I'm not a master at overclocking and I only ever mined with Nvidia hardware but here's what I know:
Every card have different limits so you either use different speeds for each cards or use the lowest OC in a rig for all the cards.
And then different algos have different levels of overclockability - as it were - as they use cards differently. For example, you could get +300 Mhz core on some algos with no problem which would instantly crash on others. So if you mine different algos you also have to use the lowest OC.
And some algos prefer core clock, some prefer memory clock OC. And since the cards are handling power pretty intelligently, for some algos it's beneficial if you downclock the memory so the card can allocate more power to core therefore boost higher. Rarely it's the opposite.
Also wirth noting that memory modules can get really hot (95+°C) on some cards mining memory heavy algos such as Eth (
source). Which you can't see with software as there are no temp sensors there usually. But with the right hardware you can see for yourself.
And while we're at temps the voltage regulation modules also can get super hot if cards are overvolted. And some models have very poor VRM cooling.
Also, an overclock might look stable, in reality it might not be and it will take a couple of hours or even days for the card to crash.
Efficiency is more important these days so reducing power limit and just letting the card boost as high as it does with a mild OC is the safest bet and figuring out the maximum before crashing levels of OC just doesn't worth the effort - especially for multiple rigs.
My recommendation is to just set a mild OC across rigs and if cards crash in a day or two decrease the OC (and if it always happen to the same cards, note them and downclock only those) and if not, then you can increase the OC.