VEK280 Evaluation Board - 3.5 English

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

VEK280 is a new evaluation platform for the Versal AI Edge device VE2802. VEK280 targets AI-ML applications with increased compute performance, lower latency, and higher levels of integration enabled by the Versal AI Edge family of devices. The primary focus of VEK280 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 179.2 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 VEK280 with AI Engines clocked at 1250 MHz and PL clocked at 300 MHz.

Table 1. VEK280 Performance with Batch 14
No Neural Network Input Size GOPS 1-thread(fps) Multi-thread(fps)
1 chen_color_resnet18_pt 224x224 3.627 2590.9 6067.8
2 efficientnet_lite_tf2 224x224 0.77 2228.0 5194.7
3 efficientNet-edgetpu-L_tf 300x300 19.36 629.5 961.1
4 efficientNet-edgetpu-M_tf 240x240 7.34 1476.2 3515.6
5 efficientNet-edgetpu-S_tf 224x224 4.72 1729.7 4588.8
6 face_mask_detection_pt 512x512 0.593 517.9 1043.9
7 fadnet_v2 576x960 412 9.5 18.4
8 fadnet_v2_pruned 576x960 201 9.7 19.1
9 HardNet_MSeg_pt 352x352 22.78 279.9 412.3
10 hfnet_tf 960x960 20.09 10.7 24.1
11 inception_v1_tf 224x224 3 1758.4 4611.7
12 inception_v3_pt 299x299 5.7 829.9 1573.0
13 inception_v3_tf 299x299 11.5 831.9 1571.5
14 inception_v3_tf2 299x299 11.5 891.3 1808.4
15 inception_v4_2016_09_09_tf 299x299 24.6 492.5 680.5
16 MLPerf_resnet50_v1.5_tf 224x224 8.19 1730.2 4563.7
17 mlperf_ssd_resnet34_tf 1200x1200 433 17.2 40.6
18 mobilenet_1_0_224_tf2 224x224 1.1 2376.4 5186.8
19 mobilenet_v1_0_25_128_tf 128x128 0.027 5149.5 10575
20 mobilenet_v1_1_0_224_tf 224x224 1.1 2375.6 5171.1
21 mobilenet_v2_1_0_224_tf 224x224 0.6 2292.4 5155.1
22 mobilenet_v2_1_4_224_tf 224x224 1.2 2074.9 5083.5
23 movenet_ntd_pt 192x192 0.5 240.8 422.9
24 ofa_depthwise_res50_pt 176x176 1.25 338.6 530.4
25 ofa_rcan_latency_pt 360x640 45.7 78.9 104.7
26 ofa_resnet50_0_9B_pt 160x160 1.8 2861.7 7949.4
27 ofa_yolo_pruned_0_30_pt 640x640 34.71 182.2 390.8
28 ofa_yolo_pruned_0_50_pt 640x640 24.62 202.6 413.9
29 ofa_yolo_pt 640x640 48.88 172.5 370.2
30 pointpillars_kitti_12000_pt 12000x100x4 10.8 58.8 71.7
31 rcan_pruned_tf 360x640 86.95 66.9 84.6
32 RefineDet-Medical_EDD_tf 320x320 9.8 648.5 1362.0
33 resnet_v1_101_tf 224x224 14.4 1448.0 3010.6
34 resnet_v1_152_tf 224x224 21.8 1207.3 2146.0
35 resnet_v1_50_tf 224x224 7 1795.4 4875.0
36 resnet_v2_101_tf 299x299 26.78 615.4 945.6
37 resnet_v2_152_tf 299x299 40.47 498.5 696.1
38 resnet_v2_50_tf 299x299 13.1 801.3 1471.3
39 resnet50_pt 224x224 4.1 1725.1 4558.2
40 resnet50_tf2 224x224 7.7 1749.1 4709.2
41 SESR_S_pt 360x640 7.48 398.9 627.5
42 squeezenet_pt 224x224 0.82 3929.0 9230.1
43 ssd_mobilenet_v1_coco_tf 300x300 2.5 906.5 1950.3
44 ssd_mobilenet_v2_coco_tf 300x300 3.8 821.679 1912.33
45 superpoint_tf 480x640 52.4 56.0723 124.5
46 unet2d_tf2 144x144 24.6 759.1 1608.5
47 vehicle_make_resnet18_pt 224x224 3.627 2585.1 6148.7
48 vehicle_type_resnet18_pt 224x224 3.627 2610.6 6294.0
49 vgg_16_tf 224x224 31 582.6 733.1
50 vgg_19_tf 224x224 39.3 572.7 718.4
51 yolov3_coco_416_tf2 416x416 65.9 271.0 532.8
52 yolov3_voc_tf 416x416 65.6 302.2 543.3
53 yolov4_csp_pt 640x640 121 78.9 117.4
54 yolov4_leaky_416_tf 416x416 60.3 197.7 365.7
55 yolov4_leaky_512_tf 512x512 91.2 139.8 240.2
56 yolov5_large_pt 640x640 109.6 135.2 268.9
57 yolov5_nano_pt 640x640 4.6 249.0 475.0
58 yolov5l_pt 640x640 109.6 94.4 163.7
59 yolov5m_pt 640x640 109.6 148.2 325.3
60 yolov5s6_pt 640x640 17 50.8 98.1
61 yolov6m_pt 640x640 82.2 39.7 51.6
62 yolov7_pt 640x640 104.8 101.8 161.7
63 yolov8m_pt 640x640 78.9 31.2 69.1
64 yolox_nano_pt 416x416x3 1 711.2 1404.8