Alveo U55C High Performance Compute Card - 2.0 English

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

The Xilinx® Alveo™ U55C high performance compute card provides optimized acceleration for workloads in high performance computing (HPC), big data analytics and search, financial computing, computational storage and machine learning. In this release, the DPU is implemented in program logic for deep learning inference acceleration.

U55C Performance with 11PE300 MHz DPUCAHX8H-DWC

Refer to the following table for the throughput performance (in frames/sec or fps) for various neural network samples on U55C with DPUCAHX8H-DWC running at 11PE@300 MHz.

Table 1. U55C Performance with 11PE300 MHz DPUCAHX8H-DWC
No Neural Network Input Size GOPS DPU Frequency (MHz) Performance (fps) (Multiple thread)
1 densebox_320_320 320x320 0.49 300 4922.83
2 densebox_640_360 360x640 1.1 300 2205.46
3 drunet_pt 528x608 2.59 300 479.90
4 ENet_cityscapes_pt 512x1024 8.6 300 144.02
5 face_landmark 96x72 0.14 300 19902.40
6 face-quality 80x60 0.06 300 30667.30
7 face-quality_pt 80x60 0.06 300 30562.20
8 facerec_resnet20 112x96 3.5 300 2500.45
9 facerec-resnet20_mixed_pt 112x96 3.5 300 2499.11
10 facerec_resnet64 112x96 11 300 907.31
11 facereid-large_pt 96x96 0.5 300 15187.80
12 facereid-small_pt 80x80 0.09 300 33132.60
13 FairMot_pt 640x480 36 300 239.92
14 fpn 256x512 8.9 300 725.33
15 FPN_Res18_Medical_segmentation 320x320 45.3 300 182.27
16 FPN-resnet18_covid19-seg_pt 352x352 22.7 300 408.05
17 inception_resnet_v2_tf 299x299 26.4 300 301.68
18 inception_v1 224x224 3.2 300 2135.29
19 inception_v1_tf 224x224 3 300 2214.22
20 inception_v2 224x224 4 300 1731.11
21 inception_v3 299x299 11.4 300 697.38
22 inception_v3_pt 299x299 5.7 300 697.54
23 inception_v3_tf 299x299 11.5 300 698.57
24 inception_v3_tf2 299x299 11.5 300 714.44
25 inception_v4 299x299 24.5 300 326.77
26 inception_v4_2016_09_09_tf 299x299 24.6 300 327.53
27 medical_seg_cell_tf2 128x128 5.3 300 2021.76
28 MLPerf_resnet50_v1.5_tf 224x224 8.19 300 1015.18
29 mlperf_ssd_resnet34_tf 1200x1200 433 300 25.83
30 mobilenet_1_0_224_tf2 224x224 1.1 350 5340.33
31 mobilenet_v1_0_25_128_tf 128x128 0.027 350 18874.90
32 mobilenet_v1_0_5_160_tf 160x160 0.15 350 12898.20
33 mobilenet_v1_1_0_224_tf 224x224 1.1 350 5325.89
34 mobilenet_v2 224x224 0.6 350 5370.48
35 multi_task 288x512 14.8 300 536.58
36 ofa_resnet50_0_9B_pt 160x160 0.9 300 2942.50
37 openpose_pruned_0_3 368x368 49.9 300 57.89
38 person-orientation_pruned_558m_pt 176x80 0.558 300 11906.10
39 personreid-res18_pt 176x80 1.1 300 7096.63
40 personreid-res50_pt 256x128 5.4 300 1637.65
41 plate_detection 320x320 0.49 300 7895.80
42 plate_num 96x288 1.75 300 2312.90
43 pmg_pt 224x224 2.28 300 1995.08
44 pointpainting-nuscenes 40000x64x16 112 300 21.28
pointpainting_nuscenes_40000_64_0_pt
pointpainting_nuscenes_40000_64_1_pt
45 pointpillars_nuscenes 40000x64x5 108 300 42.5002
pointpillars_nuscenes_40000_64_0_pt
pointpillars_nuscenes_40000_64_1_pt
46 refinedet_baseline 480x360 123 300 94.32
47 RefineDet-Medical_EDD_tf 320x320 9.8 300 797.82
48 refinedet_pruned_0_8 360x480 25 300 331.45
49 refinedet_pruned_0_92 360x480 10.1 300 717.55
50 refinedet_pruned_0_96 360x480 5.1 300 1011.15
51 refinedet_VOC_tf 320x320 81.9 300 138.97
52 reid 80x160 0.95 300 7476.71
53 resnet18 224x224 3.7 300 2609.77
54 resnet50 224x224 7.7 300 1178.18
55 resnet50_pt 224x224 4.1 300 1015.14
56 resnet50_tf2 224x224 7.7 300 1178.36
57 resnet_v1_101_tf 224x224 14.4 300 611.32
58 resnet_v1_152_tf 224x224 21.8 300 407.69
59 resnet_v1_50_tf 224x224 7 300 1178.59
60 retinaface 360x640 1.11 350 1764.23
61 salsanext_pt 64x2048 20.4 300 152.34
62 salsanext_v2_pt 64x2048 32 300 58.46
63 SemanticFPN_cityscapes_pt 256x512 10 300 782.05
64 SemanticFPN_Mobilenetv2_pt 512x1024 5.4 350 230.77
65 semantic_seg_citys_tf2 512x1024 54 300 90.59
66 SESR_S_pt 360x640 7.48 300 290.95
67 sp_net 128x224 0.55 300 5669.09
68 squeezenet 227x227 0.76 300 6173.84
69 squeezenet_pt 224x224 0.82 300 6560.42
70 ssd_adas_pruned_0_95 360x480 6.3 300 994.51
71 ssdlite_mobilenet_v2_coco_tf 300x300 1.5 350 2135.17
72 ssd_mobilenet_v1_coco_tf 300x300 2.5 350 2180.88
73 ssd_mobilenet_v2 360x480 6.6 350 720.55
74 ssd_mobilenet_v2_coco_tf 300x300 3.8 350 1472.47
75 ssd_pedestrian_pruned_0_97 360x360 5.9 300 904.75
76 ssd_resnet_50_fpn_coco_tf 640x640 178.4 300 59.37
77 ssd_traffic_pruned_0_9 360x480 11.6 300 672.57
78 tiny_yolov3_vmss 416x416 5.46 300 1634.88
79 tsd_yolox_pt 640x640 73 300 132.59
80 ultrafast_pt 288x800 8.4 300 481.47
81 unet_chaos-CT_pt 512x512 23.3 300 138.36
82 vgg_16_tf 224x224 31 300 295.64
83 vgg_19_tf 224x224 39.3 300 246.94
84 vpgnet_pruned_0_99 480x640 2.5 300 946.99
85 yolov2_voc 448x448 34 300 318.21
86 yolov2_voc_pruned_0_66 448x448 11.6 300 779.56
87 yolov2_voc_pruned_0_71 448x448 9.9 300 906.92
88 yolov2_voc_pruned_0_77 448x448 7.8 300 1091.51
89 yolov3_adas_pruned_0_9 256x512 5.5 300 1229.14
90 yolov3_bdd 288x512 53.7 300 147.58
91 yolov3_voc 416x416 65.4 300 152.95
92 yolov3_voc_tf 416x416 65.6 300 152.84
93 yolov4_leaky_spp_m 416x416 60.1 300 156.64
94 yolov4_leaky_spp_m_pruned_0_36 416x416 38.2 300 166.85