TJMax is not an exact number. I always had this belief that once we found exactly what TJMax was then this problem would all be solved and we'd have 100% accurate core temperatures but that's not true.

Intel's TJ Target number that they released does not directly translate into a TJMax number. Maybe TJ Target is some sort of design goal they were shooting for. Depending on whether you have a 45nm, 65nm, QX or Xeon CPU will somewhat determine how close TJ Target is to your actual TJMax.

The E8x00 45nm Dual Core series is a good example. Users agree that TJMax = 100C but there is a certain amount of manufacturing error in that number so actual TJMax for each CPU will be different than that theoretical TJMax. There is no available documentation which shows what the average, minimum or maximum TJMax is for a CPU line and no documentation showing how much that average value varies by.

You would think that Intel would have some information somewhere which shows how much error is typical for each CPU but if they do, they're not willing to share it. Telling us that the amount of error equals X is pretty useless information.

Slope error only explains a small part of what's going on here. There are 45nm Quads where actual TJMax is different for each set of cores. I think JohnZS has one of the more extreme examples where TJMax seems to be different for each individual core.

i43: I apologize for confusing your testing.

I just don't believe that the 95C Intel TJ Target number accurately reflects your actual TJMax. RealTemp is getting ready for the next official release so I wanted to get that adjusted and taken care of.

I had a look at this picture of yours:
http://img521.imageshack.us/img521/3...25244lchx8.png

It shows during the Cool Down Test that core0/core1 changes by about 7. Core 3 is firmly stuck at 68 so I'll ignore that one. Core 2 is moving at the start of the test but gets stuck at 74. In theory it should move a similar amount to core0/core1 during this test or maybe a degree or two less. Quads tend to do that.

Instead of moving 7, if it didn't get stuck, I think it would move at least 5 so it should be reading about 76 at the Idle stage. If you were to use TJMax = 95C, your reported core temperature for this core would be 19C or 4C below your room temperature which isn't very believable. With your cooler and room temperature, it should be reading somewhere around 31C so it is off by 12C.

The difference in slope between your two sets of cores is minimal. It's not the cause of that much error. If it was my CPU, I'd set core0/core1 to TJMax = 100C and now here's the crazy part. I'd set TJMax = 105C for core2/core3. After doing this I think your reported core temperatures for all 4 cores would be a lot closer to the actual temperature. Far closer than using Intel's TJ Target value of 95.

I can't ever see users willing to accept different TJMax values on the same CPU but that's the only logical explanation for the data I'm seeing. It also agrees with what Intel has told us about the manufacturing process and the errors that occur when calibrating TJMax for each CPU in an assembly line environment.

Of course burebista's idea about forgetting about accurate core temperatures and just going by Distance to TJMax is probably the smartest thing to do.