ZCU102 Evaluation Kit - 2.0 English

Vitis AI Library User Guide (UG1354)
Document ID
UG1354
Release Date
2022-01-20
Version
2.0 English

The ZCU102 evaluation kit uses the mid-range ZU9 UltraScale+™ device. There are two different hardware versions of the ZCU102 evaluation kit, one with the serial number 0432055-04 as the header, and the other with the serial number 0432055-05 as the header. The performance of the Vitis AI Library varies between the two hardware versions (because of different DDR memory performance). Because the 0432055-04 version of ZCU102 has been discontinued, the following table only shows the performance for the ZCU102 (0432055-05) evaluation kit. In the ZCU102 evaluation kit, triple B4096F DPU cores are implemented in the program logic.

Refer to the following table for throughput performance (in frames/sec or fps) for various neural network samples on ZCU102 (0432055-05) with DPU running at 281 MHz.

Note: The DPU on the ZCU102 has a hardware softmax acceleration module. Due to the limitation of the hardware softmax module, the software softmax is faster when the number of categories reaches 1000. Set XLNX_ENABLE_C_SOFTMAX=1 to enable the software softmax: softmax_c. The default value of XLNX_ENABLE_C_SOFTMAX is 0, which means the softmax method is selected according to the following priorities.
  1. Neon Acceleration
  2. Hardware Softmax
  3. Software Softmax_c

For ZCU102, use the following command to test the performance of classification.

env XLNX_ENABLE_C_SOFTMAX=1 ./test_performance_classification resnet50 test_performance_classification.list -t 8 -s 60
Table 1. ZCU102 (0432055-05) Performance
No Neural Network Input Size GOPS Performance (fps) (Single thread) Performance (fps) (Multiple thread)
1 bcc_pt 800x1000 268.9 3.2 10.4
2 centerpoint_0_ptcenterpoint_1_pt 2560x40x4 54 16 48.2
3 densebox_320_320 320x320 0.49 432.2 1654.7
4 densebox_640_360 360x640 1.1 211.7 818.9
5 efficientNet-edgetpu-L_tf 300x300 19.36 31.4 87.8
6 efficientNet-edgetpu-M_tf 240x240 7.34 78.2 205.1
7 efficientNet-edgetpu-S_tf 224x224 4.72 112.3 308.2
8 ENet_cityscapes_pt 512x1024 8.6 9.3 36.8
9 face_landmark 96x72 0.14 890.2 1552.9
10 face-quality 80x60 0.06 2584.7 8015.8
11 face-quality_pt 80x60 0.06 2538.9 7962.3
12 facerec_resnet20 112x96 3.5 164.9 334
13 facerec-resnet20_mixed_pt 112x96 3.5 166.7 336.3
14 facerec_resnet64 112x96 11 71 177.7
15 facereid-large_pt 96x96 0.5 889.2 2232.6
16 facereid-small_pt 80x80 0.09 2093.8 6299.2
17 fpn 256x512 8.9 34.5 151.6
18 FPN_Res18_Medical_segmentation 320x320 45.3 12.5 47.1
19 FPN-resnet18_covid19-seg_pt 352x352 22.7 36 106.1
20 FPN-resnet18_Endov 240x320 13.75 36.4 156.5
21 hourglass-pe_mpii 256x256 10.2 18.2 73.8
22 inception_resnet_v2_tf 299x299 26.4 22.4 50.8
23 inception_v1 224x224 3.2 181.8 470.2
24 inception_v1_tf 224x224 3 184.7 467.7
25 inception_v2 224x224 4 132.4 300.7
26 inception_v2_tf 224x224 3.88 91.7 230
27 inception_v3 299x299 11.4 59.1 135.9
28 inception_v3_pt 299x299 5.7 59.1 136.2
29 inception_v3_tf 299x299 11.5 59 135.1
30 inception_v3_tf2 299x299 11.5 58.6 136.7
31 inception_v4 299x299 24.5 28.6 68.5
32 inception_v4_2016_09_09_tf 299x299 24.6 28.7 68.6
33 medical_seg_cell_tf2 128x128 5.3 155.8 394.8
34 MLPerf_resnet50_v1.5_tf 224x224 8.19 71.6 168.5
35 mlperf_ssd_resnet34_tf 1200x1200 433 1.8 7
36 mobilenet_1_0_224_tf2 224x224 1.1 301.2 940.6
37 mobilenet_edge_0_75_tf 224x224 0.62 241.2 706.4
38 mobilenet_edge_1_0_tf 224x224 0.99 200.8 546.9
39 mobilenet_v1_0_25_128_tf 128x128 0.027 1216.3 4500.6
40 mobilenet_v1_0_5_160_tf 160x160 0.15 749.9 2778.7
41 mobilenet_v1_1_0_224_tf 224x224 1.1 305.8 951.3
42 mobilenet_v2 224x224 0.6 254.3 733.3
43 mobilenet_v2_1_0_224_tf 224x224 0.6 248.1 687.8
44 mobilenet_v2_1_4_224_tf 224x224 1.2 181.6 467.7
45 mobilenet_v2_cityscapes_tf 1024x2048 132.74 1.7 5.4
46 MT-resnet18_mixed_pt 512x320 13.65 31.2 101.9
47 multi_task 288x512 14.8 38 127
48 multi_task_v3_pt 320x512 25.44 16.7 60.9
49 openpose_pruned_0_3 368x368 49.9 3.8 15.2
50 personreid-res18_pt 176x80 1.1 356.5 683.8
51 personreid-res50_pt 256x128 5.4 97.3 228.1
52 plate_detection 320x320 0.49 520.4 2080.5
53 plate_num 96x288 1.75 191.4 558.4
54 pmg_pt 224x224 2.28 145.6 362.9
55 pointpainting-pointpainting_nuscenes_40000_64_0_ptpointpainting_nuscenes_40000_64_1_pt 40000x64x16 112 1.3 4.4
56 pointpillars_kitti_12000_0_ptpointpillars_kitti_12000_1_pt 12000x100x4 10.8 19.8 49.9
57 pointpillars_nuscenes_40000_64_0_ptpointpillars_nuscenes_40000_64_1_pt 40000x64x5 108 2.3 9.9
58 rcan_pruned_tf 360x640 86.95 7.6 17.1
59 refinedet_baseline 480x360 123 8.5 24.7
60 RefineDet-Medical_EDD_tf 320x320 9.8 66.3 230.4
61 refinedet_pruned_0_8 360x480 25 32.2 97.9
62 refinedet_pruned_0_92 360x480 10.1 62.4 197.3
63 refinedet_pruned_0_96 360x480 5.1 86 278.2
64 refinedet_VOC_tf 320x320 81.9 11.3 34.4
65 reid 80x160 0.95 357.9 692.9
66 resnet18 224x224 3.7 184.9 477.3
67 resnet50 224x224 7.7 80.3 188.2
68 resnet50_pt 224x224 4.1 71 168.4
69 resnet50_tf2 224x224 7.7 78.8 186.2
70 resnet_v1_101_tf 224x224 14.4 42.7 106.4
71 resnet_v1_152_tf 224x224 21.8 29.2 74.1
72 resnet_v1_50_tf 224x224 7 80 185.6
73 resnet_v2_101_tf 299x299 26.78 20.6 54.4
74 resnet_v2_152_tf 299x299 40.47 14.5 37.5
75 resnet_v2_50_tf 299x299 13.1 35.5 95.7
76 retinaface 360x640 1.11 126.2 541.9
77 salsanext_pt 64x2048 20.4 5.5 21.1
78 salsanext_v2_pt 64x2048 32 4 11.1
79 SemanticFPN_cityscapes_pt 256x512 10 34.1 162.1
80 SemanticFPN_Mobilenetv2_pt 512x1024 5.4 10.2 54
81 semantic_seg_citys_tf2 512x1024 54 7.1 23.9
82 sp_net 128x224 0.55 557 1626.5
83 squeezenet 227x227 0.76 273.9 1078.9
84 squeezenet_pt 224x224 0.82 298.9 1161.2
85 ssd_adas_pruned_0_95 360x480 6.3 87.2 296.8
86 ssd_inception_v2_coco_tf 300x300 9.6 39.1 102.3
87 ssdlite_mobilenet_v2_coco_tf 300x300 1.5 99.1 307.8
88 ssd_mobilenet_v1_coco_tf 300x300 2.5 108 337
89 ssd_mobilenet_v2 360x480 6.6 39 117.4
90 ssd_mobilenet_v2_coco_tf 300x300 3.8 79.4 212.2
91 ssd_pedestrian_pruned_0_97 360x360 5.9 76.2 278.3
92 ssd_resnet_50_fpn_coco_tf 640x640 178.4 2.9 5.2
93 ssd_traffic_pruned_0_9 360x480 11.6 54.9 200.1
94 tiny_yolov3_vmss 416x416 5.46 118.3 393.2
95 unet_chaos-CT_pt 512x512 23.3 22.3 69.5
96 vgg_16_tf 224x224 31 20.1 41.1
97 vgg_19_tf 224x224 39.3 17.3 36.5
98 vpgnet_pruned_0_99 480x640 2.5 94.8 351.7
99 yolov2_voc 448x448 34 26.2 69.3
100 yolov2_voc_pruned_0_66 448x448 11.6 63.1 191.8
101 yolov2_voc_pruned_0_71 448x448 9.9 72.6 224.1
102 yolov2_voc_pruned_0_77 448x448 7.8 84.9 268.9
103 yolov3_adas_pruned_0_9 256x512 5.5 91.4 263.6
104 yolov3_bdd 288x512 53.7 12.4 32.8
105 yolov3_voc 416x416 65.4 12.6 33.1
106 yolov3_voc_tf 416x416 65.6 13.2 34.3
107 yolov4_leaky_spp_m 416x416 60.1 13.2 33.6
108 yolov4_leaky_spp_m_pruned_0_36 416x416 38.2 18 45.3
109 ultrafast_pt 288x800 8.4 33.3 93.9
110 HardNet_MSeg_pt 352x352 22.78 23.3 60.1
111 drunet_pt 528x608 2.59 45.7 161.1
112 person-orientation_pruned_558m_pt 224x112 0.558 613.3 1388.6
113 ofa_resnet50_0_9B_pt 160x160 0.9 169.5 348.1
114 SESR_S_pt 360x640 7.48 81 135
115 c2d2_lite 512x512 6.86 2.9 5.4
116 ofa_depthwise_res50_pt 176x176 1.25 102.8 376.3
117 FairMot_pt 640x480 36 21.9 65.6
118 mobilenet_v3_small_1_0_tf2 224x224 0.132 1.7 6.6
119 clocs 12000x100x4 41 2.9 10.4
120 tsd_yolox_pt 640x640 73 12.9 33.7
121 fadnet_pruned 576x960 154 1.7 2.4
122 ssr_pt 256x256 39.72 5.7 14.2
123 fadnet 576x960 441 1.1 1.5