Thorium
Accelerated by NVIDIA Jetson 
Accelerated by NVIDIA Jetson
Thorium is engineered for AI applications where conventional embedded systems no longer suffice and where cloud or server-based solutions are technically or economically impractical. It combines extreme compute performance with industrial-grade mechanics, enabling complex AI models to run directly at the edge – deterministic, low-latency, and independent of centralized infrastructure.
- Server-class AI performance at the edge: Over 2000 FP4 TFLOPS enable deployment of very large models directly in the field.
- High system integration: Compute, networking, storage, and security combined in a compact, industrial-grade platform.
- Built for demanding environments: IP67 rating and rugged industrial connectors.
- Scalable & production-ready: Designed for series products and long-term availability.
Thorium: Server-class AI performance. Directly at the edge.
Thorium is engineered for AI applications where conventional embedded systems no longer suffice and where cloud or server-based solutions are technically or economically impractical. It combines extreme compute performance with industrial-grade mechanics, enabling complex AI models to run directly at the edge – deterministic, low-latency, and independent of centralized infrastructure.
- Server-class AI performance at the edge: Over 2000 FP4 TFLOPS enable deployment of very large models directly in the field.
- High system integration: Compute, networking, storage, and security combined in a compact, industrial-grade platform.
- Built for demanding environments: IP67 rating and rugged industrial connectors.
- Scalable & production-ready: Designed for series products and long-term availability.
Accelerated by NVIDIA Jetson
NVIDIA Jetson is the leading AI-at-the-edge computing platform with over half a million developers. With pre-trained AI models, developer SDKs and support for cloud-native technologies across the full Jetson lineup, manufacturers of intelligent machines and AI developers can build and deploy high-quality, software-defined features on embedded and edge devices targeting robotics, AIoT, smart cities, healthcare, industrial applications, and more. Cloud-native support helps manufacturers and developers implement frequent improvements, improve accuracy, and use the latest features with Jetson-based edge AI devices.
| Jetson T5000 | Jetson T4000 | |
|---|---|---|
| KI Performance | 2070 TOPS | 1200 TOPS |
| GPU | 2560-core NVIDIA Blackwell GPU with 96 Tensor Cores | 1536-core NVIDIA Blackwell GPU with 64 Tensor Cores |
| CPU | 14-core Arm® Neoverse®-V3AE 64-bit CPU | 12-core Arm® Neoverse®-V3AE 64-bit CPU |
| RAM | 128 GB 256-bit LPDDR5X | 64 GB 256-bit LPDDR5X |
| Storage | 2x NVMe M-Key Support 2280 & 2242 | |
| Power consumption | 40W - 130W | 40W - 70W |
Technical details
| Supported NVIDIA SOMs | Jetson T5000 or T4000 |
| Operating System | NVIDIA JetPack™ SDK 7.0: Ubuntu 24.04 (Linux Kernel 6.8) |
| Power Supply | 24V – 48V DC |
| Power Consumption | Max. 460 W |
| Cooling | Passive Cooling: Customizable by the customer or implemented by PCB Arts Active Cooling: With DIN-rail option or wall-mount solution |
| Standard Interfaces | 2× USB-C USB 3.2 (10 Gbit/s, screw terminal), 1× USB-C USB 2.0 for flashing, 1× Gigabit Ethernet (M12), 1× 10 Gbit/s Ethernet (M12) |
| Integrated Features | 8× opto-isolated I/Os, 2× opto-isolated CAN, 1× opto-isolated I²C, IMU: ICM-42670-P, Crypto Chip: ATSHA204A, external RTC |
| M.2 Slots | 1× M-Key 2280 for NVMe, 1× M-Key 2242 for NVMe, 1× E-Key 2230 for WiFi, 1× B-Key for LTE/5G |
| Debug Header | Via USB-C with PCB Arts adapter |
| Display Output | Via USB-C with PCB Arts adapter |
| Housing | Anodized aluminum |
| Dimensions | 210 mm × 103 mm × 61 mm Height TBD mm (with active cooling solution) |
| Weight | TBD |
| Operating Temperature | -20 °C ... +60 °C (with integrated heater) |
| IP Rating | IP67 |
| Certifications | CE, EMC, RoHS |
