The ALPON brings AI on edge closer to sensors and cameras so you do not have to stream video to the cloud for inference. It runs computer vision workloads (object detection, classification, segmentation, OCR, pose, face) at up to 25 TOPS on the DEEPX DX-M1 NPU while handling LTE / Wi-Fi / Ethernet connectivity and fleet management in a single fanless box.

The ALPON X5 AI is the AI-capable successor to the X4. It adds a DEEPX DX-M1 NPU (25 TOPS), upgrades the compute to the Raspberry Pi CM5, and exposes an NVMe SSD slot. The X4 is an edge IoT computer based on the CM4 and does not ship with an AI accelerator.

For a full side-by-side comparison, see Compare ALPON Devices.

The ALPON fits deployments where vision or sensor data needs to be processed locally with low latency and resilient connectivity:

• Smart cities: traffic analytics, ANPR, pedestrian counting.
• Retail: shelf monitoring, queue analytics, loss prevention.
• Industry: defect detection, PPE compliance, line-of-sight safety.
• Transportation: in-vehicle driver monitoring, fleet telematics with vision.
• Agriculture and energy: remote assets monitored over LTE with AI triage at the gateway.

Every ALPON X5 AI ships with:

• 1x ALPON X5 AI edge AI computer
• 1x USB-C PD power adapter (15 V / 27 W)
• 1x 3-pin DC terminal block connector
• 1x 2x LTE antennas, 1x GNSS antenna, 1x Wi-Fi / BLE antenna
• Quick-start card with QR-linked provisioning

DIN-rail and wall-mount brackets are optional accessories and not included by default.

In most cases yes, but verify three things first:

1. Power: the X5 AI requires 12 to 32 V DC or 15 V USB-C PD. A 9 V supply is acceptable on the X4 but puts the X5 AI into restricted mode.
2. Footprint: mechanical dimensions are close but not identical. Check mounting-hole positions before committing.
3. Containers: rebuild any application container for arm64 on the CM5 toolchain.

Order units and accessories from sixfab.com. For volume pricing, custom variants (PoE, memory, storage), or evaluation loaners, reach out through Sixfab contact.

The host compute is the Raspberry Pi Compute Module 5 (CM5): a quad-core Cortex-A76 running at 2.4 GHz with up to 16 GB LPDDR4 memory and on-board eMMC. The CM5 handles the operating system, networking stack, application containers, and feeds the DEEPX NPU over PCIe Gen3.

The DEEPX DX-M1 sits on the CM5 PCIe Gen3 x1 bus as an M.2 module. It appears as character device /dev/dxrt0 once the kernel driver loads. The NPU has its own 4 GB on-chip memory pool and does not share system RAM with the CM5.

The ALPON pairs the CM5's on-board eMMC (factory flashed with ALPON OS) with an internal NVMe M.2 SSD slot for user workloads, models, and logs. Typical variants ship with a pre-installed NVMe; verify your SKU to confirm capacity.

The ADDON port uses an RJ45 physical socket but is not Ethernet. It exposes 3.3 V GPIO with SPI, UART, and I2C for Sixfab add-on modules. Plugging an Ethernet or PoE cable into this port will permanently damage the device.

No. Use ETHERNET 0 for anything performance-sensitive or PoE-powered.

• ETHERNET 0: native CM5 Gigabit PHY, PoE+ capable on applicable variants, the only port that accepts PoE power.
• ETHERNET 1: Gigabit Ethernet bridged over USB 3.0 through the TUSB8020 hub. Shares upstream bandwidth with one of the USB 3.0 ports. Cannot accept PoE.

Two hardware switches are exposed for field configuration:

• SW1 (Watchdog): ON enables the hardware watchdog, which resets the device if no keep-alive is received for 5 minutes. The watchdog can only be disabled with SW1 in software cannot override it.
• SW2 (Boot / Burn): OFF is normal boot. ON is image-flashing mode, used only during supervised re-flashing.

One status LED is reserved for user applications. The remaining LEDs are driven by Sixfab system services (power, Sixfab Connect agent, cellular, Ethernet link).

Yes. The ALPON includes a TPM 2.0 chip wired to the CM5 over SPI0. Use it from userspace with the standard tpm2-tools package:

sudo apt install -y tpm2-tools
tpm2_getcap properties-fixed

Train or obtain your model in any common framework (PyTorch, TensorFlow, Keras, XGBoost, MXNet) and export it to ONNX. The dx-com compiler then converts ONNX into the on-device .dxnn format. The DEEPX Runtime loads only .dxnn artifacts.

The DX-M1 is optimized for image-based CNNs. Transformer architectures are not supported on the current runtime.

• Object detection: YOLO v5, v7, v8, YOLOX; SSD variants
• Classification: ResNet, MobileNet, EfficientNet
• Segmentation: U-Net, semantic segmentation backbones
• OCR: PaddleOCR-based pipelines
• Face detection, recognition, pose estimation: CNN-based pipelines

Throughput depends on the model variant, input resolution, and whether the model was compiled with PPU (post-processing unit) support. Indicative numbers for nano-class YOLO variants compiled with --ppu:

Larger model variants trade FPS for accuracy. Benchmark your own pipeline; numbers vary with pre-processing, decoding, and rendering costs outside the NPU.

Yes. The DEEPX Runtime supports concurrent model execution on the DX-M1. Aggregate throughput is bounded by total NPU compute and the 4 GB on-chip memory footprint. Plan memory first, then latency.

The hard limit is the DX-M1's 4 GB on-chip memory. A .dxnn whose compiled footprint exceeds 4 GB will not load. If you hit the ceiling, recompile at a lower input resolution, switch to a lighter variant, or quantize before ONNX export.

The DEEPX Runtime provides first-class bindings for Python and C++:

• Python: dx_engine, bundled inside /usr/lib/libdxrt/dxrt-venv. Ideal for prototyping, OpenCV / GStreamer pipelines, and glue code.
• C++: dxrt_api.h. Use this for deterministic latency, embedded pipelines, and production inference services.

Map the character device into the container and run with --privileged:

docker run --rm --privileged   --device /dev/dxrt0   -v /opt/models:/models   alpon-infer:latest

For docker-compose, set privileged: true and devices: ["/dev/dxrt0"]. Build every image for arm64.

Use dxtop, the DEEPX command-line monitor. It reports per-device utilization, memory use, and temperature in real time, similar to htop for CPUs or nvidia-smi for GPUs.

dxtop
# live NPU utilization, memory, and temperature

dxtop runs over SSH or in the Sixfab Connect remote terminal.

No. The DX-M1 does not expose software-controlled performance or low-power profile switching on the current runtime. The NPU runs at a single operating profile.

Pre-loaded example models are not bundled in the ALPON OS base image. Reference pipelines (YOLOv8 detection, ResNet classification, U-Net segmentation) are published in the AI Development guide and the DEEPX Model Zoo, ready to pull into an arm64 container.

The ALPON ships with five independent network paths, all managed by the ALPON Cloud agent with automatic failover:

• LTE Cat 4 (150 Mbps DL / 50 Mbps UL) with on-board eSIM (eUICC SGP.22)
• Wi-Fi 5 dual-band 2.4 / 5 GHz, IEEE 802.11 b/g/n/ac
• Bluetooth 5.0 / BLE
• Gigabit Ethernet x 2 (ETHERNET 0 native, ETHERNET 1 USB-bridged)
• GNSS positioning (GPS + multi-constellation)

The cellular module covers a broad global band set:

B1 B2 B3 B4 B5 B7 B8 B12 B13 B18 B19 B20 B25 B26 B28 B34 B38 B39 B40 B41 B66

Connectivity on non-4G / LTE networks is not guaranteed. For full RF specs see Connectivity & Antenna Specifications.

Connector genders are not uniform across ports. Match the antenna gender to the port when sourcing replacements.

Yes. The ALPON ships with an on-board, soldered eUICC (SGP.22) eSIM that supports remote profile provisioning. Carrier switching and profile download are driven from Sixfab Connect, so the device can be activated without physical SIM handling.

Yes. Sixfab offers an optional 3-meter IP67-rated outdoor combination antenna that covers LTE main, LTE diversity, Wi-Fi, and GNSS with surface mounting. Use it when the device is installed inside a metal enclosure or indoors away from windows.

The ALPON accepts three independent power inputs and draws from the highest-voltage source connected at any moment:

All three can be connected simultaneously, but using one source at a time simplifies diagnostics.

The device enters restricted mode. The CM5, LTE modem, and both Ethernet ports stay active. The USB ports, PCIe bus (NVMe SSD and DEEPX DX-M1 NPU), ADDON port, and camera port are disabled. AI inference is not available in this mode because the NPU sits on the disabled PCIe bus. Restore ≥12 V to exit.

No. The ALPON is fanless by design. The top aluminum surface is the primary heatsink for the CM5, the DEEPX DX-M1 NPU, and the voltage regulators. Mounting orientation materially affects sustained AI throughput: prefer upright or cooling-surface-up mounting.

The ALPON operates from -20 to +60 °C ambient and tolerates storage from -40 to +85 °C. Humidity is rated up to 95% non-condensing. The aluminum enclosure is classified IP40.

Yes under AI load. The CM5 plus DEEPX DX-M1 consume more than a CM4-only X4, which is why the minimum PD wattage is higher (27 W vs 15 W typical on the X4) and a 45 W supply is recommended for peak inference. Idle and non-AI workloads are comparable.

The ALPON ships with ALPON OS, a Sixfab-maintained 64-bit Debian-based Linux distribution for the Raspberry Pi CM5. It includes the DEEPX DX-M1 kernel driver, Docker, the Sixfab Connect agent, and the cellular / Wi-Fi stacks preinstalled. A future Yocto-based image is planned for long-term support.

The default user is alpon with password sixfab.

The hostname is derived from the MAC address as alpon-<MAC>, so you can reach it over mDNS on the local network:

# By hostname (mDNS)
ssh alpon@alpon-<MAC>.local
 
# Or by IP
ssh alpon@<DEVICE_IP>

When the local network is unavailable, use the browser-based remote terminal in Sixfab Connect. It works over cellular alone.

Use the standard Debian tooling:

sudo apt update
sudo apt upgrade -y

Reboot if a kernel or firmware package was upgraded. For fleets, stagger rollouts one device at a time and watch ALPON Cloud for health reports before promoting the update.

The ALPON OS image is factory-installed and is not distributed as a public download. If your device appears to need re-flashing, contact Sixfab support with the device serial number for the correct recovery procedure. Unguided re-flashing can brick the CM5 eMMC.

The ALPON ships with a hardware-backed security foundation:

• TPM 2.0 on SPI0 for sealed storage and secure boot keys
• Factory-provisioned device identity bound to Sixfab Connect
• Signed ALPON OS packages delivered through authenticated APT repositories
• Per-user SSH keys and password rotation through Sixfab Connect
• Hardware watchdog (SW1) for autonomous recovery from hung services

Sixfab Connect is the fleet-management console for ALPON devices. It handles device registration, health monitoring, remote terminal access, container deployment, OTA updates, eSIM profile management, and network failover configuration. Every ALPON ships with a Sixfab Connect agent pre-installed.

1. Open connect.sixfab.com and sign in.
2. Scan the QR code on the device label, or select Manual registration and enter the serial number.
3. Bind the device to an organization and asset group.
4. Power the device on. First provisioning takes up to 3 minutes.

The device appears as Online once the Status LED turns solid green.

Build your image for arm64, push it to a registry Sixfab Connect can reach, and schedule a deployment to the target device or asset group. Typical command for a local build:

docker buildx build --platform linux/arm64   -t registry.example.com/alpon-infer:1.0   --push .

In Sixfab Connect, create a Deployment referencing that image and include the NPU device mapping (/dev/dxrt0) and privileged flag.

Yes. Open the device in Sixfab Connect and launch the remote terminal, then run dxtop. The terminal is a browser session bound to the cloud agent, so it works even if only cellular is up. Fleet-wide NPU metrics are on the roadmap; today the live view is per-device.

The ALPON is certified or in progress for:

CE FCC IC UKCA RoHS WEEE REACH

Safety is tested against EN IEC 62368-1. EMC is tested against EN 55032 and EN 55035. Radio conformity covers EN 300 328, EN 301 511, EN 301 908-1, and EN 301 908-13. FCC is covered under 47 CFR Part 15 Subpart B and Subpart C.

The ALPON X5 AI ships with a 2-year limited warranty against manufacturing defects under normal use. The warranty does not cover damage from out-of-range power, connecting Ethernet or PoE to the ADDON port, unauthorized modification, operation outside the rated environmental envelope, or physical and electrostatic damage.

• Supplying voltage outside the rated input range on the terminal block
• Plugging Ethernet or PoE into the ADDON port
• Opening the enclosure for reasons other than swapping the NVMe SSD or the DEEPX DX-M1 M.2 module (with ESD precautions)
• Operating outside the -20 to +60 °C ambient envelope
• Impact damage, liquid ingress, or ESD damage

See Safety & Compliance for the full list.

The ALPON complies with the EU WEEE directive. Do not dispose of it in unsorted household waste. Return the device through an authorized WEEE collection point, a local electronics recycler, or to Sixfab for responsible processing. Erase sensitive data from the NVMe SSD before disposal.