In this page, we provide performance benchmarks of DeGirum ORCA1 AI accelerator on various models. The frames per second (FPS) numbers are obatained by running the single model performance test.ipynb jupyter notebook on a machine equipped with ORCA1. The script can also be run on the cloud platform to estimate the performance. All FPS numbers are for batch_size=1. This page will be periodically updated to reflect the latest performance numbers. As our compiler and software mature, we expect to add more models and also improve the performance.
Last updated on: May 22, 2023
| Model Name | FPS |
|---|---|
| efficientnet_es_imagenet--224x224_quant | 187 |
| mobiledet_coco--320x320_quant | 128 |
| mobilenet_v1_imagenet--224x224_quant | 407 |
| mobilenet_v2_imagenet--224x224_quant | 360 |
| resnet50_imagenet--224x224_pruned_quant | 250 |
| yolo_v5s_face_det--512x512_quant | 126 |