From nRF9160 to NVIDIA Jetson: Choosing the Right Edge Hardware

The Decision Tree

Start with what you need the device to do, and the right edge device becomes obvious. The Powoflow hardware portfolio is designed so that each device occupies a distinct niche. There is no overlap, and no device tries to do everything.

Do you need GPS location tracking?

If your primary requirement is tracking the location of mobile assets — vehicles, containers, heavy equipment, personnel — the KA1 edge device is the answer. Built around the Nordic Semiconductor nRF9160 SiP, it provides LTE-M/NB-IoT cellular connectivity with integrated GNSS. The SMART location strategy (90% thin scans, 10% full fixes) delivers 56-day battery life on a single charge. It reports position, motion state, battery level, and temperature over cellular networks worldwide.

Do you need video analytics or AI vision?

If you need to analyze camera feeds in real time — person detection, PPE compliance, license plate recognition, wildlife monitoring — the Mesa N1 edge device is purpose-built for that workload. It runs NVIDIA Jetson compute modules (Orin Nano, Orin NX, or AGX Orin) with TensorRT- optimized YOLOv8 models. It connects to ONVIF cameras, processes frames at the edge, and sends only detection metadata to the cloud.

Do you need environmental sensing?

If you need to measure temperature, humidity, barometric pressure, CO2, PM2.5, soil moisture, water level, or other environmental parameters, the S21XX LoRaWAN sensor family covers the broadest range of measurement types. These are low-power, battery-operated sensors that communicate via LoRaWAN to a gateway and then to the cloud. Battery life is measured in years, not months.

Do you need solar-powered remote monitoring?

For off-grid installations that need both connectivity and extended autonomy, the KA1-CO (Connected Operations) variant adds solar charging, expanded I/O, and BMS (Battery Management System) integration. It is designed for solar farm monitoring, remote infrastructure, and unmanned sites where mains power is unavailable.

Connectivity Comparison

The connectivity technology determines range, power consumption, data throughput, and operating cost. Each edge device uses the connectivity stack that best matches its use case.

LTE-M / NB-IoT (KA1, KA1-CO)

Cellular LPWAN (Low Power Wide Area Network) provides global coverage using existing mobile infrastructure. LTE-M supports mobility and handover between towers, making it ideal for asset tracking. NB-IoT provides deeper indoor penetration and lower power consumption for stationary deployments. Data costs are typically $1 to $5 per month per device through MVNO agreements. Range is effectively unlimited within cellular coverage.

LoRaWAN (S21XX sensors)

LoRaWAN provides license-free, long-range communication (up to 10 km in open terrain, 2 to 5 km in built environments) at extremely low power. A single LoRaWAN gateway can support thousands of sensors. There are no per-device data costs beyond the gateway's backhaul connection. Battery life for LoRaWAN sensors typically exceeds 5 years with 15-minute reporting intervals.

WiFi / Ethernet (Mesa N1)

Edge AI gateways require high-bandwidth, low-latency connectivity for video streaming and model updates. The Mesa N1 connects via Gigabit Ethernet or WiFi 6 to the site network. Video streams from ONVIF cameras are consumed locally; only detection metadata (typically under 1 KB per event) is sent to the cloud. Firmware and model updates are delivered over-the-air via a managed edge component runtime.

Power Comparison

Power budget is often the deciding constraint for edge hardware selection. The range spans from microwatts to tens of watts.

S21XX LoRaWAN sensors — Operate on 2x AA lithium batteries (3.6V, 2600 mAh). Sleep current is under 5 microamps. Active transmit current is 40 to 120 mA for less than 2 seconds per cycle. Expected battery life: 3 to 7 years depending on reporting interval.
KA1 cellular tracker — Rechargeable 6000 mAh LiPo battery. Sleep current is under 20 microamps. GNSS fix and LTE-M transmit consume 100 to 250 mA for 10 to 30 seconds per cycle. The SMART location strategy achieves 56-day battery life with hourly location updates.
KA1-CO solar variant — Same battery as KA1 plus a 2W solar panel and MPPT charge controller. In locations with 4+ hours of daily sunlight, the device runs indefinitely without manual recharging.
Mesa N1 edge AI — Requires mains power (12V DC or PoE). Typical power consumption is 7W to 30W depending on the Jetson module and inference workload. The Orin Nano draws 7 to 15W; the AGX Orin can peak at 60W under full GPU load. Not a battery-operated device.

Compute Comparison

The compute architecture varies from bare-metal MCU firmware to full Linux with GPU acceleration.

MCU-Class (nRF9160)

The KA1 runs on the Nordic nRF9160 System-in-Package: an ARM Cortex-M33 at 64 MHz with 1 MB flash and 256 KB RAM. The firmware is written in C using the Zephyr RTOS. There is no operating system overhead — every byte of memory and every CPU cycle is dedicated to the application. This extreme efficiency is what enables the 56-day battery life. The tradeoff is that the device runs a fixed firmware image; it cannot run arbitrary applications or AI models.

GPU-Class (NVIDIA Jetson)

The Mesa N1 runs on NVIDIA Jetson modules: full ARM-based Linux systems with dedicated CUDA GPU cores. The Orin Nano provides 26 TOPS (Tera Operations Per Second) of AI compute; the AGX Orin exceeds 100 TOPS. These devices run Ubuntu, Docker containers, Python, and TensorRT-compiled neural networks. They can process multiple video streams simultaneously, run multiple detection models, and stream results via MQTT to the cloud — all at the edge.

The compute gap between these two categories is enormous and intentional. You would never put a Jetson on a battery-powered tracker (it would last hours), and you would never try to run YOLOv8 on an nRF9160 (it has no GPU). Each device is optimized for its specific operational role.

Deployment Scenarios

Real-world deployments typically combine multiple edge device types across a single site or portfolio.

Remote Solar Farm

A 50 MW solar farm in a rural area with limited cellular coverage: one Mesa N1 with 4 cameras for perimeter security and wildlife monitoring, twelve S21XX sensors measuring irradiance, panel temperature, ambient temperature, and humidity, two KA1-CO units on mobile maintenance vehicles for location tracking, and a LoRaWAN gateway connecting the S21XX sensors to the site's backhaul.

Mining Operation

An open-pit mine with haul roads, processing plant, and tailings storage: three Mesa N1 units covering pit entry points, processing plant access, and the tailings dam. Twenty KA1 trackers on haul trucks, water carts, and light vehicles. Fifteen S21XX sensors monitoring vibration on crushers, mill bearing temperatures, and tailings dam piezometer levels. Environmental S21XX sensors for dust (PM2.5/PM10), noise, and weather.

Port and Maritime Facility

A container port with wharf, laydown area, and vessel berths: two Mesa N1 units for gate access (license plate OCR) and wharf safety (PPE detection, exclusion zone monitoring). KA1 trackers on reach stackers, mobile cranes, and pilot boats. S21XX sensors for tide level, wind speed, and air quality at the berth face.

Distributed Infrastructure Network

A utility with 200+ remote substations and pump stations: KA1-CO solar-powered units at each unmanned site, reporting location, temperature, and tamper detection. S21XX sensors on transformers and pumps for condition monitoring. Mesa N1 edge AI at high-value substations for intrusion detection. All feeding into a single Powoflow platform with geo-intelligence overlay for natural hazard awareness.

Making the Choice

The decision is not about which device is “best” — it is about which device fits each specific operational requirement. The simplified decision matrix:

Need to track where something is? KA1.
Need to see and analyze what is happening? Mesa N1.
Need to measure an environment? S21XX.
Need to operate off-grid with solar? KA1-CO.

Most deployments combine two or three device types. The Powoflow platform manages all of them through a unified interface — same map, same alarm engine, same work order system, same mobile app. The hardware is diverse; the operational experience is unified.