VCK5000 Evaluation Kit

The Versal ACAP VCK5000 is a Versal AI Core series evaluation kit that enables designers to develop solutions using AI and DSP engines capable of delivering over 100X greater compute performance compared to current server class CPUs. For this release, DPU core with batch=8 is implemented using AI Engines.

VCK5000 Performance with 8PE350 MHz DPUCVDX8H

The following table lists the throughput performance (in frames/sec or fps) for various neural network samples on the Versal ACAP VCK5000 Gen3x16 with DPUCVDX8H running at 8PE@350 MHz.

Table 1. VCK5000 Performance with 8PE350 MHz DPUCVDX8H
No	Neural Network	Input Size	GOPS	DPU Frequency (MHz)	Performance (fps) (Multiple thread)
1	densebox_320_320	320x320	0.49	350	5902.4
2	densebox_640_360	360x640	1.1	350	2802.39
3	ENet_cityscapes_pt	512x1024	8.6	350	140.057
4	face_landmark	96x72	0.14	350	14111.7
5	face-quality_pt	80x60	0.06	350	41833.6
6	fpn	256x512	8.9	350	1074.4
7	FPN_Res18_Medical_segmentation	320x320	45.3	350	535.843
8	FPN-resnet18_covid19-seg_pt	352x352	22.7	350	1071.96
9	inception_v1	224x224	3.2	350	4105.35
10	inception_v1_tf	224x224	3	350	4362.47
11	medical_seg_cell_tf2	128x128	5.3	350	1955.31
12	MLPerf_resnet50_v1.5_tf	224x224	8.19	350	4425.9
13	multi_task	288x512	14.8	350	694.169
14	openpose_pruned_0_3	368x368	49.9	350	168.812
15	plate_detection	320x320	0.49	350	7812.45
16	refinedet_baseline	480x360	123	350	291.851
17	RefineDet-Medical_EDD_tf	320x320	9.8	350	1287.96
18	refinedet_pruned_0_8	360x480	25	350	654.851
19	refinedet_pruned_0_92	360x480	10.1	350	844.655
20	refinedet_pruned_0_96	360x480	5.1	350	888.547
21	refinedet_VOC_tf	320x320	81.9	350	378.612
22	reid	80x160	0.95	350	10467.2
23	resnet18	224x224	3.7	350	6434.83
24	resnet50	224x224	7.7	350	4516.51
25	resnet50_pt	224x224	4.1	350	4453.62
26	resnet50_tf2	224x224	7.7	350	4515.15
27	resnet_v1_101_tf	224x224	14.4	350	2938.99
28	resnet_v1_152_tf	224x224	21.8	350	2095.6
29	resnet_v1_50_tf	224x224	7	350	4846.78
30	salsanext_pt	64x2048	20.4	350	171.977
31	SemanticFPN_cityscapes_pt	256x512	10	350	1080.18
32	sp_net	128x224	0.55	350	7654.76
33	squeezenet	227x227	0.76	350	8256.32
34	squeezenet_pt	224x224	0.82	350	5147.72
35	ssd_adas_pruned_0_95	360x480	6.3	350	930.74
36	ssd_traffic_pruned_0_9	360x480	11.6	350	894.817
37	tiny_yolov3_vmss	416x416	5.46	350	2612.41
38	unet_chaos-CT_pt	512x512	23.3	350	229.231
39	vpgnet_pruned_0_99	480x640	2.5	350	600.461
40	yolov2_voc	448x448	34	350	949.278
41	yolov2_voc_pruned_0_66	448x448	11.6	350	1574.72
42	yolov2_voc_pruned_0_71	448x448	9.9	350	1727.27
43	yolov2_voc_pruned_0_77	448x448	7.8	350	1698.6
44	yolov3_adas_pruned_0_9	256x512	5.5	350	1379.55
45	yolov3_bdd	288x512	53.7	350	383.884
46	yolov3_voc	416x416	65.4	350	459.207
47	yolov3_voc_tf	416x416	65.6	350	459.307