FPGA architecture allows for many algorithm implementations where the sustained performance is much closer to the device’s peak performance when compared to quad-core CPUs or GPGPUs. The strong benchmarking results from FPGA accelerators will continue to improve with appropriate focus on the silicon, arithmetic, and library foundations. As even the largest FPGAs currently consume less than 30W of power, FPGA roadmaps are well below datacenter power and cooling limitations.
Author: Martin Langhammer, Principal Scientist, Altera Corporation