AMD uses ECC on newer cards and if any error occurs, it resends data or recomputes until it reaches target.
When there is so much instability, it stucks in an infinite re-send loop or bsods. So it is very hard to find a point which it is %0 error compute frequency. An error could be present even at 1 MHz(not even GHz) working frequencies because of quantum-mechanics' tunneling probability. Anything can happen in nano world at anytime.
Only you can know if error happens.
Lets say r9-280x starts artifacting %99.99 at 2GHz(before it explodes) .
@1.5 GHz, it is %90 per cycle (just trivial numbers to have an idea of tunneling)
@1.3 GHz, it is %10 per cycle (trivial again), ECC could stand for a second before getting in bsod or smt like that
@500 MHz, it is %0.01 per cycle but ECC takes care of it and performance impact is minimal.
@1 MHz, it is %0.000000000000001 but it is there and will not vanish. ECC makes it nearly impossible to lock your game/compute.
If gaming cards were %100 stable, there wouldnt be a need for server-grade components to run a web site 7/24.
You just need to watch out for decreased gaming/computing experience when crossing factory frequencies.
If it is stable for 24 hours, it may stay stable for another 24 hours(or not).
So, if it is not stable for 24 hours for any frequency range, then you should not use it for anything.