ESP32-S3 N16R8 (16MB flash, 8MB PSRAM)

The ESP32-S3 adds to the WROOM what matters for intelligent edge: vector instructions that accelerate neural networks and signal processing, plus 8 MB of PSRAM (N16R8 config) to handle substantial buffers — spectra, audio windows, small images. It is the TinyML board: vibration or acoustic anomaly detection right at the foot of the machine, without a cloud round-trip.

It remains a microcontroller: no Linux OS, no heavy Python framework, a memory footprint to manage down to the KB. For a model that does not fit these constraints, or for real-time vision, you need the Raspberry Pi 5 (general-purpose) or the Jetson Orin Nano (CUDA). The S3 excels when the model is small, latency is critical and consumption is counted.