Yeah, not the case. Boost is a set MHz rating above "stock". It gets lowered if power usage or temps exceed their set points. Otherwise, they'd not be selling cards with different clocks, and there'd be no OC'ing, since it'd be automatic.
This is true, but not the whole truth. (Yeah, that's DEEP!)
If you push Power Limit to max and manually push core clocks, you can still see boost clock doing some work.
1. it will limit the clocks whenever it hits the temp limit or comes close to it (pretty advanced detection here, it generally clocks back just right to get temps in check)
2. it will push beyond your max core clocks with the same boost bins as it had before your OC, if there is power/temperature headroom
Therefore analyzing boost clock behaviour is a very good tool to see quickly if your OC is actually making sense. If it is not using all the original boost bins, you might as well dial back the core clock OC. In a sense, this makes the boost clocks a very efficient OC stability tool, because they are a warning sign that you are approaching your GPU's limits in power delivery/achievable core clock.
Controlling boost clocks through just the temperature slider is generally a bad idea because it will increase the aggressiveness of the temperature limiting through boost clocks. If your airflow is right, any card should be fine with a very loose temp setting because the driver will limit the card in any case. If you have temperature issues, the best way to solve them is to adjust the fan profile to keep a card within 'boostable' temps.
To chime in on trog100: the reason stock power settings limit boost clocks in some games while they do not in others, is because core clock is not the whole story of GPU utilization. Some games will use more of the GPU internals than others. Tesselation is a good example of this - generally games that use a lot of tesselation will cause a hotter GPU while showing similar utilization on the OSD. If you test overclocks; use precisely these games to really test your OC. This is also a reason the Valley, and in a greater sense the Heaven bench are very good for stability testing of OCs (and yes, their GPU clock readings are off by a large margin).
Last, boost clocks do not affect memory at all, contrary to what Toothless assumed.