NVIDIA Jetson Orin Nano 8GB Dev Kit

The Jetson Orin Nano 8 GB is made for one thing that the other boards cannot do: deep neural network inference in real time at the foot of the machine. Its 1024 CUDA cores and 32 Tensor cores process video streams, object detection and heavy models locally — visual quality control, counting, guidance — where a Raspberry Pi plateaus. NVIDIA re-rated the module from 40 to 67 TOPS (INT8) via a JetPack software update.

It is not a sensor or a microcontroller: it consumes 7–25 W, costs far more, and demands the CUDA/TensorRT ecosystem (a real learning curve). You deploy it only when the model requires it — deep vision, critical latency. For tabular ML, signal or threshold monitoring, an ESP32-S3 or an RPi 5 is the right level, at a fraction of the cost and consumption. The edge guide details this trade-off.