I suspect, even more now with others reporting the same thing, that it is a temperature sampling issue. I dual boot PureOS and Arch/Antergos, and running TLP (power management) on the latter. When running on battery (Arch), the CPU scaling governor keeps the processor’s frequency well below 1 GHz. However, connecting the laptop to its charger increases the minimum frequency to roughly 2 GHz. This in itself isn’t the problem, and the initial fan spin up resulting from this increase is quite quickly calmed, but stressing the CPU from here on will increase the likelihood of peaks/erroneous data.
So, yes, this part is probably related to the Turboboost-part. I’m guessing that it’s a simple division error, in which a batch of temperature data is divided by old (read: lower clock speed) processor data. Less aggressive scaling will then simply remove most of the trouble.
However, what might be most convenient, is to set the fan’s (or processor’s) temperature reference to the sensor called “pch_skylake-virtual-0” instead of the two “coretemp-isa-0000” core_0 and core_1 values. From what I’ve read, people seem to suggest it’s a real sensor, but it averages perfectly between the two core temperatures without peaks/errors.
I don’t know if this is possible in userspace, as I haven’t found any information on how to set it (as stated in my previous reply: I haven’t even found the fan), but setting it up this way might be preferable in the short run.