Acceleration of Deep Neural Network Training Using Field Programmable Gate Arrays

Table 2

Compare with other devices.

Platform	CPU	GPU	FPGA
	Intel® Core i5-4300	AMD Radeon (TM) R5 M330	DE1_SoC
Technology	22 nm	28 nm	28 nm
Power (Watt)	58.5	94.50	18.5
Throughput (GOPS)	28.50	280.60	203.75
Energy efficiency (GOPS/W)	0.487	2.969	11.013