ArmSoM
A Powerful Replacement for Raspberry Pi CM4

ArmSoM CM5

More Performance · Better Value · Stronger Graphics

Powered by the Rockchip RK3576 (8nm AIoT platform): 8-core CPU, 6 TOPS NPU, up to 16GB LPDDR5, up to 4K@120fps video decoding and 4K@60fps video encoding, onboard WiFi 6 + BT 5.3, and a 55×40mm form factor compatible with Raspberry Pi CM4 carriers.

8 Core
A72 @2.2GHz · A53 @1.8GHz
6 TOPS
NPU AI Performance
4K@120
4K@120 decode / 4K@60 encode
16 GB
Max LPDDR5 Memory
ArmSoM CM5 compute module front side (RK3576 + LPDDR5 + eMMC + WiFi 6)

Six Core Features

Built on the Rockchip RK3576 second-generation 8nm AIoT platform and compatible with Raspberry Pi CM4 carriers, designed for embedded productization

⚡

8-Core 64-bit CPU

4× Cortex-A72 @ 2.2GHz + 4× Cortex-A53 @ 1.8GHz on an advanced 8nm process, with dedicated NEON coprocessors for strong multithreaded performance

8nm | 8-Core | A72@2.2GHz · A53@1.8GHz
🤖

6 TOPS NPU

A built-in AI accelerator supporting INT4/INT8/INT16/FP16/BF16/TF32 mixed-precision operations, ready to deploy mainstream models such as YOLO and ResNet

6 TOPS@INT8 | TensorFlow | PyTorch | ONNX
🎮

Mali-G52 MC3 GPU

Compatible with OpenGL ES 1.1/2.0/3.2, OpenCL 2.0 and Vulkan 1.1, plus a dedicated 2D hardware engine with MMU for smooth display performance

OpenGL ES 3.2 | OpenCL 2.0 | Vulkan 1.1
🎥

4K@120 Video Pipeline

Decoding supports H.264/H.265/VP9/AV1/AVS2 up to 8K@30fps or 4K@120fps (JPEG 4K@60fps); encoding supports H.264/H.265 and JPEG up to 4K@60fps

Decode 8K@30 / 4K@120 | Encode 4K@60
💾

Up to 16GB Memory

32-bit LPDDR5 high-speed memory available in 2GB / 4GB / 8GB / 16GB (8GB default); onboard 32/64/128GB eMMC (64GB default), with MicroSD and M.2 M-KEY NVMe SSD expansion

LPDDR5 16GB | eMMC 128GB | NVMe SSD

We plan to introduce CM5 v2.0 with LPDDR4 memory to offer more flexibility as memory prices change.

💧

CM4-Compatible B2B Module

2×100-pin board-to-board (B2B) connectors in the same specification as Raspberry Pi CM4, so it plugs directly into the official Raspberry Pi CM4 IO Board; the compact 55×40mm size fits easily into industrial equipment and custom products

55×40mm | 2×100pin B2B | CM4 Carrier Compatible

Product Gallery

A compact 55×40mm compute module integrating RK3576 SoC + LPDDR5 + eMMC + WiFi 6, ready for CM4 carriers or the official CM5-IO board

CM5 front side
CM5 Front
RK3576 SoC | LPDDR5 | eMMC | WiFi 6
CM5 back side
CM5 Back
2×100-pin B2B | Power Management | Passives
CM5 with IO carrier board
Module + CM5-IO Carrier
Pair with the official carrier for fast evaluation
CM5 hardware interfaces
Hardware Interfaces
USB 3.0 | PCIe 2.0 | HDMI 2.1 | DP 1.4 | MIPI

RK3576 — Second-Generation 8nm AIoT Platform

RK3576 combines an 8-core CPU, Mali-G52 MC3 GPU and 6 TOPS NPU, bringing RK3588-family architecture at a better price-performance ratio

RK3576
Second-generation 8nm high-performance AIoT SoC
CPU
A72×4 + A53×4
Max Frequency
2.2 GHz
GPU
Mali-G52 MC3
NPU
6 TOPS
Process
8nm
Decoding
8K@30 / 4K@120
Encoding
4K@60

ArmSoM CM5 vs Raspberry Pi CM4 at a Glance

CM5 and Raspberry Pi CM4 share the same 55×40mm footprint and board-to-board connector specification, so CM5 can use Raspberry Pi CM4 carriers directly; at the same time it upgrades the CPU architecture, NPU performance, memory capacity, video codec and high-speed expansion.

SpecificationRaspberry Pi CM4ArmSoM CM5
SoCBroadcom BCM2711Rockchip RK3576
CPUA72×4A72×4 + A53×4
GPUVideoCore VIMali-G52 MC3
NPUNone6 TOPS
Memory1/2/4/8GB LPDDR42/4/8/16GB LPDDR5
StorageOptional 8/16/32GB eMMC (Lite has no eMMC)32/64/128GB eMMC
Video DecodingH.265 up to 4Kp60H.265/AV1 8K@30 / 4K@120
Display Output2× HDMI 2.01× HDMI 2.1 + 1× DP 1.4
Camera1× 4-lane + 1× 2-lane CSI1× 4-lane + 1× 2-lane CSI
USB1× USB 2.01× USB 3.0 + 1× USB 2.0
Expansion1× PCIe 2.02× PCIe 2.0 / USB 3.0
Ethernet1× Gigabit1× Gigabit
GPIO40-PIN40-PIN
Dimensions55×40mm55×40mm
Operating SystemRaspberry Pi OS / Ubuntu, etc.Debian 12 / Android 14 / Ubuntu / Armbian
Availability—At least through August 2034

● Same as CM4, CM4 carrier reusable ● CM5 specification upgraded or differs

Performance Comparison

CM5 leads same-priced compute modules in AI performance, CPU performance and video codec

AI Performance Comparison
ProductAI AcceleratorAI PerformanceNotes
ArmSoM CM5Built-in NPU6 TOPSSupports INT4/INT8/INT16 mixed operations
Raspberry Pi CM5No built-in NPU—CPU/GPU inference
Raspberry Pi CM4No built-in NPU—CPU/GPU inference
CPU Multi-Core Performance (relative)
ArmSoM CM5
A72×4 + A53×4
Raspberry Pi CM5
A76×4
Raspberry Pi CM4
A72×4
Video Decoding Capability Comparison
ArmSoM CM5
H.265/AV1 8K@30 / 4K@120
Raspberry Pi CM5
H.265 4Kp60
Raspberry Pi CM4
H.265 4Kp60

AI Edge Inference

The 6 TOPS NPU supports mainstream deep learning frameworks — one solution from prototype to deployment

The built-in NPU supports INT4 / INT8 / INT16 mixed-precision operations with up to 6 TOPS

TensorFlow
PyTorch
ONNX
Caffe
TFLite
MXNet
Android NN
RKNN Toolkit
👓
Machine Vision
Face detection, object recognition, behavior analysis, industrial quality inspection
🎧
Intelligent Voice
Voice wake-up, speaker recognition, noise reduction, speech synthesis
📷
Smart Security
Intrusion detection, license plate recognition, people counting, video structuring
🤖
Robotics
SLAM navigation, path planning, sensor fusion, edge decision-making

Multi-Display Output

RK3576 supports independent HDMI, DP and MIPI DSI outputs for digital signage, HMI, multi-screen workstations and more

4K@120fps
HDMI 2.1 TX
3840×2160
4K@120fps
DP 1.4 Output
DisplayPort
2K@60fps
MIPI DSI (4-lane)
Embedded Display Interface

1× HDMI 2.1, 1× DP 1.4 and 1× MIPI DSI can drive multiple independent displays simultaneously, ideal for digital signage, HMI and multi-screen workstations

Rich Interfaces, Unlimited Expansion

All high-speed interfaces are routed through 2×100-pin board-to-board (B2B) connectors — the same specification as Raspberry Pi CM4, so existing CM4 carrier designs can be reused

🖧
USB
1× USB 3.0 + 1× USB 2.0
🖧
HDMI / DP
1× HDMI 2.1 + 1× DP 1.4 (4K@120)
📡
PCIe 2.0
2× PCIe 2.0 1-lane
💾
M.2 M-KEY
PCIe 2.1 NVMe SSD expansion
🌐
Ethernet
1× Gigabit GMAC (RGMII/RMII)
📶
WiFi 6 / BT 5.3
Onboard WiFi 6 + Bluetooth 5.3
📷
MIPI CSI
1× 4-lane + 1× 2-lane camera
📺
MIPI DSI
1× 4-lane (up to 2K@60fps)
🔈
Audio
SPDIF_Tx ×1 / AUDIO_R/L ×1 / i2s ×1
💾
Storage
Onboard eMMC + MicroSD (TF card boot)
⚙
UART DEBUG
Debug UART ×1
⚙
FAN / RTC
Fan connector ×1 / RTC ×1
🔌
Power
4.5V ~ 5.5V DC input
🧩
GPIO
40-PIN (routed via carrier board)
ArmSoM CM5 and CM5-IO hardware block diagram

Use Cases

From edge AI to industrial control, from digital signage to smart security, CM5 powers every industry

🤖
Edge AI & Machine Learning
Deploy AI models locally for low-latency inference and data privacy
📺
Digital Signage
4K@120 video playback with multi-screen output for large commercial displays
🎥
Media Center
4K@120 decoding for a home-theater-grade playback experience
💻
Development & Prototyping
Validate quickly with the IO carrier, from prototype to mass production
🔌
IoT Gateway
Multi-protocol connectivity plus edge computing to connect everything
🛠
Industrial HMI
Touch-panel human-machine interface with RS485/CAN for shop-floor control
📷
Smart Security NVR
Multi-channel video input with AI analysis, intelligent recording and alerts
🧬
Robotics
SLAM navigation, sensor fusion, real-time decision-making and control

Technical Specifications

Complete hardware specifications for carrier board design and product integration

ArmSoM CM5 specification overview: front component callouts and 55×40mm back dimensions
ItemSpecification
SoCRockchip RK3576, second-generation 8nm AIoT platform
CPU4× Cortex-A72 @ 2.2GHz + 4× Cortex-A53 @ 1.8GHz, with dedicated NEON coprocessors
GPUARM Mali-G52 MC3, compatible with OpenGL ES 1.1/2.0/3.2, OpenCL 2.0 and Vulkan 1.1, dedicated 2D hardware engine
NPU6 TOPS@INT8, supports INT4/INT8/INT16/FP16/BF16/TF32 mixed precision; supports TensorFlow / Caffe / TFLite / PyTorch / ONNX / Android NN
Memory32-bit LPDDR5, options 2GB / 4GB / 8GB / 16GB (8GB default)
Version NoteThe current CM5 uses LPDDR5. We plan to introduce CM5 v2.0 with LPDDR4 memory to offer more flexibility as memory prices change.
StorageOnboard eMMC 32GB / 64GB / 128GB (64GB default); MicroSD expansion supported; M.2 M-KEY supports PCIe 2.1 NVMe SSD
Video DecodingH.264 / H.265 / VP9 / AV1 / AVS2 up to 8K@30fps or 4K@120fps; JPEG up to 4K@60fps
Video EncodingH.264 / H.265 and JPEG up to 4K@60fps
ISP16MP ISP
Display Output1× HDMI OUT 2.1 (4K@120fps); 1× DP 1.4 (4K@120fps); 1× MIPI DSI (up to 2K@60fps)
Camera1× 4-lane MIPI CSI + 1× 2-lane MIPI CSI (up to 2.5Gbps per lane)
Network1× Gigabit GMAC (RGMII/RMII); onboard IEEE 802.11a/b/g/n/ac/ax WiFi 6 + BT 5.3
USB1× USB 3.0 + 1× USB 2.0
PCIe2× PCIe 2.0 1-lane
AudioSPDIF_Tx ×1; AUDIO_R/L ×1; i2s ×1
Other InterfacesUART DEBUG ×1 / FAN ×1 / RTC ×1
GPIO40-PIN (routed via carrier board)
Connector2×100-pin board-to-board (B2B) connectors, same specification as Raspberry Pi CM4, compatible with CM4 carriers
Dimensions / WeightCM5: 55 mm × 40 mm × 4.9 mm, approx. 12g; CM5-IO carrier: 100 × 80 × 29 mm
Power Input4.5V ~ 5.5V
Operating Temperature0°C ~ 80°C
Operating SystemDebian 12 / Android 14 / Ubuntu 22.04 / Armbian / ubuntu-rockchip
AvailabilityAt least through August 2034

Compatible Ecosystem

Same footprint and connectors as Raspberry Pi CM4, so CM4 carriers work out of the box, and the official CM5-IO carrier lowers the barrier to development

🍓
Raspberry Pi CM4 Official IO Board
Same size and connectors — plug in directly and reuse the mature CM4 carrier ecosystem
🖥
ArmSoM CM5-IO Carrier
Official companion carrier that exposes all interfaces for fast functional validation
💻
2×100-pin B2B Carrier Design
Same connector specification as CM4, with reference designs and custom carrier support
🔧
OEM/ODM Customization
Reference designs and schematics available, with support for deep customization
📦
Volume Supply
Bulk purchasing across 2GB~16GB configurations, available at least through 2034
ArmSoM CM5 installed directly on the official Raspberry Pi CM4 IO Board

Frequently Asked Questions

What is the difference between CM5 and CM7s?
CM5 is based on the Rockchip RK3576 (4× Cortex-A72 + 4× Cortex-A53, Mali-G52 MC3 GPU), while CM7s is based on the RK3588S (4× Cortex-A76 + 4× Cortex-A55, Mali-G610 MP4 GPU). CM7s offers higher performance and up to 32GB of memory for compute-intensive workloads; CM5 is more cost-effective, uses the 55×40mm form factor and is compatible with Raspberry Pi CM4 carriers, making it better suited to embedded productization.
Can CM5 use Raspberry Pi CM4 carrier boards directly?
Yes. CM5 uses the same 55×40mm dimensions, 2×100-pin board-to-board connectors and 40-PIN GPIO definition as the Raspberry Pi CM4, so it plugs directly into the official Raspberry Pi CM4 IO Board. Note that, limited by the RK3576 pinout, CM5 provides 1× HDMI + 1× DP + 1× DSI, while CM4 provides 2× HDMI + 2× DSI (1×4-lane + 1×2-lane); CM5 also adds PCIe and USB 3.0.
What upgrades does CM5 offer over Raspberry Pi CM4?
The CPU is upgraded from a 4-core Cortex-A72 to 4× Cortex-A72 + 4× Cortex-A53; a 6 TOPS NPU is added; memory is upgraded to up to 16GB 32-bit LPDDR5; 32/64/128GB eMMC is onboard; video decoding is upgraded from H.265 4Kp60 to 8K@30 / 4K@120 (H.265/AV1); and USB 3.0 plus a second PCIe 2.0 lane are added.
Which operating systems are supported?
Officially supported: Debian 12, Android 14 and Ubuntu 22.04. Community support includes Armbian and ubuntu-rockchip, with prebuilt images and flashing tutorials available.
How can I choose the memory and storage configuration?
Memory options are 2GB / 4GB / 8GB / 16GB 32-bit LPDDR5 (8GB default); storage is onboard eMMC at 32GB / 64GB / 128GB (64GB default), with MicroSD boot and M.2 M-KEY NVMe SSD expansion supported. We plan to introduce CM5 v2.0 with LPDDR4 memory to offer more flexibility as memory prices change.
What are the operating temperature and supply availability?
CM5 has a standard operating temperature of 0°C ~ 80°C, and ArmSoM commits to availability at least through August 2034. For special temperature ranges or industrial customization, contact sales@armsom.org.
How do I deploy AI models on the NPU?
Use the Rockchip RKNN Toolkit to convert TensorFlow, PyTorch, ONNX and other framework models into RKNN format, then run them on the 6 TOPS NPU; for large language models, refer to the official RKNN-LLM solution. ArmSoM provides complete model-conversion tutorials and sample code.

Ready to integrate CM5 into your product?

From sample orders to volume customization, ArmSoM provides end-to-end support

Or email us at sales@armsom.org