The Firefly EC-AGXOrin is an edge-AI inference platform built around the NVIDIA Jetson AGX Orin 64 GB module. While its target deployment mirrors rugged edge AI systems such as AAEON BOXER-8645AI and Vecow RAC-1000, the EC-AGXOrin pulls ahead in interface provision, processing capability and camera-front-end support. It is designed for domains such as in-vehicle computing, robotic control, machine vision, intelligent video analytics and autonomous mobile robotics.
Core Hardware Configuration & Performance
Firefly has configured the EC-AGXOrin for high extensibility and strong compute throughput. Key parameters include:
- Camera & video processing: The system provides 8 × GMSL2 (Gigabit Multimedia Serial Link 2) interfaces (input via two 4-pin Mini FAKRA connectors). It supports up to 22 channels of 1080p decoding or a single 8K@30fps H.265 stream.
- AI compute and memory/storage: Leveraging the Jetson AGX Orin 64GB module, it delivers up to 275 TOPS (INT8) AI performance. The on-board memory is 64 GB LPDDR5, and storage includes 64 GB eMMC. Expansion is provided via an M.2 NVMe slot and a MicroSD card slot.
- Interfaces & connectivity: A comprehensive I/O list includes 1 × 10 GbE RJ45 port, 5 × Gigabit-Ethernet RJ45 ports, USB 3.0 ports, HDMI 2.0 output, RS-232/RS-485 serial ports, CAN bus interface; wireless connectivity supports WiFi 6, Bluetooth 5.2, 4G/5G cellular as well as GPS/GNSS positioning.
Technical Specification Table
Below is a refined specification table for the EC-AGXOrin:
| Category | Specification |
|---|---|
| System Module (SoM) | NVIDIA Jetson AGX Orin 64 GB |
| CPU | 12-core Arm Cortex-A78AE v8.2 (64-bit) at up to 2.2 GHz, with ~3 MB L2 + ~6 MB L3 cache |
| GPU / AI Accelerator | NVIDIA Ampere architecture: 2 048 CUDA cores + 64 Tensor cores; plus 2× NVDLA v2, PVA v2 vision accelerator |
| AI Compute Performance | Up to ~275 TOPS (INT8) at ~60 W |
| Video Encoding (H.265) | 2× 4K@60fps • 4× 4K@30fps • 8× 1080p@60fps • 16× 1080p@30fps |
| Video Decoding (H.265) | 1× 8K@30fps • 3× 4K@60fps • 7× 4K@30fps • 11× 1080p@60fps • 22× 1080p@30fps |
| System Memory | 64 GB LPDDR5 (256-bit, ≈204.8 GB/s bandwidth) |
| Internal Storage | 64 GB eMMC 5.1 |
| Storage Expansion | M.2 2280 M-Key (NVMe SSD) • MicroSD (TF) card slot |
| Video Output | HDMI 2.0 (up to 4K@60Hz) |
| Camera Interfaces | 8× GMSL2 via two Mini FAKRA 4-pin connectors |
| Audio Interface | 3.5 mm audio jack (supports microphone input, CTIA standard) |
| Networking | 1× 10 GbE RJ45 • 5× GbE RJ45 • M.2 E-Key (WiFi/BT) • Mini PCIe (4G LTE) • M.2 B-Key (5G) • GNSS support • 7 antenna holes |
| USB Interfaces | 4× USB 3.0 Type-A • 2× USB Type-C (OTG & debug) |
| Serial Interfaces | 1× RS-232 • 1× RS-485 (via Phoenix terminal) |
| Expansion Interfaces | M.2 2230 E-Key (WiFi/BT) • M.2 B-Key (4G LTE) • M.2 2280 M-Key (NVMe SSD) • Mini PCIe (4G expansion) • 8-channel DIO via DB-9 • 2×12-pin Phoenix terminal (RS-485, RS-232, CAN 2.0, UART, IO pins, 3.3/5 V power, 3.5 mm pitch) |
| Other Features | Power button, Reset button, Recovery button; Run & Power status LEDs; Power input: 24 V DC (5.5×2.1 mm jack, wide input support 9–36 V); Dimensions: 277.95×136.09×88.0 mm; Operating temp: -20 °C to 60 °C; Storage temp: -20 °C to 70 °C; Humidity: 10%–90% non-condensing |
- The stated “up to 275 TOPS” performance aligns with NVIDIA’s specification for the Jetson AGX Orin series, which delivers up to 275 TOPS of AI inference capability under Sparse INT8 precision.
- The memory bandwidth figure of approximately 204.8 GB/s for the 64 GB 256-bit LPDDR5 configuration matches the expected value for the Jetson AGX Orin 64 GB module.
- The video encoding and decoding channel counts correspond to Firefly’s official specification for the EC-AGXOrin. While NVIDIA’s base Jetson AGX Orin documentation may list slightly fewer decode channels, the listed capabilities—such as support for up to 22 streams of 1080p @ 30 fps—are consistent with optimized partner implementations.
- The CPU cache configuration of 3 MB L2 + 6 MB L3 is based on Firefly’s documentation for this model. While NVIDIA’s reference brief details smaller per-core and per-cluster caches, Firefly’s published configuration reflects the effective total cache allocation on this device.
- Regarding connectivity, the EC-AGXOrin supports Wi-Fi 6, Bluetooth 5.2, and 4G/5G cellular networking, with optional GPS/GNSS positioning via modular expansion. Actual wireless and navigation functions may vary depending on the installed M.2 or Mini PCIe modules.
- The 8-channel GMSL2 camera input capability is fully supported, allowing simultaneous high-bandwidth video capture for multi-camera AI inference scenarios.
The EC-AGXOrin features an industrial-grade aluminium enclosure with dual active cooling fans, ensuring reliable 24/7 operation in harsh edge-AI environments. Multiple design views are provided for integration and service planning:
- Front-face view: shows the overall device layout and form factor.
- Front-panel close-up: highlights key elements such as buttons and LED indicators.
- Rear camera-input view: emphasizes the eight GMSL2 connectors for camera signal input.
- Full rear-panel view: displays all available I/O interfaces for efficient cabling and deployment.
The mechanical design focuses on durability, thermal performance, and serviceability, making it suitable for autonomous systems, robotics, and outdoor edge deployments.
The EC-AGXOrin supports a wide range of AI development and deployment frameworks:
Operating system and core libraries:
Runs Ubuntu 22.04 LTS with full desktop and graphical acceleration, supporting CUDA, TensorRT, and cuDNN for optimized inference performance.
Model and framework compatibility:
Compatible with modern robotics and AI frameworks such as ROS, and capable of running large-language models (e.g., Llama, ChatGLM, Qwen) and vision models (e.g., EfficientViT, NanoSAM, TAM). It can also host image-generation models like Stable Diffusion and Flux for creative or synthetic-data applications.
Development and deployment flexibility:
Supports PyTorch, TensorFlow, MATLAB, and PaddlePaddle, with cuDNN acceleration enabled. Developers can build custom operators, deploy workloads via Docker containers, and run local inference pipelines using frameworks such as Ollama and ComfyUI for visual AI workflows.
This versatility makes the EC-AGXOrin ideal for robotics control, machine vision, video analytics, autonomous navigation, and edge-based generative AI.
Within the edge-AI hardware segment, the EC-AGXOrin represents a next-generation industrial platform. Compared to earlier Jetson AGX Xavier-based solutions—which typically supported fewer camera channels and lower compute power—the EC-AGXOrin offers enhanced expandability, 8-channel GMSL2 support, and significantly higher AI performance through the Jetson AGX Orin 64 GB module.
As of now, official pricing has not been publicly released. Prospective buyers and integrators are encouraged to contact Firefly’s sales team or authorized distributors for detailed quotation and availability information.
