Troubleshooting
Diagnostic reference for the ALPON X5 AI edge AI computer: power, LEDs and boot,
connectivity (LTE, Wi-Fi, Ethernet), the DEEPX DX-M1 NPU, ALPON X5 AI OS,
and the ALPON Cloud agent. Each issue lists the symptom, the most likely cause, and
the verified fix.
Start from the physical layer and move up. Verify power first
(15 V / 27 W USB-C PD, DC 12 V to 24 V, or PoE+), confirm the
Status LED reports a healthy state, check network links
(LTE, Wi-Fi, Ethernet) in that order, verify the DEEPX DX-M1 NPU
enumerates at /dev/dxrt0 with dxtop, and finally inspect
ALPON X5 AI OS logs with journalctl -xe.
Most "dead on arrival" reports resolve with a 15 V / 27 W USB-C PD adapter or a 12 V to 24 V DC supply on the terminal block. Standard 5 V phone chargers do not negotiate enough voltage to boot the Raspberry Pi CM5 and DEEPX DX-M1 NPU. Start every diagnostic here.
The Status and Cellular LEDs surface device health without requiring shell access. Read these two indicators first; they narrow most issues to a single category in the accordion below. The full four-LED map (Power, Status, Cellular, User RGB) is in Physical Layout & I/O.
ETHERNET 0 to an upstream router.
Each row below is one symptom. Click a row to open it; the body covers what you see, why it happens, and the verified fix. Rows are colour-coded by category, with the dot, the category pill, and the left rail all matching, so you can scan the list by category at a glance.
Nothing lights up when power is connected.
The adapter does not negotiate a 12 V or 15 V USB-C PD profile (27 W minimum). Standard USB-C phone chargers deliver 5 V only and the ALPON X5 AI will not boot.
Pin 1 = Chassis Ground, Pin 2 = DC Negative, Pin 3 = DC Positive.The Raspberry Pi CM5 comes up and Ethernet links, but /dev/dxrt0 is missing, the NVMe SSD does not appear, USB devices are not detected, and the ADDON port is inactive.
The device has entered restricted mode. Input voltage is below 12 V, so the PCIe bus and peripheral rails are disabled by design.
Restricted mode prevents brownouts at low input voltage. The Raspberry Pi CM5 and both Ethernet ports stay alive; USB, PCIe (NVMe + DEEPX DX-M1), ADDON, and camera rails are disabled. AI inference is not available until the device leaves restricted mode. Supply options and sizing are covered in Power & Mounting.
The Raspberry Pi CM5 logs voltage or thermal warnings during sustained NPU inference and occasionally reboots.
The adapter delivers the minimum 27 W but cannot hold it under peak load, or the enclosure is mounted so airflow across the aluminum heatsink is blocked.
vcgencmd get_throttled that the Raspberry Pi CM5 is not hitting the under-voltage bit.The Raspberry Pi CM5 starts and powers down every few minutes without reaching a login prompt.
The hardware watchdog (SW1) is enabled but no userspace service is feeding it, so it resets the device every 5 minutes.
Always disconnect all power before flipping SW1 (Watchdog) or SW2 (Boot/Burn). Changing SW2 while powered can corrupt the Raspberry Pi CM5 eMMC.
The Sixfab Connect agent cannot reach the cloud. Cellular and Ethernet are both down, or the device has not completed provisioning.
ETHERNET 0 into an internet-connected router to give the device a known-good path.L.systemctl status sixfab-connect to inspect the agent.Weak LTE signal, missing or loose antenna, or the eSIM (eUICC) profile has not activated on the local network.
L (LTE Main and LTE Diversity). Never connect the Wi-Fi antenna (RP-SMA male) to an LTE port.mmcli -L mmcli -m 0 --signal-get
registered but throughput is low, switch carriers from the ALPON Cloud console.The Wi-Fi antenna is missing, the SSID uses an unsupported security mode, or the device is on the edge of the access point's coverage.
W.nmcli or Sixfab Connect.nmcli device wifi list nmcli device wifi connect "YOUR_SSID" password "YOUR_PASS"
ETHERNET 1 does not link up or shows low throughputETHERNET 1 is a USB-bridged Gigabit Ethernet port, not a native Raspberry Pi CM5 PHY. It shares the TUSB8020 hub upstream with one of the USB 3.0 ports. Heavy concurrent USB traffic reduces its throughput.
ETHERNET 0 for performance-sensitive traffic. It is the native Raspberry Pi CM5 Gigabit Ethernet and the only PoE-capable port.ETHERNET 1, avoid attaching high-bandwidth USB 3.0 devices to the shared hub.ETHERNET 1. It does not accept PoE.The injector is legacy IEEE 802.3af (not 802.3at Class 4), the cable is connected to ETHERNET 1 instead of ETHERNET 0, or your variant does not include the PoE module.
ETHERNET 0 only. ETHERNET 1 cannot accept power./dev/dxrt0 is missing or dxtop reports no deviceThe device is in restricted mode (PCIe disabled), the kernel driver failed to load, or the DEEPX DX-M1 M.2 module is not seated correctly.
lspci | grep -i deepx lsmod | grep dxrt ls -l /dev/dxrt*
sudo apt update sudo apt install --reinstall -y libdxrt
lspci does not list the DEEPX device, power down, open the enclosure, and reseat the DEEPX DX-M1 M.2 module. Apply ESD precautions.ImportError: No module named dx_engineThe script is running under the system Python. The dx_engine binding lives inside the bundled virtual environment at /usr/lib/libdxrt/dxrt-venv.
source /usr/lib/libdxrt/dxrt-venv/bin/activate python -c "import dx_engine; print(dx_engine.__version__)"
For scripts, invoke the venv Python directly: /usr/lib/libdxrt/dxrt-venv/bin/python my_script.py.
Cannot open /dev/dxrt0 inside a Docker containerThe container was started without the NPU character device mapped in, or without the privileges needed to attach to it.
ALPON X5 AI OS does not include Docker; install it first (see the ALPON X5 AI OS page).
Start the container with the device and --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"].
The compiled .dxnn footprint exceeds the DEEPX DX-M1's 4 GB on-chip memory. The NPU has its own dedicated memory pool, independent of Raspberry Pi CM5 system RAM.
nano or small instead of large or x).dxcom.The model was compiled without PPU (post-processing unit) support. Post-processing (NMS, box decoding) then runs on the Raspberry Pi CM5 CPU and becomes the bottleneck.
dxcom on a host PC:
dxcom \ -m /path/to/yolov8n.onnx \ -c yolov8n_config.json \ -o yolov8n-output
ppu block, calibration settings, and the matching preprocessing pipeline:
{
"inputs": {
"images": [1, 3, 640, 640]
},
"calibration_num": 100,
"calibration_method": "ema",
"train_batchsize": 32,
"num_samples": 1024,
"default_loader": {
"dataset_path": "./calibration_dataset",
"file_extensions": ["jpeg", "jpg", "png", "JPEG"],
"preprocessings": [
{ "resize": { "mode": "pad", "size": 640,
"pad_location": "edge",
"pad_value": [114, 114, 114] } },
{ "div": { "x": 255 } },
{ "convertColor": { "form": "BGR2RGB" } },
{ "transpose": { "axis": [2, 0, 1] } },
{ "expandDim": { "axis": 0 } }
]
},
"ppu": {
"type": 1,
"conf_thres": 0.25,
"num_classes": 80,
"layer": [
{ "bbox": "Mul_441", "cls_conf": "Sigmoid_442" }
]
}
}
.dxnn. PPU-compiled nano-class YOLO models reach roughly 50 FPS at 1280×720 and 20 to 25 FPS at 1920×1080.dxtop to confirm the NPU is saturated rather than the CPU.The bbox and cls_conf layer names (Mul_441, Sigmoid_442) come from the YOLOv8 ONNX graph and will differ across exports. Open the model in Netron to confirm the correct node names before compiling.
dxcom reports an unsupported operatorThe ONNX graph contains an attention block, transformer layer, or another operator outside the DEEPX DX-M1 supported set. The current dxrt-runtime targets image-based CNNs; transformers are not yet supported.
Wrong username, wrong IP or hostname, firewall blocking port 22, or the device is on a different subnet than your workstation.
ssh alpon@<DEVICE_IP> or ssh alpon@alpon-<MAC>.local.sudo passwd alpon to set a new password.The alpon account is used internally by Sixfab Connect and other system services. Change the password and, if you need a separate admin identity, create a new user with sudo rights. Deleting alpon breaks core system functions.
apt update or apt upgrade failsNo internet path, broken GPG key for the Sixfab or DEEPX repositories, or a partial upgrade that left dpkg in an inconsistent state.
# Verify connectivity first ping -c 3 deb.debian.org # Repair any half-configured packages sudo dpkg --configure -a sudo apt -f install # Refresh and retry sudo apt update sudo apt upgrade -y
If a repository's GPG key is missing, reinstall it from the Sixfab or DEEPX documentation before running apt update again. Reboot when a kernel or firmware package was upgraded.
The HDMI display was connected after boot, the cable does not support the negotiated mode, or the Raspberry Pi CM5 has not completed early boot.
Offline in Sixfab ConnectNo upstream path is currently reachable, or the agent lost its session and has not re-attached.
ETHERNET 0 into a wired router.systemctl status sixfab-connect journalctl -u sixfab-connect -n 200
sudo systemctl restart sixfab-connect
The device is offline, the container image exceeds available NVMe free space, or the image was not built for ARM64.
Online and the Sixfab Connect agent is healthy.docker buildx build --platform linux/arm64 -t my-image .df -h /. Prune unused images: docker image prune -a.
Seven quick checks that resolve more than 90 % of field issues. Each takes under a minute.
The chip on the right of each row tells you what to do — $ means a shell command
to run, → means a visual or label to confirm.
$vcgencmd get_throttled
15 V / 27 W USB-C PD or 12 V to 24 V DC. No under-voltage flag set.
→SW1 · SW2 · OFF
SW1 OFF = watchdog enabled (default); SW2 (Boot/Burn) OFF. Toggle only with all power disconnected.
→Status · Cellular
Note the exact state of Status and Cellular. See the LED reference.
→L · L · W · G
Four SMA connectors hand-tight. Wi-Fi/Bluetooth = RP-SMA male; LTE Main, LTE Diversity, GNSS/GPS = SMA female.
$ping -c 3 8.8.8.8
At least one of LTE, Wi-Fi, or ETHERNET 0 reaches the internet.
$dxtop
dxtop shows /dev/dxrt0 with live utilization and temperature.
→SN-XXXXXXXX
Read it from the enclosure label, next to the QR code. Include it in the support ticket.
If you can run a shell on the device (SSH or HDMI), this single block captures the journal, kernel ring buffer, PCIe topology, and OS release into one archive. Attach it to your ticket if you can; otherwise the checks above are enough to start.
# Single-shot bundle, run on the device over SSH sudo journalctl -b -n 2000 > ~/alpon-journal.log dmesg > ~/alpon-dmesg.log lspci -vv > ~/alpon-lspci.log cat /etc/os-release > ~/alpon-os.log tar czf ~/alpon-support.tar.gz ~/alpon-*.log # Copy the archive to your computer with scp, WinSCP, or Cyberduck.
Send a plain-text email with the device serial number, the observed LED state, and a one-line description of what you were doing when the issue appeared. Sixfab support can guide you through log collection from there.
If your symptom is not covered above, contact Sixfab support. Including the seven checks from the section above (or the support bundle) in your ticket turns a multi-day debugging cycle into a single round-trip.
Couldn't fix it with the cards above?
Attach the support bundle (~/alpon-support.tar.gz) along with the
observed LED state and the device serial number. Typical first response in one
business day.