Individual CPU threads are substantially slower than modern GPUs, and from what I’ve seen generally the single GPU thread will be substantially faster than all the CPU threads put together. Probably this is a closer race on a 16 or 32 core CPU, but in most cases enabling CPU+GPU is a modest performance boost, and you want to use a 2.80 or 2.79 experimental build so you can set small rendering tile sizes so that when you get to the end of the render you don’t have an idle GPU and have to wait for all the slow CPU tiles to complete which can slow down the render time a lot. Best to do even tiny 16x16 tiles in many cases now.