ZCU102 Performance - 1.1 English

The ZCU102 evaluation board uses the mid-range ZU9 UltraScale+ device. There are two different hardware versions of ZCU102 board, one with the serial number 0432055-04 as the header and the other with the serial number 0432055-05 as the header. The performance of the Vitis AI Library varies between the two hardware versions (because of different DDR performance). Since 0432055-04 version of ZCU102 has been discontinued, the following table only shows the performance of ZCU102 (0432055-05). In ZCU102 board, triple B4096F DPU cores are implemented in program logic.

Refer to the following table for throughput performance (in frames/sec or fps) for various neural network samples on ZCU102 (0432055-05) with DPU running at 281 MHz.

Table 1. ZCU102 (0432055-05) Performance
No	Neural Network	Input Size	GOPS	Performance (fps) (Single thread)	Performance (fps) (Multiple thread)
1	inception_resnet_v2_tf	299x299	26.4	23.3	51
2	inception_v1_tf	224x224	3.0	172.1	455.6
3	inception_v3_tf	299x299	11.5	55.9	136.8
4	inception_v4_2016_09_09_tf	299x299	24.6	28.1	71.4
5	mobilenet_v1_0_25_128_tf	128x128	0.027	848.6	2260.9
6	mobilenet_v1_0_5_160_tf	160x160	0.15	577.2	1913.7
7	mobilenet_v1_1_0_224_tf	224x224	1.1	261.3	788.6
8	mobilenet_v2_1_0_224_tf	224x224	0.60	218.6	598.7
9	mobilenet_v2_1_4_224_tf	224x224	1.2	162.4	412
10	resnet_v1_101_tf	224x224	14.4	43.5	94.5
11	resnet_v1_152_tf	224x224	21.8	29.9	66.1
12	resnet_v1_50_tf	224x224	7.0	78.3	164.7
13	vgg_16_tf	224x224	31.0	19.8	44.7
14	vgg_19_tf	224x224	39.3	17.1	40.3
15	ssd_mobilenet_v1_coco_tf	300x300	2.5	87.2	323.5
16	ssd_mobilenet_v2_coco_tf	300x300	3.8	63	198.7
17	ssd_resnet_50_fpn_coco_tf	640x640	178.4	1.3	5
18	yolov3_voc_tf	416x416	65.6	13.5	37.8
19	mlperf_ssd_resnet34_tf	1200x1200	433	1.8	7.7
20	resnet50	224x224	7.7	72.8	155.5
21	resnet18	224x224	3.7	174.7	461.6
22	inception_v1	224x224	3.2	166.6	444
23	inception_v2	224x224	4.0	136.7	335.6
24	inception_v3	299x299	11.4	56	138.5
25	inception_v4	299x299	24.5	28.1	71.3
26	mobilenet_v2	224x224	0.6	214.7	580.6
27	squeezenet	227x227	0.76	269.4	1045.9
28	ssd_pedestrain_pruned_0_97	360x360	5.9	75.7	294.4
29	ssd_traffic_pruned_0_9	360x480	11.6	54.4	206.7
30	ssd_adas_pruned_0_95	360x480	6.3	82.6	296.3
31	ssd_mobilenet_v2	360x480	6.6	38.4	112.5
32	refinedet_pruned_0_8	360x480	25	31.7	103.8
33	refinedet_pruned_0_92	360x480	10.1	59.8	204
34	refinedet_pruned_0_96	360x480	5.1	82.6	290.1
35	vpgnet_pruned_0_99	480x640	2.5	102.3	397.3
36	fpn	256x512	8.9	59.7	185.2
37	sp_net	128x224	0.55	491.6	1422.8
38	openpose_pruned_0_3	368x368	49.9	3.5	15.3
39	densebox_320_320	320x320	0.49	388.3	1279
40	densebox_640_360	360x640	1.1	195	627.8
41	face_landmark	96x72	0.14	846.7	1379.9
42	reid	80x160	0.95	361.9	672.8
43	multi_task	288x512	14.8	35.4	133
44	yolov3_adas_pruned_0_9	256x512	5.5	83.8	235.3
45	yolov3_voc	416x416	65.4	13.5	38.2
46	yolov3_bdd	288x512	53.7	13	37.1
47	yolov2_voc	448x448	34	24.8	77.1
48	yolov2_voc_pruned_0_66	448x448	11.6	53.2	194.2
49	yolov2_voc_pruned_0_71	448x448	9.9	59.8	224
50	yolov2_voc_pruned_0_77	448x448	7.8	68	266.2