VCK5000 Performance with a 4PE DPUCVDX8H @350MHz - 3.0 English

Vitis AI Library User Guide (UG1354)
Document ID
UG1354
Release Date
2023-01-12
Version
3.0 English

The following table lists the throughput performance (in frames/sec or fps) for various neural network samples on the Versal ACAP VCK5000 Gen4x8 with DPUCVDX8H running at 4PE@350 MHz.

Table 1. VCK5000 Performance with a 4PE DPUCVDX8H @350MHz
No Neural Network Input Size GOPS DPU Frequency (MHz) Performance (fps) (Multiple thread)
1 chen_color_resnet18_pt 224x224 3.627 350 3480.5
2 clocs 12000x100x4 41 350 11.6
3 drunet_pt 528x608 2.59 350 89.0
4 efficientdet_d2_tf 768x768 11.06 350 17.4
5 efficientnet_lite_tf2 224x224 0.77 350 2456.8
6 efficientnet-b0_tf2 224x224 0.36 350 286.1
7 efficientNet-edgetpu-L_tf 300x300 19.36 350 391.6
8 efficientNet-edgetpu-M_tf 240x240 7.34 350 957.1
9 efficientNet-edgetpu-S_tf 224x224 4.72 350 1629.8
10 ENet_cityscapes_pt 512x1024 8.6 350 65.1
11 face_mask_detection_pt 512x512 0.593 350 678.2
12 fadnet 576x960 441 350 9.5
13 fadnet_pruned 576x960 154 350 9.9
14 fadnet_v2_pruned_pt 576x960 201 350 12.3
15 fadnet_v2_pt 576x960 412 350 11.2
16 HRNet_pt 1024x2048 1511.9 350 5.6
17 inception_resnet_v2_tf 299x299 26.4 350 292.2
18 inception_v1_pruned_0_087_tf 224x224 2.73 350 1607.1
19 inception_v1_pruned_0_157_tf 224x224 2.52 350 1627.5
20 inception_v1_tf 224x224 3 350 1524.4
21 inception_v2_tf 224x224 3.88 350 206.9
22 inception_v3_pt 299x299 5.7 350 494.9
23 inception_v3_pruned_0_3_pt 299x299 8 350 564.0
24 inception_v3_pruned_0_4_pt 299x299 6.8 350 605.4
25 inception_v3_pruned_0_5_pt 299x299 5.7 350 654.7
26 inception_v3_pruned_0_6_pt 299x299 4.5 350 747.0
27 inception_v3_tf 299x299 11.5 350 494.6
28 inception_v3_pruned_0_2_tf 299x299 9.1 350 536.5
29 inception_v3_pruned_0_4_tf 299x299 6.9 350 566.4
30 inception_v3_tf2 299x299 11.5 350 487.0
31 inception_v4_2016_09_09_tf 299x299 24.6 350 280.9
32 inception_v4_pruned_0_2_tf 299x299 19.56 350 287.8
33 inception_v4_pruned_0_4_tf 299x299 14.79 350 332.7
34 medical_seg_cell_tf2 128x128 5.3 350 855.9
35 MLPerf_resnet50_v1.5_tf 224x224 8.19 350 1476.8
36 mlperf_ssd_resnet34_tf 1200x1200 433 350 50.3
37 mobilenet_1_0_224_tf2 224x224 1.1 350 6708.6
38 mobilenet_edge_0_75_tf 224x224 0.62 350 3151.4
39 mobilenet_edge_1_0_tf 224x224 0.99 350 2943.6
40 mobilenet_v1_0_25_128_tf 128x128 0.027 350 20846.7
41 mobilenet_v1_0_5_160_tf 160x160 0.15 350 9609.9
42 mobilenet_v1_1_0_224_pruned_0_11_tf 224x224 1.02 350 7205.6
43 mobilenet_v1_1_0_224_pruned_0_12_tf 224x224 1 350 7208.1
44 mobilenet_v1_1_0_224_tf 224x224 1.1 350 6721.4
45 mobilenet_v2_1_0_224_tf 224x224 0.6 350 3931.5
46 mobilenet_v2_1_4_224_tf 224x224 1.2 350 3054.9
47 mobilenet_v2_cityscapes_tf 1024x2048 132.74 350 16.6
48 mobilenet_v3_small_1_0_tf2 224x224 0.132 350 1430.3
49 movenet_ntd_pt 192x192 0.5 350 2831.5
50 ofa_depthwise_res50_pt 176x176 1.25 350 3273.4
51 ofa_rcan_latency_pt 360x640 45.7 350 33.1
52 ofa_resnet50_0_9B_pt 160x160 0.9 350 2154.1
53 ofa_resnet50_baseline_pt 224x224 15 350 673.8
54 ofa_resnet50_pruned_0_45_pt 224x224 8.2 350 868.1
55 ofa_resnet50_pruned_0_60_pt 224x224 6 350 914.7
56 ofa_resnet50_pruned_0_74_pt 192x192 3.6 350 1412.3
57 ofa_yolo_pruned_0_30_pt 640x640 34.71 350 272.6
58 ofa_yolo_pruned_0_50_pt 640x640 24.62 350 316.8
59 ofa_yolo_pt 640x640 48.88 350 227.0
60 pmg_pt 224x224 2.28 350 1998.4
61 pointpainting 40000x64x16 112 350 19.0
62 pointpillars_kitti_12000_pt 12000x100x4 10.8 350 3.1
63 rcan_pruned_tf 360x640 86.95 350 35.6
64 refinedet_VOC_tf 320x320 81.9 350 196.4
65 RefineDet-Medical_EDD_baseline_tf 320x320 81.28 350 196.6
66 RefineDet-Medical_EDD_pruned_0_5_tf 320x320 41.42 350 324.0
67 RefineDet-Medical_EDD_pruned_0_75_tf 320x320 20.54 350 410.0
68 RefineDet-Medical_EDD_pruned_0_85_tf 320x320 12.32 350 518.8
69 RefineDet-Medical_EDD_tf 320x320 9.8 350 536.3
70 resnet_v1_101_pruned_0_346_tf 224x224 9.4 350 1101.7
71 resnet_v1_101_pruned_0_568_tf 224x224 6.21 350 1222.9
72 resnet_v1_101_tf 224x224 14.4 350 1027.2
73 resnet_v1_152_pruned_0_51_tf 224x224 10.68 350 845.9
74 resnet_v1_152_pruned_0_60_tf 224x224 8.82 350 882.7
75 resnet_v1_152_tf 224x224 21.8 350 734.3
76 resnet_v1_50_pruned_0_38_tf 224x224 4.3 350 1715.9
77 resnet_v1_50_pruned_0_65_tf 224x224 2.45 350 2036.2
78 resnet_v1_50_tf 224x224 7 350 1644.7
79 resnet_v2_101_tf 299x299 26.78 350 236.7
80 resnet_v2_152_tf 299x299 40.47 350 165.3
81 resnet_v2_50_tf 299x299 13.1 350 415.6
82 resnet50_pruned_0_3_pt 224x224 5.8 350 1529.9
83 resnet50_pruned_0_4_pt 224x224 4.9 350 1581.4
84 resnet50_pruned_0_5_pt 224x224 4.1 350 1639.4
85 resnet50_pruned_0_6_pt 224x224 3.3 350 1730.7
86 resnet50_pruned_0_7_pt 224x224 2.5 350 1820.0
87 resnet50_pt 224x224 4.1 350 1512.3
88 resnet50_tf2 224x224 7.7 350 1645.2
89 salsanext_pt 64x2048 20.4 350 103.0
90 salsanext_v2_pt 64x2048 32 350 58.7
91 semantic_seg_citys_tf2 512x1024 54 350 59.4
92 SemanticFPN_cityscapes_pt 256x512 10 350 690.2
93 SemanticFPN_Mobilenetv2_pt 512x1024 5.4 350 198.6
94 SESR_S_pt 360x640 7.48 350 107.6
95 solo_pt 640x640 107 350 55.1
96 squeezenet_pt 224x224 0.82 350 3198.3
97 ssd_inception_v2_coco_tf 300x300 9.6 350 104.7
98 ssd_mobilenet_v1_coco_tf 300x300 2.5 350 2663.9
99 ssd_mobilenet_v2_coco_tf 300x300 3.8 350 1259.7
100 ssd_resnet_50_fpn_coco_tf 640x640 178.4 350 83.8
101 ssdlite_mobilenet_v2_coco_tf 300x300 1.5 350 1816.6
102 vehicle_make_resnet18_pt 224x224 3.627 350 3476.6
103 vehicle_type_resnet18_pt 224x224 3.627 350 3476.4
104 vgg_16_pruned_0_43_tf 224x224 17.67 350 589.7
105 vgg_16_pruned_0_5_tf 224x224 15.64 350 638.4
106 vgg_16_tf 224x224 31 350 325.8
107 vgg_19_pruned_0_24_tf 224x224 29.79 350 357.8
108 vgg_19_pruned_0_39_tf 224x224 23.78 350 451.2
109 vgg_19_tf 224x224 39.3 350 291.0
110 xilinxSR_pt 360x640 182.44 350 8.4
111 yolov3_coco_416_tf2 416x416 65.9 350 274.3
112 yolov3_voc_tf 416x416 65.6 350 275.4
113 yolov4_csp_pt 640x640 121 350 127.9
114 yolov4_leaky_416_tf 416x416 60.3 350 235.2
115 yolov4_leaky_512_tf 512x512 91.2 350 134.5
116 yolov5_large_pt 640x640 109.6 350 152.7
117 yolov5-nano_pt 640x640 4.6 350 656.7
118 yolov5s6_pt 640x640 17 350 126.4
119 yolov6m_pt 640x640 82.2 350 205.6
120 yolox_nano_pt 416x416 1 350 1372.1