[confluence]

GPUs suck at any problem that cannot be easily divided. If you can map a function over arbitrary chunks of self-contained data GPUs will perform better.