AI Edge Computing Box Based on Rockchip RK3588 + Cambricon | 38TOPS INT8 Computing Power for Intelligent Security, Smart Construction Sites, Smart City Management, and Smart Gas Stations
Edge Computing Box
Rockchip RK3588 + Cambricon | 38TOPS INT8 Computing Power
● Features a Big-Little CPU architecture with quad-core A76 and quad-core A55, CPU frequency up to 2.4GHz, equipped with 1MB L2 Cache and 3MB L3 Cache for enhanced CPU performance.
● High-performance quad-core Mali-G610 GPU, supports at least dual 4K UI displays, capable of smoothly running complex graphics processing tasks.
● Expandable with 4G/Wi-Fi 6/BT wireless communication modules, facilitating edge computing deployment.
● Supports mSATA SSD expansion for large-scale data storage needs.
● Supports 1–2 AI acceleration modules, with up to 32TOPS@INT8 expandable AI computing power, achieving a maximum combined computing power of 38TOPS@INT8, delivering exceptional AI performance.
● Compatible with mainstream deep learning frameworks including TensorFlow, Caffe, PyTorch, TFLite, and ONNX; supports network models such as face detection, tracking, recognition, posture estimation, and hard hat detection.
The XM-RK3588 edge computing box is a high-performance, low-power edge computing device developed by Xinmai Technology based on the RK3588. It integrates an NPU with up to 6.0TOPS@INT8 computing power and offers powerful video encoding and decoding capabilities. It supports up to 32 channels of 1080P@30fps video decoding and 16 channels of 1080P@30fps encoding. It also supports HDMI output of 4K@120fps or 8K@30fps. With a lightweight and flexible design, it is widely applicable in smart transportation, smart parks, smart gas stations, transparent kitchens, and other scenarios, delivering optimized performance for various AI applications.
Item
Type
Model Specifications
Description
C
o
r
e
C
o
n
f
i
g
u
r
a
t
i
o
n
Processor
RK3588
CPU
Quad-core ARM Cortex-A76@2.4GHz
Quad-core ARM Cortex-A55@1.8GHz
GPU
Quad-core ARM Mali-G610 MP4
NPU
6.0TOPS@INT8 computing power
Memory
LPDDR4X
Default Configuration
4 Gbyte
Optional Configuration
8/16 Gbyte
Storage
eMMC
Default Configuration
32 Gbyte
Optional Configuration
64 Gbyte
Two AI module expansion interfaces reserved
(Per Module Specification)
NPU
16TOPS@INT8
Video Decoding
16 x 1080P@30fps H.264/H.265 video decoding
Video Encoding
8 x 1080P@30fps H.264/H.265 video encoding
Image Codec
1080P@800fps JPEG format
GMAC
2x
Supports 10/100/1000Mbps operation modes, supports RGMII mode
RS232
2x
Two 3-pin Phoenix terminals
RS485
2x
Two 3-pin Phoenix terminals
MINI-PCIe
1x
For expansion of mSATA SSD storage devices and 4G wireless communication modules
USB
4x
Two standard USB3.0 ports and two standard USB2.0 ports
DEBUG
1x
System debug serial port
WiFi+BT
1x
(Optional) Wi-Fi 6 + BT: 1x dual-band 2.4GHz/5GHz Wi-Fi 6, Bluetooth 5.0; requires enclosure with external antenna support and Wi-Fi antenna
TF
1x
External TF card slot; Class 10 or higher recommended
LED
3x
Power indicator, HDD indicator, SYS system indicator
SIM Card Slot
1x
For use with 4G/5G wireless communication modules
M.2
1x
For 4G/5G module expansion only
HDMI OUT
1x
Supports 4K@120fps / 8K@30fps output, HDMI 2.1 protocol
AUDIO OUT
1x
Headphone audio output interface
M
e
c
h
a
n
i
c
a
l
P
a
r
a
m
e
t
e
r
s
Cooling Method
Passive cooling
Dimensions
214mm x 165mm x 54mm
Electrical Parameters
Operating Voltage
Operating Current
Power Consumption
Test Ambient Temperature
Test Condition
DC 12V
2A
3.5W
Room temperature
No load
Linux
Bootloader
Version
U-boot v2017.09
Boot Mode
eMMC
Download Mode
TF/Serial port
Kernel
Version
Linux 5.10
Device Drivers
Standard drivers for USB, Ethernet, TF, RS232/RS485, etc.
File System
System Version
Debian 11
System Services
Supports SSH, telnetd, NFS, TFTP, etc.
Development Components
Media Processing
GPU, VPU (video encoding/decoding, image scaling, etc.)
Intelligent Analytics Processing
NPU (supports INT4, INT8, INT16 integer and FP16 floating-point calculations)
AI Development
Supports classification, detection, tracking, recognition, segmentation, and other algorithms
Suitable for applications in facial recognition, intelligent security, smart construction sites, smart city management, smart gas stations, communities, smart factories, transparent kitchens, and more.
| Mechanical Interface
