Radxa AIcore DX-M1/DX-M1M
Product overview
- DX-M1
- DX-M1M
Radxa AICore DX-M1 is a high-performance edge AI acceleration computing module based on DEEPX's DX-M1 module. It features high energy efficiency, high-precision computing, and ease of use, specifically optimized for edge computing scenarios and machine vision.
The Radxa AICore DX-M1 module is currently compatible with various single-board computers (SBCs), such as Radxa ROCK 5B/5B+ and other boards.
- High Energy Efficiency Design
Features an energy-efficient design with excellent inference performance and low power consumption, achieving 25 TOPS computing power with only 3-5W power consumption.
- High-Precision Computing
Utilizes IQ8™ (Intelligent Quantization Integer 8) technology, maintaining the same precision level as GPU's FP32.
- Ease of Use
- AI Frameworks: Supports ONNX, PyTorch, TensorFlow.
- System Support: Supports Ubuntu, Debian.
- Standardized Interface: Adopts standard M.2 2280 M Key form factor, compatible with industrial and embedded devices.
Radxa AICore DX-M1M is a compact edge AI acceleration module powered by the DEEPX DX-M1M, designed for fast and efficient on-device inference in embedded and industrial systems.
This AI processor delivers a stunning 25 TOPS (INT8) while maintaining ultra-low power consumption at just 3W, enabling high performance within a highly efficient power budget. It integrates onboard memory and QSPI flash in a standard M.2 M + B Key (PCIe Gen3 ×2) form factor. With broad host and OS compatibility, AICore DX-M1M provides a scalable, plug-in path for accelerating edge AI workloads, from proof-of-concept to production deployment.
- High Energy Efficiency Design
Achieves up to 25 TOPS of inference performance while consuming only up to 3W, providing an exceptional performance‑per‑watt ratio for edge and embedded deployments.
- Seamless Developer Experience
Built on a standard PCIe‑based module interface with broad OS and framework support, the AICore DX‑M1M ensures quick integration and shortened time‑to‑deployment.
- Standard Interface: M.2 M + B Key (PCIe Gen3 x2)
- Host Platform: x86 / ARM
- Software Support: Windows 10 / 11, Ubuntu 24.04 / 22.04 / 20.04 LTS, Docker
- AI Frameworks: Supports TensorFlow, ONNX, Keras, PyTorch via DX-COM compiler
Product appearance
- DX-M1
- DX-M1M
AIcore DX-M1 module (front)
AIcore DX-M1 module (back)
AIcore DX-M1M module (front)
AIcore DX-M1M module (back)
Specifications
- DX-M1
- DX-M1M
| Product | AIcore DX-M1M |
|---|---|
| Type | AI accelerator module |
| AI performance | 25 TOPS |
| Form factor | M.2 M key |
| Dimensions | 22 × 80 mm |
| Interface | PCIe Gen 3 ×4 |
| Memory configuration | 4GB LPDDR5 + 1Gbit QSPI NAND Flash |
| Debug interface | UART0, JTAG1 |
| Supported AI frameworks | PyTorch, ONNX, TensorFlow |
| Supported OS | Linux (Ubuntu, Debian) |
| Supported architectures | ARM |
| Product | AICore DX-M1M |
|---|---|
| Type | AI Acceleration Module |
| AI Performance | 25 TOPS |
| Memory | 1GB LPDDR4X@4266 MT/s |
| Storage | QSPI 1Gbit NAND / NOR Flash |
| Module Interface | PCIe Gen3 x2 |
| Host Interface | PCIe Gen3 x4 (Supports Gen 1/2/3 & x1/x2) |
| Software | Support Windows 10 / 11, Ubuntu 24.04 / 22.04 / 20.04 LTS, Docker |
| Al Frameworks | Support TensorFlow, ONNX, Keras, PyTorch by DX-COM compiler converted |
| Operating Temperature | -25 ~ 65°C (Non-Throttling) / -25 ~ 85°C (Throttling) |
| Host Platform | x86 / ARM |
| Power Consumption | 3W (Typical) |
| Form Factor | M.2 M + B Key, 22mm x 42mm |
Application scenarios
- Smart cameras
- Autonomous driving
- Consumer electronics
- Security surveillance
- Edge computing
- Smart mobility