V70 Versal Development Card - 3.5 English

Vitis AI Library User Guide (UG1354)
Document ID
UG1354
Release Date
2023-06-29
Version
3.5 English

V70 is a new evaluation platform for the Versal AI data center application. It is based on the VC2802 device and is designed for high throughput AI inference and signal processing computer performance. The primary focus of the V70 card is to enable solution demos to make it easier for customers to develop their own applications.

In Vitis AI 3.5, a C20B14CU1 DPU core is implemented. It delivers 1.05G * 8192 * (20/16) * 14 = 150.5 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 the V70 card with AI Engines clocked at 1050 MHz and PL clocked at 250 MHz.

Table 1. V70 Performance with Batch 14
No Neural Network Input Size GOPS DPU Frequency (MHz) Performance (fps) (Multiple thread)
1 chen_color_resnet18_pt 224x224 3.627 250 9679.80
2 efficientnet_lite_tf2 224x224 0.77 250 9842.18
3 efficientNet-edgetpu-L_tf 300x300 19.36 250 757.68
4 efficientNet-edgetpu-M_tf 240x240 7.34 250 2843.53
5 efficientNet-edgetpu-S_tf 224x224 4.72 250 3654.75
6 face_mask_detection_pt 512x512 0.593 250 2173.85
7 inception_v1_pruned_0_087_tf 224x224 2.73 250 3258.85
8 inception_v1_pruned_0_157_tf 224x224 2.52 250 3008.47
9 inception_v1_tf 224x224 3 250 3692.12
10 inception_v3_pruned_0_2_tf 299x299 9.1 250 1234.00
11 inception_v3_pruned_0_3_pt 299x299 8 250 1289.55
12 inception_v3_pruned_0_4_pt 299x299 6.8 250 1366.46
13 inception_v3_pruned_0_4_tf 299x299 6.9 250 1355.14
14 inception_v3_pruned_0_5_pt 299x299 5.7 250 1482.21
15 inception_v3_pruned_0_6_pt 299x299 4.5 250 1684.68
16 inception_v3_pt 299x299 5.7 250 1143.74
17 inception_v3_tf 299x299 11.5 250 1296.26
18 inception_v3_tf2 299x299 11.5 250 1503.39
19 inception_v4_2016_09_09_tf 299x299 24.6 250 403.08
20 inception_v4_pruned_0_2_tf 299x299 19.56 250 410.59
21 inception_v4_pruned_0_4_tf 299x299 14.79 250 441.69
22 MLPerf_resnet50_v1.5_tf 224x224 8.19 250 3792.32
23 mlperf_ssd_resnet34_tf 1200x1200 433 250 70.16
24 mobilenet_1_0_224_tf2 224x224 1.1 250 14585.10
25 mobilenet_v1_0_25_128_tf 128x128 0.027 250 63108.60
26 mobilenet_v1_1_0_224_pruned_0_11_tf 224x224 1.02 250 14905.70
27 mobilenet_v1_1_0_224_pruned_0_12_tf 224x224 1 250 14936.70
28 mobilenet_v1_1_0_224_tf 224x224 1.1 250 14581.60
29 mobilenet_v2_1_0_224_tf 224x224 0.6 250 11544.90
30 mobilenet_v2_1_4_224_tf 224x224 1.2 250 8081.16
31 movenet_ntd_pt 192x192 0.5 250 8326.82
32 ofa_depthwise_res50_pt 176x176 1.25 250 12925.70
33 ofa_rcan_latency_pt 360x640 45.7 250 53.00
34 ofa_resnet50_0_9B_pt 160x160 0.9 250 7780.10
35 ofa_resnet50_pruned_0_45_pt 224x224 8.2 250 3165.36
36 ofa_resnet50_pruned_0_60_pt 224x224 6 250 3609.05
37 ofa_resnet50_pruned_0_74_pt 192x192 3.6 250 5265.23
38 ofa_yolo_pruned_0_30_pt 640x640 34.71 250 323.62
39 ofa_yolo_pruned_0_50_pt 640x640 24.62 250 456.91
40 ofa_yolo_pt 640x640 48.88 250 295.25
41 pointpillars_kitti_12000_pt 12000x100x4 10.8 250 187.11
42 rcan_pruned_tf 360x640 86.95 250 43.08
43 RefineDet-Medical_EDD_baseline_tf 320x320 81.28 250 339.95
44 RefineDet-Medical_EDD_pruned_0_5_tf 320x320 41.42 250 683.10
45 RefineDet-Medical_EDD_pruned_0_75_tf 320x320 20.54 250 1101.20
46 RefineDet-Medical_EDD_pruned_0_85_tf 320x320 12.32 250 1561.50
47 RefineDet-Medical_EDD_tf 320x320 9.8 250 1769.71
48 resnet_v1_101_pruned_0_346_tf 224x224 9.4 250 2966.77
49 resnet_v1_101_pruned_0_568_tf 224x224 6.21 250 3638.47
50 resnet_v1_101_tf 224x224 14.4 250 2542.30
51 resnet_v1_152_pruned_0_51_tf 224x224 10.68 250 2394.19
52 resnet_v1_152_pruned_0_60_tf 224x224 8.82 250 2614.17
53 resnet_v1_152_tf 224x224 21.8 250 1792.50
54 resnet_v1_50_pruned_0_38_tf 224x224 4.3 250 5038.76
55 resnet_v1_50_pruned_0_65_tf 224x224 2.45 250 6836.10
56 resnet_v1_50_tf 224x224 7 250 4329.41
57 resnet_v2_101_tf 299x299 26.78 250 760.86
58 resnet_v2_152_tf 299x299 40.47 250 564.92
59 resnet_v2_50_tf 299x299 13.1 250 1160.64
60 resnet50_pruned_0_3_pt 224x224 5.8 250 4132.65
61 resnet50_pruned_0_4_pt 224x224 4.9 250 4401.23
62 resnet50_pruned_0_5_pt 224x224 4.1 250 4671.20
63 resnet50_pruned_0_6_pt 224x224 3.3 250 5126.95
64 resnet50_pruned_0_7_pt 224x224 2.5 250 5842.50
65 resnet50_pt 224x224 4.1 250 3792.31
66 resnet50_tf2 224x224 7.7 250 3912.34
67 SESR_S_pt 360x640 7.48 250 298.31
68 squeezenet_pt 224x224 0.82 250 4500.98
69 ssd_mobilenet_v1_coco_tf 300x300 2.5 250 6100.36
70 ssd_mobilenet_v2_coco_tf 300x300 3.8 250 2952.77
71 unet2d_tf2 144x144 24.6 250 1177.35
72 vehicle_make_resnet18_pt 224x224 3.627 250 9686.00
73 vehicle_type_resnet18_pt 224x224 3.627 250 9690.45
74 vgg_16_pruned_0_43_tf 224x224 17.67 250 1958.96
75 vgg_16_pruned_0_5_tf 224x224 15.64 250 2190.40
76 vgg_16_tf 224x224 31 250 619.22
77 vgg_19_pruned_0_24_tf 224x224 29.79 250 759.97
78 vgg_19_pruned_0_39_tf 224x224 23.78 250 1020.87
79 vgg_19_tf 224x224 39.3 250 691.80
80 yolov3_coco_416_tf2 416x416 65.9 250 414.54
81 yolov3_voc_tf 416x416 65.6 250 421.89
82 yolov4_csp_pt 640x640 121 250 88.08
83 yolov4_leaky_416_tf 416x416 60.3 250 232.00
84 yolov4_leaky_512_tf 512x512 91.2 250 96.77
85 yolov5_nano_pt 640x640 4.6 250 539.70
86 yolov5s6_pt 640x640 17 250 85.28
87 yolov6m_pt 640x640 82.2 250 279.67
88 yolov7_pt 640x641 104.8 250 78.08
89 yolov8m_pt 640x642 78.9 250 161.25
90 yolox_nano_pt 416x416 1 250 1253.45