The proposed FPGA-based architecture was conceptually evaluated against conventional CPU and GPU platforms across three key metrics: power consumption, inference throughput, and energy efficiency.
FPGAs demonstrate significantly lower power consumption compared to GPUs while maintaining competitive inference throughput suitable for real-time edge applications. The architecture achieves superior performance-per-watt, making it highly suitable for resource-constrained embedded environments such as IoT devices, autonomous systems, and smart healthcare applications.
Dynamic voltage-frequency scaling further reduces power consumption during low-workload periods, while quantization and pruning techniques maintain model accuracy within acceptable bounds for edge deployment.