I've programmed both FPGAs and GPUs. They are entirely different and shouldn't be compared with the exception of the small-ish class of tasks that map equally well to both (smaller than you think). FPGAs are more suitable to, say, prototype a new GPU or some other custom IC. It's more "arcane" and "less cost-effective" by definition because you are one abstraction level lower. For further reading, you really need to understand how a hardware design and manufacturing pipeline works to see why FPGAs and GPUs are so apples and oranges.

For consumers it won't really matter, but for server applications this can be profound.

If Intel can clean up the tooling for these things it could make them a lot more popular.