The ZCU104 evaluation board uses the mid-range ZU7ev UltraScale+ device. Dual B4096F DPU cores are implemented in program logic and delivers 2.4 TOPS INT8 peak performance for deep learning inference acceleration.
Refer to the following table for the throughput performance (in frames/sec or fps) for various neural network samples on ZCU104 with DPU running at 300 MHz.
No | Neural Network | Input Size | GOPS | Performance (fps) (Single thread) | Performance (fps) (Multiple thread) |
---|---|---|---|---|---|
1 | densebox_320_320 | 320x320 | 0.49 | 434.7 | 1560.6 |
2 | densebox_640_360 | 360x640 | 1.1 | 225.4 | 784.1 |
3 | ENet_cityscapes_pt | 512x1024 | 8.6 | 9.3 | 39.8 |
4 | face_landmark | 96x72 | 0.14 | 891.3 | 1601.7 |
5 | face-quality | 80x60 | 0.06 | 2187.5 | 6350.5 |
6 | face-quality_pt | 80x60 | 0.06 | 2176.5 | 6612.1 |
7 | facerec_resnet20 | 112x96 | 3.5 | 174.2 | 308.5 |
8 | facerec-resnet20_mixed_pt | 112x96 | 3.5 | 175.9 | 311.1 |
9 | facerec_resnet64 | 112x96 | 11 | 75.8 | 146 |
10 | facereid-large_pt | 96x96 | 0.5 | 884.6 | 1995.9 |
11 | facereid-small_pt | 80x80 | 0.09 | 1914.5 | 5617.9 |
12 | fpn | 256x512 | 8.9 | 34.1 | 128.6 |
13 | FPN_Res18_Medical_segmentation | 320x320 | 45.3 | 12.9 | 33.6 |
14 | FPN-resnet18_covid19-seg_pt | 352x352 | 22.7 | 38.1 | 80.7 |
15 | FPN-resnet18_Endov | 240x320 | 13.75 | 34.7 | 133.8 |
16 | hourglass-pe_mpii | 256x256 | 10.2 | 18.6 | 76 |
17 | inception_resnet_v2_tf | 299x299 | 26.4 | 23.7 | 44.2 |
18 | inception_v1 | 224x224 | 3.2 | 178.6 | 371.3 |
19 | inception_v1_tf | 224x224 | 3 | 182.1 | 377.2 |
20 | inception_v2 | 224x224 | 4 | 132.9 | 263.4 |
21 | inception_v2_tf | 224x224 | 3.88 | 93.7 | 188.2 |
22 | inception_v3 | 299x299 | 11.4 | 62.1 | 121.4 |
23 | inception_v3_pt | 299x299 | 5.7 | 62.2 | 121.7 |
24 | inception_v3_tf | 299x299 | 11.5 | 61.5 | 120.3 |
25 | inception_v3_tf2 | 299x299 | 11.5 | 60.6 | 118.9 |
26 | inception_v4 | 299x299 | 24.5 | 30.4 | 59 |
27 | inception_v4_2016_09_09_tf | 299x299 | 24.6 | 30.4 | 59 |
28 | medical_seg_cell_tf2 | 128x128 | 5.3 | 163.2 | 341.5 |
29 | MLPerf_resnet50_v1.5_tf | 224x224 | 8.19 | 75.6 | 146.2 |
30 | mlperf_ssd_resnet34_tf | 1200x1200 | 433 | 1.8 | 5.1 |
31 | mobilenet_1_0_224_tf2 | 224x224 | 1.1 | 306.4 | 739.8 |
32 | mobilenet_edge_0_75_tf | 224x224 | 0.62 | 249.8 | 569.6 |
33 | mobilenet_edge_1_0_tf | 224x224 | 0.99 | 208.6 | 454.5 |
34 | mobilenet_v1_0_25_128_tf | 128x128 | 0.027 | 1147.4 | 3973.5 |
35 | mobilenet_v1_0_5_160_tf | 160x160 | 0.15 | 736.7 | 2132.6 |
36 | mobilenet_v1_1_0_224_tf | 224x224 | 1.1 | 311.1 | 750.2 |
37 | mobilenet_v2 | 224x224 | 0.6 | 262.3 | 599.2 |
38 | mobilenet_v2_1_0_224_tf | 224x224 | 0.6 | 256.8 | 578.4 |
39 | mobilenet_v2_1_4_224_tf | 224x224 | 1.2 | 189.2 | 399 |
40 | mobilenet_v2_cityscapes_tf | 1024x2048 | 132.74 | 1.6 | 4.6 |
41 | MT-resnet18_mixed_pt | 512x320 | 13.65 | 30.5 | 90 |
42 | multi_task | 288x512 | 14.8 | 36.8 | 108.2 |
43 | openpose_pruned_0_3 | 368x368 | 49.9 | 3.7 | 11 |
44 | personreid-res18_pt | 176x80 | 1.1 | 371.8 | 684.5 |
45 | personreid-res50_pt | 256x128 | 5.4 | 102.8 | 198.2 |
46 | plate_detection | 320x320 | 0.49 | 513.8 | 1909.6 |
47 | plate_num | 96x288 | 1.75 | 161.9 | 446 |
48 | pointpillars_kitti_12000_0_pt pointpillars_kitti_12000_1_pt | 12000x100 | 10.8 | 20.2 | 48.8 |
49 | refinedet_baseline | 480x360 | 123 | 9.1 | 18.4 |
50 | RefineDet-Medical_EDD_tf | 320x320 | 9.8 | 69.6 | 169.6 |
51 | refinedet_pruned_0_8 | 360x480 | 25 | 34.6 | 74.5 |
52 | refinedet_pruned_0_92 | 360x480 | 10.1 | 66.6 | 154.7 |
53 | refinedet_pruned_0_96 | 360x480 | 5.1 | 91.9 | 227.2 |
54 | refinedet_VOC_tf | 320x320 | 81.9 | 10.8 | 25.8 |
55 | reid | 80x160 | 0.95 | 372.9 | 698.6 |
56 | resnet18 | 224x224 | 3.7 | 193.4 | 414.6 |
57 | resnet50 | 224x224 | 7.7 | 77.2 | 149.7 |
58 | resnet50_pt | 224x224 | 4.1 | 73.1 | 142.4 |
59 | resnet50_tf2 | 224x224 | 7.7 | 75.1 | 146.2 |
60 | resnet_v1_101_tf | 224x224 | 14.4 | 45.3 | 87.9 |
61 | resnet_v1_152_tf | 224x224 | 21.8 | 31 | 60.2 |
62 | resnet_v1_50_tf | 224x224 | 7 | 84.5 | 162.7 |
63 | resnet_v2_101_tf | 299x299 | 26.78 | 21.8 | 45.1 |
64 | resnet_v2_152_tf | 299x299 | 40.47 | 15.4 | 31 |
65 | resnet_v2_50_tf | 299x299 | 13.1 | 37 | 81.6 |
66 | retinaface | 360x640 | 1.11 | 129.5 | 480.8 |
67 | salsanext_pt | 64x2048 | 20.4 | 5.3 | 19.8 |
68 | SemanticFPN_cityscapes_pt | 256x512 | 10 | 34 | 132.6 |
69 | semantic_seg_citys_tf2 | 512x1024 | 54 | 7.1 | 24.2 |
70 | sp_net | 128x224 | 0.55 | 507.9 | 1134 |
71 | squeezenet | 227x227 | 0.76 | 275.6 | 980 |
72 | squeezenet_pt | 224x224 | 0.82 | 227.9 | 737.4 |
73 | ssd_adas_pruned_0_95 | 360x480 | 6.3 | 91 | 237.9 |
74 | ssd_inception_v2_coco_tf | 300x300 | 9.6 | 40.8 | 85.8 |
75 | ssdlite_mobilenet_v2_coco_tf | 300x300 | 1.5 | 104.9 | 265.7 |
76 | ssd_mobilenet_v1_coco_tf | 300x300 | 2.5 | 112.8 | 295.3 |
77 | ssd_mobilenet_v2 | 360x480 | 6.6 | 25.6 | 100.3 |
78 | ssd_mobilenet_v2_coco_tf | 300x300 | 3.8 | 80.6 | 190.3 |
79 | ssd_pedestrian_pruned_0_97 | 360x360 | 5.9 | 79.7 | 214.9 |
80 | ssd_resnet_50_fpn_coco_tf | 640x640 | 178.4 | 2.9 | 5.2 |
81 | ssd_traffic_pruned_0_9 | 360x480 | 11.6 | 57.5 | 150 |
82 | tiny_yolov3_vmss | 416x416 | 5.46 | 123.2 | 328.8 |
83 | unet_chaos-CT_pt | 512x512 | 23.3 | 21.9 | 59.1 |
84 | vgg_16_tf | 224x224 | 31 | 21.4 | 37.1 |
85 | vgg_19_tf | 224x224 | 39.3 | 18.5 | 32.7 |
86 | vpgnet_pruned_0_99 | 480x640 | 2.5 | 100.5 | 311.4 |
87 | yolov2_voc | 448x448 | 34 | 28 | 57.4 |
88 | yolov2_voc_pruned_0_66 | 448x448 | 11.6 | 66.4 | 153.3 |
89 | yolov2_voc_pruned_0_71 | 448x448 | 9.9 | 76.2 | 180.7 |
90 | yolov2_voc_pruned_0_77 | 448x448 | 7.8 | 88.6 | 216.8 |
91 | yolov3_adas_pruned_0_9 | 256x512 | 5.5 | 92.8 | 231.9 |
92 | yolov3_bdd | 288x512 | 53.7 | 13.1 | 26.7 |
93 | yolov3_voc | 416x416 | 65.4 | 13.5 | 27.2 |
94 | yolov3_voc_tf | 416x416 | 65.6 | 14.1 | 28.2 |
95 | yolov4_leaky_spp_m | 416x416 | 60.1 | 13.9 | 28.5 |