOS reading 'cause they're looking in the same place and getting the same info. That will not necessarily mean the readings will be correct if the BIOS or the board aren't up to snuff.
I find Sandra and Aida find the same sensor and have the same readings. MBM5 and ET4 don't seem to get it right on my board, though they may get it right on others. I think there may be some different interpretations going on in these programs which may give differing figures. I find Aida is generally lower, though at the moment Aida and Sandra are reporting the same figures.
The correct CPU reading should be from inside the CPU itself and may tend to be higher and may not be a good indication of what's actually getting dissapated out the HSF. On the other hand, if you have a good HSF with an excellent thermal bond with the CPU, the reading from inside the CPU will be a good indication of whether you are getting the heat out of there.
If, under load, you see your temps soaring out of control, either your HSF isn't the right one or the thermal bond isn't what it should be. Or the hot air isn't getting out of the box.
I find it takes a lot of observation and experimenting to find out what is right and what works.