Benchmark 是了解硬件平台网络模型运行速度的最佳途径。以下数据基于 AXera-Pi Pro 测试获取，仅供社区参考，不代表商业交付最终性能。
- Pulsar2 3.2
| Models | Input Size | Inference Time(ms)@1 Core | Inference Time(ms)@3 Cores |
|---|---|---|---|
| Inceptionv1 | 224 | 1.166 | 0.495 |
| Inceptionv3 | 299 | 4.370 | 1.670 |
| MobileNetv1 | 224 | 0.567 | 0.277 |
| MobileNetv2 | 224 | 0.595 | 0.282 |
| SqueezeNetv11 | 227 | 0.471 | 0.200 |
| SqueezeNetv10 | 227 | 0.904 | 0.346 |
| ResNet18 | 224 | 1.269 | 0.540 |
| ResNet50 | 224 | 2.867 | 1.335 |
| VGG16 | 224 | 14.135 | 8.141 |
| YOLOv3 | 608 | 43.198 | 14.983 |
| YOLOv5s | 640 | 6.658 | 2.600 |
| YOLOv6s | 640 | 8.913 | 3.126 |
| YOLOv7-Tiny | 640 | 5.938 | 2.771 |
| YOLOv7 | 640 | 34.787 | 12.544 |
| YOLOv8s | 640 | 10.422 | 3.592 |
| YOLOv10s | 640 | 9.099 | 3.300 |
| YOLO11s | 640 | 8.886 | 3.192 |
| YOLOX_s | 640 | 9.668 | 3.405 |
| YOLO-Nas_s | 640 | 12.061 | 4.270 |
| PPYOLOE+_s | 640 | 11.439 | 3.045 |
| Deit_t | 224 | 1.723 | 0.780 |
| Swin_t | 224 | 6.334 | 2.492 |
| ViT_b | 224 | 17.369 | 5.725 |