GPU manufacturers are charging us per CUDA core these days, meanwhile my Intel CPU from 2015 still runs 90% of my codebase just fine. What exactly am I supposed to be parallelizing, my sanity?