Please don't take the power usage here, let's consider only the hash power and the cost between XCVU9P ($4,000 FPGA) vs GTX 1080 Ti ($800 GPU)
The difference in cost is 5x, does it apply to the hashpower or not?
How quickly does FPGA programming adapt to new algos or the change of old algos?
Unfortunately, you cannot ignore power consumption, it is the primary reason to use them in the first place.
An efficient implementation of a new algorithm on an FPGA can take months, depending on it's complexity and the skill of the circuit designer.