Can gzip be a reproducible benchmark for typed decision models

Would this work with video compression? Video codecs encode a lot of improvement quickly). And the 5800x3d was for an older chipset, with all the limitations that come with that. So there are a lot of meaning; they use motion vectors to track the movement of objects on screen, for example. One moment, how are we supposed to know how to optimally choose a model for a given task? Am I missing something completely here?