Rugged Edge Computers Leverage NVIDIA Jetson Orin
A new wave of rugged edge computers based on the NVIDIA Jetson Orin platform is targeting harsh environments. Connect Tech introduced its Anvil-T5, a GPGPU-based computer for multi-sensor edge AI in defense and industrial markets that operates up to 60°C. Similarly, Twowin released the T218 Edge Computer, which uses the Jetson Orin Nano for demanding security and surveillance applications.
- The NVIDIA Jetson Orin modules are built on the NVIDIA Ampere architecture, which features third-generation Tensor Cores that can double throughput for AI inferencing by utilizing a technique called sparsity. The top-tier AGX Orin module offers up to 275 Tera Operations Per Second (TOPS) for INT8 compute, while the smaller Orin Nano module provides up to 40 TOPS. - The Jetson AGX Orin system-on-module (SoM) includes up to a 12-core Arm Cortex-A78AE CPU and a GPU with 2048 CUDA cores and 64 Tensor cores. In contrast, the more compact Orin Nano module features a 6-core Arm Cortex-A78AE CPU, a 1024-core GPU with 32 Tensor cores, and operates at a lower peak power of 15W compared to the AGX Orin's 75W. - Beyond its temperature range, the Twowin T218's rugged design includes two RJ45 Gigabit Ethernet ports with optional Power over Ethernet (PoE), an M.2 Key M slot for NVMe storage, and an M.2 Key E slot for WiFi, all within a fan-cooled enclosure. This allows it to power connected cameras directly and simplifies deployment in environments like outdoor perimeters. - The Connect Tech Anvil system provides extensive high-speed I/O, including dual 10GbE ports, multiple USB 3.2 ports, and two NVMe M.2 Key M slots for fast storage. It also supports a wide range of camera inputs, such as GMSL, FPD-Link III, and MIPI CSI, making it suitable for multi-sensor fusion applications. - General-Purpose GPUs (GPGPUs) are critical for edge AI because their architecture, with thousands of cores, excels at the parallel processing required by deep learning and neural networks. This allows for real-time, low-latency decision-making directly on the device, reducing reliance on the cloud. - The growth in the edge computing market is heavily driven by industrial automation, the expansion of IoT devices, and the need for real-time analytics in manufacturing. Hardware, such as rugged nodes, accounted for over 40% of edge-related spending in 2024. - To create different performance tiers, the Orin Nano module omits certain hardware blocks found in the larger AGX Orin. The Nano does not have the dedicated Deep Learning Accelerators (DLA) or the Programmable Vision Accelerator (PVA) present in the AGX variant. - Development on the Jetson Orin platform is enabled by the NVIDIA JetPack SDK, which is based on Ubuntu Linux. This software stack includes tools like TensorRT for optimizing AI inference, along with libraries for computer vision and GPU-accelerated computing.
