In general GPU's are same or better in floating points per given area of silicon. on top of that , FPGA's add other expenses due to lower market size. Maybe FP

In general GPU's are same or better in floating points per given area of silicon. on top of that , FPGA's add other expenses due to lower market size. Maybe FPGA's are competitive in integer math, but that's a pretty small niche ,not big enough for a coprocessor.