Embedded AI Computer
Highlights
- Up to 100 TOPS AI performance on NVIDIA Jetson Orin platform
- Compact DIN-rail enclosure (6 TE) for control cabinet installation
- Headless design for autonomous applications without monitor
- 8 galvanically isolated digital inputs, 2 analog inputs, 1 switching output
- Pre-installed JetPack/Ubuntu – ready to use
- Directly combinable with LyconSys LTE/IoT modems and USB cameras
Compact industrial PC for edge AI applications based on NVIDIA Jetson Orin. Up to 100 TOPS AI performance in DIN-rail enclosure. Four model variants for different performance requirements.
Up to 100 TOPS AI performance
NVIDIA Jetson Orin NX 16 GB: 100 TOPS, 30–60 FPS object detection with YOLOv8 at < 10 ms latency, TensorRT-optimised.
Compact DIN-rail format
6 TE on DIN rail (EN 50022): fits standard control cabinets alongside modems and routers, 9-24V DC, IP30.
NVIDIA JetPack pre-installed
Ubuntu, CUDA, TensorRT and cuDNN pre-installed from factory. First SSH connection typically in under 5 minutes.
Industrial I/O interfaces
8× galvanically isolated digital inputs, 2 analog inputs, 1 switching output, 3× USB 3.1 for cameras and modems.
The Embedded AI Computer is a compact industrial PC for edge AI applications, based on the NVIDIA Jetson Orin platform. In the DIN-rail enclosure for control cabinet installation, it offers up to 100 TOPS AI performance for image processing, object detection and autonomous decision systems.
Cooling Variants
| Variant | Description | Application |
|---|---|---|
| Standard (Fan) | Heat sink with controlled fan | Normal industrial environments, occasional full load |
| Water Cooling (optional) | External water loop with pump and radiator | Continuous full load, fanless operation, extreme environments |
Connections & I/O
- 3× USB 3.1 Type-A: For cameras, LTE/IoT modems, storage devices
- 1× Gigabit Ethernet: Network connection, remote access
- 8× Digital inputs: Galvanically isolated, max. 30 V – for sensors, switches, signals
- 1× Switching output: MOSFET, max. 500 mA – for relays, LEDs, actuators
- 1× RS485 interface: For MODBUS-RTU
- NVMe slot: M.2 2230/2242 with 4 PCIe lanes for fast SSD storage
Headless Design
No HDMI output – the system is designed for autonomous operation without monitor. Configuration and access via network (SSH, web interface) or serial.
Application Areas
Image and camera-based systems Quality control, object detection, classification – with directly connected USB cameras or network-connected GigE Vision cameras.
Autonomous measurement systems Weather monitoring, object tracking, event detection – with local data processing and decision logic without cloud dependency.
Edge data processing Preprocessing of large data volumes, local classification, real-time decisions – reduces latency and bandwidth to the cloud.
Remote operation with service access Combination with LyconSys VPN routers and LTE modems for secure remote access to remote installations.
Software & NVIDIA Ecosystem
Pre-installed and ready to use
The Embedded AI Computer is delivered with pre-installed NVIDIA JetPack – the official SDK for Jetson platforms. The system is ready to use:
- Ubuntu Linux as base operating system
- CUDA for GPU-accelerated computing
- cuDNN for optimized deep learning operations
- TensorRT for high-performance inference
- VPI (Vision Programming Interface) for image processing
Supported Frameworks
All common AI frameworks run on the Embedded AI Computer:
- PyTorch – Research and production
- TensorFlow / TensorFlow Lite – Training and inference
- ONNX Runtime – Framework-independent models
- DeepStream – Video analytics pipelines
- Isaac ROS – Robotics applications
LyconSys BSP Package
In addition to standard JetPack, we offer our own Board Support Package (BSP) with:
- Drivers for I/O interfaces (digital inputs, analog inputs, switching output)
- Sample code and documentation
- Optimized system configuration for industrial use
- OTA update support
Your own image? No problem
On request we deliver the system with bootloader only – you install your own Linux image. We support you in integrating our drivers into your build system.
System Integration with LyconSys Products
The Embedded AI Computer is directly compatible with our other DIN-rail modules:
| Product | Connection | Application |
|---|---|---|
| 4G/LTE USB Modem | USB | Cellular internet, remote access |
| IoT USB Modem | USB | Power-efficient IoT connectivity with GPS |
| VPN Router ERT1000 | Ethernet | Secure VPN connection, firewall |
| USB Cameras | USB | Image capture for AI processing |
All devices fit on the same DIN rail and use the same power supply (9-24V DC).
| Property | Value |
|---|---|
| Models | ORINHSN8, ORINHSX8, ORINHSX16 |
| Platform | NVIDIA Jetson Orin Nano / Jetson Orin NX |
| AI Performance | 40 – 100 TOPS (sparse INT8) (model-dependent) |
| RAM | 8 GB – 16 GB LPDDR5 (model-dependent) |
| Storage | NVMe M.2 (2230/2242), 4× PCIe lanes |
| USB | 3× USB 3.1 Type-A |
| Network | 1× Gigabit Ethernet (RJ45) |
| Digital Inputs | 8× galvanically isolated (max. 30 V) |
| Analog Inputs | 2× 8-bit |
| Switching Output | 1× MOSFET (max. 500 mA) |
| Display | Headless (no HDMI output) |
| Real-time Clock | RTC, battery buffered (CR2016) |
| Watchdog | Hardware Watchdog |
| Voltage Monitoring | External supply voltage monitored via PMIC |
| Cooling | Heat sink + controlled fan (standard) or water cooling (optional) |
| Power Supply | 9 V – 24 V DC (max. 28 V), max. 25 W |
| Enclosure | DIN-rail enclosure (ABS) per DIN 43880 |
| Mounting | DIN rail (EN 50022), wall mounting (optional) |
| Protection | IP30 |
| Dimensions | 105 × 54 × 82 mm (W × H × D), 6 TE |
| Weight | 462 g |
| Operating Temperature | -30°C to +70°C |
| Humidity | 5% to 95% (non-condensing) |
| Operating System | NVIDIA JetPack / Ubuntu Linux (pre-installed) |
| CE | EN 61000-6-2, EN 61000-6-4, EN 61000-4-2, EN 61000-4-3, EN 61000-4-4, EN 61000-4-6, EN 61000-4-29, EN 55022 |
| RoHS | EN 50581 |
- Embedded AI Computer (selected model variant)
- Quick start guide
Model Overview
Entry-level, simple inference
40 TOPSORINHSN8
- RAM: 8 GB
- GPU cores: 1024
- Platform: Jetson Orin Nano
- HD: 2 TB NVMe
Demanding real-time applications
70 TOPSORINHSX8
- RAM: 8 GB
- GPU cores: 1024
- Platform: Jetson Orin NX
- HD: 2 TB NVMe
Maximum performance, large models
100 TOPSORINHSX16
- RAM: 16 GB
- GPU cores: 1024
- Platform: Jetson Orin NX
- HD: 2 TB NVMe
All models use high-bandwidth LPDDR5 memory and are compatible with the same carrier board and enclosure. Upgrading to a more powerful module is possible at any time.
FAQ
That depends on the complexity of your AI models and real-time requirements:
- ORINHSN4 (20 TOPS): Simple classification, small models, low frame rate
- ORINHSN8 (40 TOPS): Standard object detection (e.g. YOLOv5/v8), 720p/1080p video
- ORINHSX8 (70 TOPS): Multiple cameras in parallel, larger models, 4K video
- ORINHSX16 (100 TOPS): Maximum performance, very large models, multiple parallel pipelines
When in doubt, we recommend starting with the ORINHSN8 – it offers good value for money and covers most applications.
The system is designed for headless operation. Access is via:
- SSH: Command line over network
- VNC/Remote Desktop: Graphical interface over network
- Web Interface: For custom applications or monitoring
- Serial console: For debugging and emergency access
In combination with a LyconSys VPN router you can also securely access the system remotely.
Standard fan cooling is sufficient for most applications, even under high load. We recommend water cooling for:
- Continuous operation under full load (e.g. 24/7 video analytics)
- Fanless operation in noise-sensitive environments
- Extreme ambient temperatures or dusty environments
Water cooling is available as optional accessory and includes pump, radiator and fittings.
The M.2 slot supports NVMe SSDs in 2230 or 2242 format with PCIe connection (4 lanes). Most common NVMe SSDs work, e.g.:
- Samsung PM991a (2230)
- Kioxia BG4/BG5 (2230)
- Western Digital SN520/SN740 (2230/2242)
We recommend SSDs with industrial temperature range for continuous operation.
Yes. Our team has years of experience with NVIDIA Jetson, CUDA and image processing. We support you with:
- Selection and training of AI models
- Optimization for TensorRT
- Integration of cameras and sensors
- Development of custom firmware
Contact us – we can implement the project jointly or as contract development.
Request now
Would you like more information, a quote or technical consultation?