Specifications
Model
Radxa NX4 system-on-module (with optional NX4 IO carrier board)
Product Type
260-pin SO-DIMM industrial system-on-module
SoC
Rockchip RK3576J (industrial-grade variant)
CPU
Octa-core — 4x Arm Cortex-A72 @ 2.2 GHz plus 4x Arm Cortex-A53 @ 2.0 GHz
GPU
Arm Mali-G52 MC3 with OpenGL ES 1.1/2.0/3.2, OpenCL 2.0 and Vulkan 1.2
NPU
6 TOPS INT8 accelerator supporting INT4, INT8, INT16, BF16 and TF32 mixed operations
Video Decode
H.265, VP9, AV1 and AVS2 up to 8Kp30 or 4Kp120; H.264/AVC and MJPEG up to 4Kp60
Video Encode
H.264, H.265 and MJPEG up to 4Kp60
System Memory
Up to 16 GB LPDDR5 at 5500 MT/s
Storage
Optional eMMC 5.1 up to 256 GB, optional UFS 2.0 up to 1 TB, optional onboard SPI flash, plus SDMMC interface
Networking
Wi-Fi 6 and Bluetooth 5.4 module with external antenna support; onboard Realtek RTL8211FSI-VS Gigabit Ethernet PHY
Video Output
HDMI 2.1 up to 4K at 120 Hz
Camera
2x 4-lane MIPI CSI
USB
1x USB 3.2, 1x USB 2.0 on the module
PCIe
1x PCIe 2.0 x1
Other I/O
UART, I²C, CAN bus, PWM, GPIO; I²S audio for external codecs or DAC/ADC
Module Connector
260-pin SO-DIMM edge connector
Supply Voltage
5 V
Dimensions
69.6 x 45 mm
Temperature Range
Commercial 0 to 60 °C; industrial –40 °C to 85 °C
NX4 IO Board — Storage
M.2 M-Key 2280 slot for NVMe SSD (plus M.2 M-Key 2230 and M.2 E-Key 2230 sockets)
NX4 IO Board — Display / Camera
1x HDMI output, 2x 4-lane MIPI CSI connectors
NX4 IO Board — Networking
Gigabit Ethernet RJ45 with PoE support via an additional PoE HAT
NX4 IO Board — USB
4x USB 3.2 Type-A plus 1x USB 3.2 Type-C
NX4 IO Board — Expansion
40-pin GPIO header with I²C, SPI, UART, PWM and GPIO; 4-pin CAN bus header; 4-pin PWM fan; 2-pin RTC battery
NX4 IO Board — Power
9–20 V DC via 5.5 x 2.5 mm barrel jack, plus 2-pin PoE backpower header
Overview
The Radxa NX4 is a 260-pin SO-DIMM system-on-module built around the Rockchip RK3576J industrial-grade SoC. The processor pairs four Cortex-A72 cores at 2.2 GHz with four Cortex-A53 cores at 2.0 GHz, adds a Mali-G52 MC3 GPU with OpenCL 2.0 and Vulkan 1.2 support, and carries a 6 TOPS INT8 NPU that handles INT4, INT8, INT16, BF16 and TF32 mixed operations — a configuration suited to vision inference at the edge rather than cloud offload.
Memory and storage are configurable to the application: up to 16 GB of LPDDR5 at 5500 MT/s, with optional eMMC 5.1 to 256 GB, UFS 2.0 to 1 TB, or onboard SPI flash, plus an SDMMC interface. Media handling reaches 8Kp30 or 4Kp120 decode across H.265, VP9, AV1 and AVS2, with H.264/MJPEG to 4Kp60 and encode to 4Kp60. The module provides HDMI 2.1 output at up to 4K120, two 4-lane MIPI CSI camera interfaces, a PCIe 2.0 x1 lane, CAN bus, and Wi-Fi 6 with Bluetooth 5.4 alongside an onboard Gigabit Ethernet PHY. It measures 69.6 x 45 mm, runs from 5 V, and is offered in commercial (0–60 °C) and industrial (–40–85 °C) grades.
Radxa also supplies the NX4 IO carrier board for evaluation and deployment. It exposes HDMI, two 4-lane MIPI CSI connectors, four USB 3.2 Type-A ports and one Type-C port, Gigabit Ethernet with optional PoE, an M.2 M-Key 2280 NVMe slot, a 40-pin GPIO header with I²C/SPI/UART/PWM, and a 4-pin CAN bus header, all powered from a 9–20 V DC barrel jack with a PoE backpower header.
Key Benefits
6 TOPS INT8 NPU with INT4/INT16/BF16/TF32 mixed precision covers on-module vision inference. Industrial –40 °C to 85 °C grade with up to 16 GB LPDDR5-5500 suits rugged edge products. UFS 2.0 to 1 TB or eMMC 5.1 to 256 GB gives storage options beyond typical SOM designs. 8Kp30 decode with AV1 and HDMI 2.1 4K120 output handles high-resolution media pipelines. A matching IO carrier board with M.2 NVMe, dual MIPI CSI, PoE-capable GbE and 9–20 V input shortens time to a working prototype.
Applications
Edge AI vision products, robotics and AMR controllers, industrial HMI and panel electronics, digital signage players, medical imaging and diagnostics front ends, smart retail terminals, and embedded systems that need an industrial-grade octa-core module with on-module NPU.
Request a Quote — RADXA NX4 — RK3576J 260-PIN SO-DIMM SOM SOM WITH 6 TOPS NPU AND 16GB LPDDR5
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →