Compute coord/atom vs compute cluster/atom

what are the difference(s) between compute coord/atom cutoff and compute cluster/atom commands, with the exception of per-atom array for coord/atom if multiple typeNs ?

what applications would you use one vs the other, and why ?