Hardly anyone has ever noticed this issue. When someone does notice, they assume that Microsoft must be right and RealTemp is wrong. Here is a quick test so you can come to your own conclusion.
I used the ThrottleStop TS Bench test which allows you to select how big a load you want to use. A Core i7-4700MQ is a hyper threaded 4 core CPU so it can process 8 threads at a time. I set the TS Bench to use 4 threads. In theory, this should create a load of 50%. It seems reasonable that the CPU will need to be active for another 0.5% to 1.0% so it can continue to process the Windows background tasks as this benchmark is running.
ThrottleStop shows the exact load on each thread. Add it all up and divide by 8 and it is showing a 50.9% load while RealTemp is showing a similar load of 50,5%. The slight difference is only because there is a slight difference in timing of these readings. On the other hand, the Task Manager is completely out to lunch. 72% is wrong.
I think the logic here is that Windows is basing its load value on this CPU having a default speed of 2400 MHz. Turbo Boost has this CPU running at a 34 multiplier not a 24 multiplier.
(34 / 24) - 1.4166
1.4166 X 50.5% = 71.5%
That is the only thing that makes some sort of sense to me but ultimately, 72% is wrong. Only half the CPU is being utilized during this test.
RealTemp and ThrottleStop use high performance timers within the CPU and follow a method recommended by Intel to determine the percentage of time the CPU is in the C0 state actively working on a task. The Windows 10 Task Manager graph is useless information but people still use this information to determine if they need a CPU with more cores.
Full Skylake support for RealTemp is on the things to do list. RealTemp 3.70 does not yet correctly support the new 6th Gen Skylake CPUs.