Because modern GPUs doesn't have a 1:1 mapping between their "cores" and general purpose logic units. Once you use a branch you effectively collapse your "cores" down to the general purpose logic associated with those "cores." (I think my gtx 660ti has like ~1400 cores and ... 8 general purpose logic units?) For graphics you essentially are just doing vector/matrix math calculations with no branching so you can use all of the "cores" in parallel. This is not to mention that there are costs in streaming data to the GPU you would need to engineer around to get this to work. All my knowledge comes from CUDA land, so maybe not 100% accurate here, but the general principle is probably correct. TL;DR - GPUs are not general purpose chips like x86 CPUs, and there are major caveats when using them.

I looked out my window, and there was a 747 150-200 feet from me. I should clearly see the pilots, and it was being tailed by a f-whatever. You feel your building being shook by that shit and I'll see if you don't go running for the door as well.

Right, it was intelligent design!

You sir don't know how women work. They want someone to listen to their problems and agree with them. They do not want someone to fix their problems. -A Male

Someone needs to play professional football, or flip my burgers if it doesn't work out for em.