SupermicroARS-E103-JONX-H2-01-G2

This is the box for analysing video or sensor signals where they are produced, such as the camera watching a production line. The Jetson Orin NX puts the Arm CPU, an Ampere-architecture GPU and 16 GB of LPDDR5 shared between them on a single module, and in Super mode (the higher clock and power profile NVIDIA enables from JetPack 6.2) it reaches 157 TOPS (trillion operations per second). With no fan or vents, the datasheet rates it from -25 to 60 °C, and the two RS-232/422/485 serial ports, CAN bus and four digital inputs and four outputs connect it to PLCs and machinery without intermediate gateways.

Those 157 TOPS are the sparse figure (which skips the model's zero weights) at the highest power mode; for dense compute NVIDIA quotes 78, and that is the number to size against. Memory also sets the ceiling, because the 16 GB is shared between the OS and the model: image detectors and classifiers fit comfortably, a large language model does not. It is also Arm, so your software and containers have to be built for aarch64, something we check before proposing it.

What we do

We deliver it with JetPack, NVIDIA's Ubuntu-based OS, with Super mode configured, and on top we install what the project calls for: DeepStream to process several cameras at once, your models exported to TensorRT, or k3s (a lightweight Kubernetes) to deploy and update containers across many units from one place. When there are dozens spread across shops or plants, that fleet management saves the most work: JetPack and container updates go out from the same point, and since the unit has no BMC, for sites without a technician we prepare a spare with the same image that can be swapped in just by plugging it in.