Hardware Performance and Efficiency Metrics on MNIST
6.13Clock Time (ns)EfficientKAN
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| EfficientKANPrecision=W4, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 6.13 | 4.03 | 24.72 | 40.5 | 1.4 | |
| PyKANPrecision=W4, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 6.13 | 3.96 | 24.25 | 41.2 | 1.43 | |
| FastKANPrecision=W4, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 6.13 | 260 | 1.59 | 627.4 | 3.32 | |
| GRAMLayerPrecision=FP, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 6.13 | 428 | 2.62 | 381.8 | — | |
| GRAMLayerPrecision=W4, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 6.13 | 271 | 1.66 | 603.4 | 1.58 | |
| EfficientKANPrecision=FP, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 8.75 | 3.96 | 34.62 | 28.9 | — | |
| PyKANPrecision=FP, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 8.75 | 3.96 | 34.66 | 28.8 | — | |
| FastKANPrecision=FP, Hardware=Xilinx Zynq UltraScale+ XCZU7EV2025.11 | 8.75 | 603 | 5.28 | 189.5 | — |