My best guess is that TJMax is not consistent across all of your cores. With the cores numbered from Core 0 to Core 5, I'd bump up TJMax on Core 3 by about 10C or 11C and I'd bump Core 4 by about 6C. That should make all 6 of them line up pretty good. They might even be more accurate too.

Prime95 Small FFTs is still the best program for equally loading each core. When you do that, most of the difference you see from one core to the next is error in how Intel sets TJMax.

A TJ target number is written into each core so in theory Intel could use a unique TJ Max value for each core but they don't bother. They just write a consistent value into all 6 cores even though actual TJMax varies a little from one core to the next.

I'll try to get RealTemp GT updated someday in the near future with better CPU name recognition and some of the newer GPU code but other than that, it looks like it still works.