VCK5000 Performance with a 6PE DPUCVDX8H-aieDWC @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-aieDWC running at 6PE@350 MHz.

Table 1. VCK5000 Performance with a 6PE DPUCVDX8H-aieDWC @350MHz
No Neural Network Input Size GOPS DPU Frequency (MHz) Performance (fps) (Multiple thread)
1 chen_color_resnet18_pt 224x224 3.627 350 6617.8
2 drunet_pt 528x608 2.59 350 150.6
3 efficientnet_lite_tf2 224x224 0.77 350 3902.2
4 efficientNet-edgetpu-M_tf 240x240 7.34 350 1339.5
5 efficientNet-edgetpu-S_tf 224x224 4.72 350 2480.9
6 ENet_cityscapes_pt 512x1024 8.6 350 159.0
7 face_mask_detection_pt 512x512 0.593 350 1420.6
8 fadnet 576x960 441 350 13.6
9 fadnet_pruned 576x960 154 350 12.4
10 fadnet_v2_pruned_pt 576x960 201 350 16.6
11 fadnet_v2_pt 576x960 412 350 14.6
12 HRNet_pt 1024x2048 1512 350 7.9
13 inception_resnet_v2_tf 299x299 26.4 350 486.1
14 inception_v1_pruned_0_087_tf 224x224 2.73 350 3780.7
15 inception_v1_pruned_0_157_tf 224x224 2.52 350 3842.1
16 inception_v1_tf 224x224 3 350 3461.2
17 inception_v2_tf 224x224 3.88 350 327.2
18 inception_v3_pt 299x299 5.7 350 935.6
19 inception_v3_pruned_0_3_pt 299x299 8 350 1065.4
20 inception_v3_pruned_0_4_pt 299x299 6.8 350 1147.6
21 inception_v3_pruned_0_5_pt 299x299 5.7 350 1214.5
22 inception_v3_pruned_0_6_pt 299x299 4.5 350 1364.9
23 inception_v3_tf 299x299 11.5 350 924.1
24 inception_v3_pruned_0_2_tf 299x299 9.1 350 992.1
25 inception_v3_pruned_0_4_tf 299x299 6.9 350 932.9
26 inception_v3_tf2 299x299 11.5 350 959.9
27 inception_v4_2016_09_09_tf 299x299 24.6 350 506.1
28 inception_v4_pruned_0_2_tf 299x299 19.56 350 495.7
29 inception_v4_pruned_0_4_tf 299x299 14.79 350 599.5
30 medical_seg_cell_tf2 128x128 5.3 350 1212.6
31 MLPerf_resnet50_v1.5_tf 224x224 8.19 350 3400.5
32 mlperf_ssd_resnet34_tf 1200x1200 433 350 77.3
33 mobilenet_1_0_224_tf2 224x224 1.1 350 8603.1
34 mobilenet_edge_0_75_tf 224x224 0.62 350 5659.1
35 mobilenet_edge_1_0_tf 224x224 0.99 350 5158.4
36 mobilenet_v1_0_25_128_tf 128x128 0.027 350 23036.6
37 mobilenet_v1_0_5_160_tf 160x160 0.15 350 15782.7
38 mobilenet_v1_1_0_224_pruned_0_11_tf 224x224 1.02 350 8813.2
39 mobilenet_v1_1_0_224_pruned_0_12_tf 224x224 1 350 8810.8
40 mobilenet_v1_1_0_224_tf 224x224 1.1 350 8607.2
41 mobilenet_v2_1_0_224_tf 224x224 0.6 350 6572.6
42 mobilenet_v2_1_4_224_tf 224x224 1.2 350 4970.4
43 mobilenet_v2_cityscapes_tf 1024x2048 132.7 350 16.4
44 movenet_ntd_pt 192x192 0.5 350 3268.7
45 ofa_depthwise_res50_pt 176x176 1.25 350 3755.2
46 ofa_rcan_latency_pt 360x640 45.7 350 49.7
47 ofa_resnet50_0_9B_pt 160x160 0.9 350 4049.4
48 ofa_resnet50_baseline_pt 224x224 15 350 1212.9
49 ofa_resnet50_pruned_0_45_pt 224x224 8.2 350 1787.1
50 ofa_resnet50_pruned_0_60_pt 224x224 6 350 1739.3
51 ofa_resnet50_pruned_0_74_pt 192x192 3.6 350 2830.2
52 ofa_yolo_pruned_0_30_pt 640x640 34.71 350 381.9
53 ofa_yolo_pruned_0_50_pt 640x640 24.62 350 431.2
54 ofa_yolo_pt 640x640 48.88 350 257.2
55 pmg_pt 224x224 2.28 350 3547.9
56 pointpainting 40000x64x16 112 350 18.1
57 rcan_pruned_tf 360x640 86.95 350 53.5
58 refinedet_VOC_tf 320x320 81.9 350 304.1
59 RefineDet-Medical_EDD_baseline_tf 320x320 81.28 350 304.6
60 RefineDet-Medical_EDD_pruned_0_5_tf 320x320 41.42 350 522.8
61 RefineDet-Medical_EDD_pruned_0_75_tf 320x320 20.54 350 633.1
62 RefineDet-Medical_EDD_pruned_0_85_tf 320x320 12.32 350 941.3
63 RefineDet-Medical_EDD_tf 320x320 9.8 350 975.1
64 resnet_v1_101_pruned_0_346_tf 224x224 9.4 350 2510.0
65 resnet_v1_101_pruned_0_568_tf 224x224 6.21 350 3115.9
66 resnet_v1_101_tf 224x224 14.4 350 2231.3
67 resnet_v1_152_pruned_0_51_tf 224x224 10.68 350 2078.8
68 resnet_v1_152_pruned_0_60_tf 224x224 8.82 350 2301.8
69 resnet_v1_152_tf 224x224 21.8 350 1586.1
70 resnet_v1_50_pruned_0_38_tf 224x224 4.3 350 4024.5
71 resnet_v1_50_pruned_0_65_tf 224x224 2.45 350 5472.2
72 resnet_v1_50_tf 224x224 7 350 3732.8
73 resnet50_pruned_0_3_pt 224x224 5.8 350 3496.4
74 resnet50_pruned_0_4_pt 224x224 4.9 350 3721.0
75 resnet50_pruned_0_5_pt 224x224 4.1 350 3975.3
76 resnet50_pruned_0_6_pt 224x224 3.3 350 4387.6
77 resnet50_pruned_0_7_pt 224x224 2.5 350 4819.4
78 resnet50_pt 224x224 4.1 350 3420.2
79 resnet50_tf2 224x224 7.7 350 3734.9
80 salsanext_pt 64x2048 20.4 350 203.2
81 salsanext_v2_pt 64x2048 32 350 87.1
82 semantic_seg_citys_tf2 512x1024 54 350 114.4
83 SemanticFPN_cityscapes_pt 256x512 10 350 1040.8
84 SemanticFPN_Mobilenetv2_pt 512x1024 5.4 350 207.2
85 SESR_S_pt 360x640 7.48 350 184.3
86 squeezenet_pt 224x224 0.82 350 6658.9
87 ssd_inception_v2_coco_tf 300x300 9.6 350 164.8
88 ssd_mobilenet_v1_coco_tf 300x300 2.5 350 2876.0
89 ssd_mobilenet_v2_coco_tf 300x300 3.8 350 2606.1
90 ssd_resnet_50_fpn_coco_tf 640x640 178.4 350 103.1
91 ssdlite_mobilenet_v2_coco_tf 300x300 1.5 350 2507.9
92 vehicle_make_resnet18_pt 224x224 3.627 350 6615.8
93 vehicle_type_resnet18_pt 224x224 3.627 350 6620.7
94 vgg_16_pruned_0_43_tf 224x224 17.67 350 980.4
95 vgg_16_pruned_0_5_tf 224x224 15.64 350 1052.0
96 vgg_16_tf 224x224 31 350 502.7
97 vgg_19_pruned_0_24_tf 224x224 29.79 350 560.5
98 vgg_19_pruned_0_39_tf 224x224 23.78 350 721.6
99 vgg_19_tf 224x224 39.3 350 445.5
100 xilinxSR_pt 360x640 182.4 350 11.4
101 yolov3_coco_416_tf2 416x416 65.9 350 386.6
102 yolov3_voc_tf 416x416 65.6 350 391.1
103 yolov4_csp_pt 640x640 121 350 174.6
104 yolov5_large_pt 640x640 109.6 350 195.7
105 yolov5-nano_pt 640x640 4.6 350 872.1
106 yolov5s6_pt 640x640 17 350 160.3
107 yolov6m_pt 640x640 82.2 350 279.5