Nevertheless, I have never seen such a vendor giving a non-fuzzy description of those features. Moreover, after the sinking of Itanium, except for IBM the other server CPU makers use CPU cores that are also designed for consumer applications, so at most the server CPUs may have additional redundancy in some execution units (for validating the results) that is disabled in the consumer variants, to reduce the power consumption. However if that were true I do not see why the CPU vendors would not brag about it, to justify the premium price that is requested for the server CPUs.

For the applications that need very high reliability, it appears that in most cases it is preferable to use multiple CPUs operating in lock-step, which can be compared, ensuring that errors will be detected regardless in which part of the CPU they appear, instead of attempting to use a single CPU that is extremely fault-tolerant by using specially designed redundant execution units (because errors can appear anywhere, including on interconnections and buffers).

Only when non-correctable errors happen, which should be very seldom (e.g. less than one per year), the CPU generates a machine error exception, for which it can be identified that the cause was a (multiple-bit) error in one of the internal cache memories.

Therefore you could sense only extremely-high radiation levels, which would also require a customized OS kernel to recover from exceptions without crashes. Such great radiation levels could be much easier detected with some discrete reverse-biased diodes, which are also easier to place wherever the radiation exists, keeping the microcontroller or whatever device records the radiation levels either away or shielded from the radiation.