Alveo U55C 高性能计算卡 - 2.5 简体中文

Vitis AI Library 用户指南 (UG1354)
Document ID
UG1354
Release Date
2022-06-15
Version
2.5 简体中文

赛灵思 Alveo™ U55C 高性能计算卡旨在为高性能计算 (HPC)、大数据分析和搜索、金融计算、计算存储和机器学习内的工作负载提供最优化加速。该版本中，DPU 是在面向深度学习推断加速的编程逻辑中实现的。

含 11PE300 MHz DPUCAHX8H-DWC 的 U55C 性能

请参阅下表，查看具有运行频率为 11PE@300 MHz 的 DPUCAHX8H-DWC 的 U55C 上的各种神经网络采样的吞吐量性能（以每秒帧数或 fps 为单位）。

表 1. 含 11PE300 MHz DPUCAHX8H-DWC 的 U55C 性能
编号 神经网络 输入大小 GOPS DPU 频率 (MHz) 性能 (fps)（多线程）
1 bcc_pt 800x1000 269 300 37.4299
2 densebox_320_320 320x320 0.49 300 4900.28
3 densebox_640_360 360x640 1.1 300 2205.27
4 efficientNet-edgetpu-M_tf 240x240 7.34 300 1204.68
5 efficientNet-edgetpu-S_tf 224x224 4.72 300 2000.42
6 ENet_cityscapes_pt 512x1024 8.6 300 147.291
7 face_landmark 96x72 0.14 300 2205.27
8 face-quality 80x60 0.06 300 30651.4
9 face-quality_pt 80x60 0.06 300 30918.6
10 facerec_resnet20 112x96 3.5 300 2424.61
11 facerec-resnet20_mixed_pt 112x96 3.5 300 2423.09
12 facerec_resnet64 112x96 11 300 896.009
13 facereid-large_pt 96x96 0.5 300 15138.2
14 facereid-small_pt 80x80 0.09 300 32819.6
15 fpn 256x512 8.9 300 721.096
16 FPN_Res18_Medical_segmentation 320x320 45.3 300 179.581
17 FPN-resnet18_covid19-seg_pt 352x352 22.7 300 401.981
18 FPN-resnet18_Endov 240x320 13.8 300 652.217
19 hourglass-pe_mpii 256x256 10.2 300 604.947
20 inception_resnet_v2_tf 299x299 26.4 300 298.634
21 inception_v1 224x224 3.2 300 2101.57
22 inception_v1_tf 224x224 3 300 2185.25
23 inception_v2 224x224 4 300 1688.47
24 inception_v2_tf 224x224 3.88 300 672.69
25 inception_v3 299x299 11.4 300 690.495
26 inception_v3_pt 299x299 5.7 300 690.401
27 inception_v3_tf 299x299 11.5 300 691.406
28 inception_v3_tf2 299x299 11.5 300 709.433
29 inception_v4 299x299 24.5 300 324.079
30 inception_v4_2016_09_09_tf 299x299 24.6 300 324.683
31 medical_seg_cell_tf2 128x128 5.3 300 1972.68
32 MLPerf_resnet50_v1.5_tf 224x224 8.19 300 1012.03
33 mlperf_ssd_resnet34_tf 1200x1200 433 300 25.8372
34 mobilenet_1_0_224_tf2 224x224 1.1 300 5305.78
35 mobilenet_edge_0_75_tf 224x224 0.62 300 4716.44
36 mobilenet_edge_1_0_tf 224x224 0.99 300 3834.4
37 mobilenet_v1_0_25_128_tf 128x128 0.03 300 18873.8
38 mobilenet_v1_0_5_160_tf 160x160 0.15 300 12862.5
39 mobilenet_v1_1_0_224_tf 224x224 1.1 300 5305.64
40 mobilenet_v2 224x224 0.6 300 5323.56
41 MT-resnet18_mixed_pt 512x320 13.7 300 511.942
42 multi_task 288x512 14.8 300 542.182
43 multi_task_v3_pt 320x512 25.4 300 291.268
44 openpose_pruned_0_3 368x368 49.9 300 57.9091
45 personreid-res18_pt 176x80 1.1 300 6648.41
46 personreid-res50_pt 256x128 5.4 300 1611.62
47 plate_detection 320x320 0.49 300 7824.46
48 plate_num 96x288 1.75 300 2297.66
49 pmg_pt 224x224 2.28 300 1989.61
50 pointpainting_nuscenes_pt 40000x64x16 112 300 20.1126
51 pointpillars_nuscenes_pt 40000x64x5 108 300 39.9327
52 rcan_pruned_tf 360x640 87 300 80.0572
53 refinedet_baseline 480x360 123 300 93.8894
54 RefineDet-Medical_EDD_tf 320x320 9.8 300 789.952
55 refinedet_pruned_0_8 360x480 25 300 327.715
56 refinedet_pruned_0_92 360x480 10.1 300 707.408
57 refinedet_pruned_0_96 360x480 5.1 300 992.461
58 refinedet_VOC_tf 320x320 81.9 300 138.342
59 reid 80x160 0.95 300 6983.54
60 resnet18 224x224 3.7 300 2603.77
61 resnet50 224x224 7.7 300 1174.56
62 resnet50_pt 224x224 4.1 300 1012.4
63 resnet50_tf2 224x224 7.7 300 1174.21
64 resnet_v1_101_tf 224x224 14.4 300 610.351
65 resnet_v1_152_tf 224x224 21.8 300 407.059
66 resnet_v1_50_tf 224x224 7 300 1175.61
67 retinaface 360x640 1.11 300 1732.13
68 salsanext_pt 64x2048 20.4 300 153.159
69 salsanext_v2_pt 64x2048 32 300 58.4121
70 SemanticFPN_cityscapes_pt 256x512 10 300 803.504
71 SemanticFPN_Mobilenetv2_pt 512x1024 5.4 300 228.29
72 semantic_seg_citys_tf2 512x1024 54 300 90.0321
73 sp_net 128x224 0.55 300 5639.48
74 squeezenet 227x227 0.76 300 6171.87
75 squeezenet_pt 224x224 0.82 350 6398.41
76 ssd_adas_pruned_0_95 360x480 6.3 300 981.077
77 ssd_inception_v2_coco_tf 300x300 9.6 300 329.922
78 ssdlite_mobilenet_v2_coco_tf 300x300 1.5 300 2061.95
79 ssd_mobilenet_v1_coco_tf 300x300 2.5 300 2115.8
80 ssd_mobilenet_v2 360x480 6.6 300 717.919
81 ssd_mobilenet_v2_coco_tf 300x300 3.8 300 1458.46
82 ssd_pedestrian_pruned_0_97 360x640 5.9 300 890.03
83 ssd_resnet_50_fpn_coco_tf 640x640 178 300 58.9891
84 ssd_traffic_pruned_0_9 360x480 11.6 300 666.004
85 tiny_yolov3_vmss 416x416 5.46 300 1631.58
86 unet_chaos-CT_pt 512x512 23.3 300 136.798
87 vgg_16_tf 224x224 31 300 283.803
88 vgg_19_tf 224x224 39.3 300 238.409
89 vpgnet_pruned_0_99 480x640 2.5 300 934.319
90 yolov2_voc 448x448 34 300 317.894
91 yolov2_voc_pruned_0_66 448x448 11.6 300 779.019
92 yolov2_voc_pruned_0_71 448x448 9.9 300 905.845
93 yolov2_voc_pruned_0_77 448x448 7.8 300 1091.31
94 yolov3_adas_pruned_0_9 256x512 5.5 300 1223.77
95 yolov3_bdd 288x512 53.7 300 147.071
96 yolov3_voc 416x416 65.4 300 152.429
97 yolov3_voc_tf 416x416 65.6 300 152.75
98 yolov4_leaky_spp_m 416x416 60.1 300 155.946
99 yolov4_leaky_spp_m_pruned_0_36 416x416 38.2 300 166.494
100 ultrafast_pt 288x800 8.4 300 474.242
101 drunet_pt 528x608 2.59 300 469.205
102 person-orientation_pruned_558m_pt 224x112 0.56 300 11379.5
103 ofa_resnet50_0_9B_pt 160x160 0.9 300 2922
104 SESR_S_pt 360x640 7.48 300 282.558
105 c2d2_lite 512x512 6.86 300 28.2529
106 ofa_depthwise_res50_pt 176x176 1.25 300 2887.1
107 FairMot_pt 640x480 36 300 239.285
108 tsd_yolox_pt 640x640 73 300 131.889
109 ssr_pt 256x256 39.7 300 62.0193
110 chen_color_resnet18_pt 224x224 3.63 300 2645.11
111 face_mask_detection_pt 512x512 0.59 300 1581.35
112 ofa_rcan_latency_pt 360x640 45.7 300 65.5305
113 vehicle_make_resnet18_pt 224x224 3.63 300 2634.76
114 vehicle_type_resnet18_pt 224x224 3.63 300 2642.15
115 ofa_yolo_pt 640x640 48.9 300 178.7
116 ofa_yolo_pruned_0_30_pt 640x640 34.7 300 223.187
117 ofa_yolo_pruned_0_50_pt 640x640 24.6 300 283.17
118 yolov3-coco_tf2 416x416 65.9 300 150.588