ncnn 小板子

ncnn is a high-performance neural network inference framework.

See https://github.com/Tencent/ncnn for more info about ncnn.

• The benchmark score is the inference time, less is better.
• Runs on all cpu big cores or vulkan, and fp16 arithmetic is enabled if supported.
• Table contents are ordered by mobilenet.
• Rock5B data from https://forum.radxa.com/t/rock5-android12-bringup-status-update/8506/17

small-board	image	spec	squeezenet	mobilenet	shufflenet
NVIDIA Jetson AGX Orin		Ampere 1.3GHz x 2048 Tensor Core x 64	2.13	2.03	2.57
NVIDIA Jetson AGX Orin		A78AE 2.2GHz x 12	3.50	3.49	5.08
ZYSJ RK3588		ARM Mali-G610	7.09	9.16	5.88
ZYSJ RK3588		RK3588 A76 2.4GHz x 4 A55 1.8GHz x 4	7.57	11.01	7.95
Radxa Rock5B		RK3588 A76 2.4GHz x 4 A55 1.8GHz x 4	9.38	11.97	8.02
NVIDIA Jetson Nano		Maxwell 921MHz x 128	12.00	16.03	14.23
Unisoc Tiger T710		Unisoc Tiger T710 A75 1.82GHz x 4 A55 1.82GHz x 4	16.25	20.31	10.59
Khadas VIM3		ARM Mali-G52	19.54	31.62	17.91
Radxa Rock3A		RK3568 A55 2.0GHz x 4	29.52	34.47	23.62
NanoPi M4		ARM Mali-T864	24.57	35.86	33.90
Radxa RockPi X		Intel HD Graphics (Cherry Trail) 500 MHz	29.91	36.41	24.54
NVIDIA Jetson Nano		A57 1.43GHz x 4	30.03	41.74	21.07
Khadas VIM3		A311D A73 2.2GHz x 4 A53 1.8GHz x 2	30.98	42.57	21.62
Odroid XU4		Exynos 5422 A15 2GHz x 4 A7 1.3GHz x 4	41.42	55.91	23.12
Debix Model A		i.MX8M Plus A53 1.6GHz x 4	44.60	59.50	30.34
Tinker Board S R2.0		ARM Mali-T764	41.78	62.67	56.83
Raspberry Pi 4B		BCM2711B0 A72 1.5GHz x 4	51.93	62.97	34.52
NanoPi M4		RK3399 A72 1.8GHz x 2 A53 1.5GHz x 4	43.73	64.28	27.43
Radxa Zero		S905Y2 A53 1.8GHz x 4	51.92	66.12	37.15
NanoPi K2		S905 A53 1.5GHz x 4	56.96	75.25	38.70
Radxa RockPi X		Atom x5-Z8350 x86-64 1.92 GHz x 4	50.22	80.12	37.96
NanoPi R2S		RK3328 A53 1.3GHz x 4	62.20	82.88	52.34
Tinker Board S R2.0		RK3288CGW A17 1.8GHz x 4	56.32	83.56	35.96
EAIDK 310		RK3228H A53 1.3GHz x 4	68.97	90.87	48.22
Radxa Zero		ARM Mali-G31 850 MHz x 2	55.29	98.45	52.24
Raspberry Pi 3B+		BCM2837B0 A53 1.4GHz x 4	93.48	111.72	63.71
Raspberry Pi Zero 2W		BCM2710A1 A53 1.0GHz x 4	119.52	162.60	72.72
EASY EAI Nano		RV1126 A7 1.5GHz x 4	105.16	163.82	68.79
RVB-ICE		T-Head C910 RV64GC 1.2GHz x 2	161.10	274.39	118.61
Loongson Pie Lite		Loongson 2K1000 GS264 1.0GHz x 2	185.28	276.03	124.39
Ingenic X2000		Ingenic X2000 XBurst2 1.2GHz x 2	175.99	282.25	118.4
Banana Pi M2 Zero 2		AllWinner H2+ A7 1.2GHz x 4	230.97	327.65	147.97
Raspberry Pi 4B		Broadcom VideoCore VI	308.47	345.16	143.88
Raspberry Pi 2B		BCM2836 A7 900MHz x 4	221.84	346.58	141.40
PICO-PI-IMX7		i.MX7D A7 1.0GHz x 2	220.10	366.92	137.02
Microphase Z7-Lite		XC7Z020CLG400 A9 766MHz x 2	389.18	623.71	212.20
MangoPI MQ-Pro		D1 C906 RV64GCV 1GHz x 1	385.21	691.47	487.67
NeZha D1		D1 C906 RV64GCV 1GHz x 1	396.18	727.97	495.87
StarFive VisionFive V1		SiFive’s U74 RV64GFC 1.5GHz x 2	609.41	993.87	358.98
Raspberry Pi 1B		ARM1176JZF-S 700MHz x 1	38957.64	85496.75	16750.39