Six Core Features
Flagship computing module based on RK3588S — retaining core computing power, optimized form factor and cost
⚡
8-Core 64-bit CPU
4× Cortex-A76 @ 2.4GHz + 4× Cortex-A55 @ 1.8GHz, 8nm advanced process, strong multi-threaded performance
8nm | 8-Core | 2.4GHz
💾
Up to 32GB Memory
LPDDR4/4X high-speed memory, optional 2GB / 4GB / 8GB / 16GB / 32GB; onboard 32/64/128/256GB eMMC storage, meeting big data scenarios
LPDDR4/4X | 32GB | eMMC 256GB
🤖
6 TOPS NPU
Built-in AI accelerator, supports INT4/INT8/INT16 mixed computing, easily deploy mainstream models such as YOLO and ResNet
INT4/INT8/INT16 | TensorFlow | PyTorch | ONNX
💧
Compact B2B Compute Module
3×100pin B2B connector (DF40C-100DP-04V), 55×40×1.6mm compact size, easily embedded into industrial equipment, smart terminals and custom products
55×40×1.6mm | 3×100pin B2B
🎮
Mali-G610 MP4 GPU
Quad-core GPU supports OpenGL ES 3.2 / OpenCL 2.2 / Vulkan 1.2, 450 GFLOPS graphics computing power, smooth 4K UI and 3D rendering
450 GFLOPS | Vulkan 1.2
🎥
8K Full-Pipeline Video
8K@60fps decoding (H.265/VP9/AVS2), 8K@30fps encoding (H.264/H.265), supports multi-channel 1080P concurrent processing
Decode 8K@60 | Encode 8K@30
Product Showcase
55×40mm compact core board — front side RK3588S SoC + LPDDR + eMMC, back side 3×100pin B2B connector
CM7s Front
RK3588S SoC (metal heatsink cover) | LPDDR4X memory | eMMC storage | RK806 PMIC
CM7s Back
3×100pin B2B connector (DF40C-100DP-04V) | Power management | RC array
RK3588S — Flagship Chip, Compact Body
RK3588S is a miniaturized and cost-optimized variant of the RK3588 family. Core computing performance is identical, with a smaller package, better suited for compact modules
RK3588S vs RK3588 Differences at a Glance
RK3588S retains the core computing units (CPU/GPU/NPU/codec), achieving a smaller package and lower cost through simplified peripheral interfaces. For applications that do not need multi-lane PCIe 3.0, HDMI RX, or dual Type-C, RK3588S is the better choice.
RK3588S
8nm Flagship AIoT SoC | 17×17mm Package
CPU
A76×4 + A55×4
GPU
Mali-G610 MP4
Process
8nm
Max Frequency
2.4 GHz
NPU
6 TOPS
Package
17×17mm
Spec | RK3588 | RK3588S (CM7s) |
CPU | A76×4 + A55×4 | A76×4 + A55×4 |
Max Frequency | 2.4GHz | 2.4GHz |
GPU | Mali-G610 MP4 | Mali-G610 MP4 |
NPU | 6 TOPS | 6 TOPS |
8K Decode | 8K@60fps | 8K@60fps |
8K Encode | 8K@30fps | 8K@30fps |
Package Size | 23×23mm | 17×17mm ▼ 45% |
PCIe 3.0 | 2×2-lane | None |
PCIe 2.0 | 3×1-lane | 2×1-lane |
SATA 3.0 | 3 channels | 2 channels |
HDMI TX / eDP | 2 channels | 1 channel |
HDMI RX | 1 channel | None |
DP Output | 2×4-lane | 1×4-lane |
MIPI CSI | 2×4-lane | 1×4-lane |
Full-featured Type-C | 2 ports | 1 port |
GMAC Ethernet | 2 channels | 1 channel |
ADC | 8 channels | 4 channels |
● Core computing performance is identical
● Peripheral interfaces are simplified
Performance Comparison
CM7s fully leads computing modules in the same price range in AI computing power, CPU performance and video codec
AI Computing Power Comparison
Product | AI Acceleration Unit | AI Computing Power | Description |
ArmSoM CM7s | Built-in NPU | 6 TOPS | Supports INT4/INT8/INT16 mixed computing |
Raspberry Pi CM5 | No built-in NPU | — | CPU/GPU inference |
Raspberry Pi CM4 | No built-in NPU | — | CPU/GPU inference |
Jetson Nano | No built-in NPU, relies on GPU | ~0.5 – 0.512 TFLOPS | 128-core Maxwell GPU inference |
CPU Multi-core Performance Comparison (Relative)
ArmSoM CM7s
A76×4 + A55×4
Raspberry Pi CM5
A76×4
Raspberry Pi CM4
A72×4
Jetson Nano
A57×4
Video Decoding Comparison
ArmSoM CM7s
8K@60fps + 32×1080P@60fps
Raspberry Pi CM5
4K@60fps
Raspberry Pi CM4
4K@30fps
Jetson Nano
4K@30fps
AI Edge Inference
6 TOPS NPU supports mainstream deep learning frameworks — from prototype to deployment, one solution does it all
Built-in NPU supports INT4 / INT8 / INT16 mixed-precision computing, with computing power up to 6 TOPS
👓
Machine Vision
Face detection, object recognition, behavior analysis, industrial quality inspection
🤖
Robotics
SLAM navigation, path planning, sensor fusion, edge decision-making
🎧
Smart Voice
Voice wake-up, voiceprint recognition, noise reduction, speech synthesis
📷
Smart Security
Intrusion detection, license plate recognition, people counting, video structuring
Multi-Display Output
RK3588S supports driving multiple independent displays simultaneously, meeting digital signage, HMI, multi-screen workstation and other scenarios
8K@60fps
HDMI 2.1 TX
7680×4320
4K@60fps
eDP 1.3 TX
3840×2160
8K@30fps
DP 1.4 (Type-C)
DisplayPort Alt Mode
4K@60fps
MIPI DSI (4-lane)
内嵌显示屏接口
Supports simultaneous output of multiple independent displays, suitable for digital signage, HMI interfaces, multi-screen workstations and other scenarios
Rich Interfaces, Infinite Expansion
All high-speed interfaces are routed out via the 3×100pin B2B connector (DF40C-100DP-04V), meeting various carrier board design needs
🖧
USB
1×USB 3.0 + 2×USB 2.0 Host
🖧
USB OTG
1×USB 3.0 OTG / Type-C
📡
PCIe 2.0
2×1-lane PCIe 2.0
💾
SATA 3.0
2 SATA interfaces
🌐
Ethernet
1×Gigabit GMAC + RGMII
📷
MIPI CSI
1×4-lane or 2×2-lane
⚙
PWM / CAN
16×PWM + 3×CAN
📺
MIPI DSI
1×4-lane display output
📥
SDIO 3.0
SDIO 3.0 interface (WiFi/BT module)
🔈
Audio
I2S / PDM / SPDIF TX
💾
SDMMC
SDMMC interface (TF card boot)
⚙
UART
Up to 10 UART channels
⚙
ADC
4 SARADC channels
⚙
SPI / I2C
5×SPI + 8×I2C
Use Cases
From edge AI to industrial control, from digital signage to smart security, CM7s empowers every industry
🤖
Edge AI & Machine Learning
Deploy AI models locally for low-latency inference and data privacy protection
🔌
IoT Gateway
Multi-protocol communication + edge computing, connecting everything intelligently
📺
Digital Signage
8K video playback + multi-screen output, driving large commercial display systems
🛠
Industrial HMI
Touchscreen HMI, RS485/CAN bus, on-site industrial control
🎥
Media Center
8K@60 decoding capability, creating a home-theater-grade playback experience
📷
Smart Security NVR
Multi-channel video input + AI analysis, smart recording and alerts
💻
Development & Prototyping
Quick validation with IO carrier board, seamless transition from prototype to mass production
🧬
Robotics
SLAM navigation, sensor fusion, real-time decision and control
Technical Specs
Complete hardware parameters for carrier board design and product integration
Item | Specification |
|---|---|
SoC | Rockchip RK3588S, 8nm LP process |
CPU | 4× Cortex-A76 @ 2.4GHz + 4× Cortex-A55 @ 1.8GHz |
GPU | ARM Mali-G610 MP4, supports OpenGL ES 3.2 / OpenCL 2.2 / Vulkan 1.2, 450 GFLOPS |
NPU | 6 TOPS@INT8, supports INT4/INT8/INT16 mixed computing |
Memory | LPDDR4/4X, optional 2GB / 4GB / 8GB / 16GB / 32GB |
Storage | Onboard eMMC 5.1, optional 32GB / 64GB / 128GB / 256GB; supports SDMMC/TF card boot |
Video Decode | 8K@60fps H.265/VP9/AVS2; 8K@30fps H.264; 4K@60fps AV1; 1080P@60fps MPEG-2/VC-1/VP8 |
Video Encode | 8K@30fps H.264/H.265; supports up to 16 channels of 1080P@30fps simultaneous encoding |
ISP | Integrated 48MP ISP, supports HDR & 3DNR |
Display Output | 1× HDMI 2.1 (8K@60); 1× eDP 1.3 (4K@60); 1× DP 1.4 via Type-C (8K@30); 1× MIPI DSI (4K@60) |
Camera | 1× 4-lane MIPI CSI or 2× 2-lane MIPI CSI (MIPI DPHY) |
USB | 1× USB 3.0 Host + 1× USB 3.0 OTG (Type-C) + 2× USB 2.0 Host |
PCIe | 2× PCIe 2.0 1-lane (shared with SATA/USB3) |
SATA | 2× SATA 3.0 (shared with PCIe/USB3) |
Ethernet | 1× Gigabit Ethernet (GMAC/RGMII) |
SDIO | SDIO 3.0 interface, expandable for WiFi/BT module |
Audio | I2S ×2 / PDM ×2 / SPDIF TX ×2 |
Serial | UART ×10 / SPI ×5 / I2C ×8 |
Other Interfaces | PWM ×16 / CAN ×3 / SARADC ×4 / GPIO |
Connector | 3×100pin B2B connector (Hirose DF40C-100DP-04V) |
PCB Size | 55 mm × 40 mm × 1.6 mm (L × W × H) |
Operating Temperature | 0°C ~ 70°C (Commercial grade); -25°C ~ 85°C (Industrial grade optional) |
Power Input | DC 5V (MAX 2500mA) |
Power Output | DC 3.3V (MAX 600mA) / DC 1.8V (MAX 600mA) |
Operating System | Android 12 / Debian 12 / Ubuntu 22.04/24.04 / Armbian / Buildroot / Kylin OS |
Compatible Ecosystem
CM7s uses a 3×100pin B2B connector, paired with official carrier boards and ecosystem carrier boards to lower the development barrier
🖥
ArmSoM CM7s IO Board
Official carrier board, routes out all interfaces for quick feature validation
💻
B2B Standard Carrier Board
Compatible with DF40C-100DP-04V connector, supports custom carrier board design
📐
Banana Pi Ecosystem
Compatible with BPI series carrier boards and expansion modules
🔧
OEM/ODM Customization
Reference designs and schematics provided, supports deep customization
📦
Volume Supply
Supports bulk purchase of multiple configurations from 2GB~32GB, industrial grade optional
Design Resources
Open hardware with complete design materials — from schematics to SDK, accelerating product development
FAQ
Q
What is the difference between CM7s and CM5?
CM7s is based on RK3588S, while ArmSoM CM5 is based on RK3576. RK3588S is comprehensively superior to RK3576 in CPU (A76 architecture), GPU (Mali-G610 MP4), and video codec capability (8K@60), making it suitable for scenarios with higher computing power requirements.
Q
Is RK3588S slower than RK3588?
Core computing performance (CPU/GPU/NPU/codec) is basically identical. The difference of RK3588S lies in simplified peripheral interfaces: no PCIe 3.0, one fewer HDMI/DP/SATA channel, no HDMI RX. For applications that do not need these interfaces, RK3588S offers the same computing power with a smaller package (17×17mm vs 23×23mm) and lower cost.
Q
Can CM7s directly replace a Raspberry Pi CM4?
CM7s uses a 3×100pin B2B connector (DF40C-100DP-04V), which has a different form factor from the Raspberry Pi CM4's SO-DIMM interface, so it cannot be directly inserted into a CM4 carrier board. However, ArmSoM provides an official carrier board that routes out all interfaces, enabling quick product validation and development. If you need a CM4-compatible solution, we recommend looking into the ArmSoM CM5 series.
Q
What operating systems are supported?
Official support for Android 12, Debian 12, Ubuntu 22.04/24.04, and Buildroot. Community support for Armbian, Arch Linux, openSUSE, Kylin OS, and more.
Q
Is an industrial-grade version available?
Yes, CM7s offers an industrial-grade version with an operating temperature range of -25°C ~ 85°C, suitable for harsh industrial environments. For bulk purchases, contact sales@armsom.org for customization.
Q
How do I deploy AI models on the NPU?
Use the Rockchip RKNN Toolkit to convert models from TensorFlow, PyTorch, ONNX and other frameworks to RKNN format, then run them on the NPU. ArmSoM provides complete model conversion tutorials and example code.
Ready to integrate CM7s into your product?
From sample procurement to volume customization, ArmSoM provides full-chain support
Or email sales@armsom.org