AkiraOS is a Zephyr-based embedded OS that runs sandboxed WebAssembly applications on microcontrollers and lets users deploy and update firmware OTA without reflashing. In other words, it’s similar to Docker containers, but for microcontrollers.

The open-source embedded platform separates the OS from the application. That means the firmware stays stable, while apps are independent .wasm binaries deployable over-the-air without touching the OS, and portable so a single binary works on ESP32-S3, nRF5x, or STM32 MCU boards.

AkiraOS highlights:

• User space
  • Up to 8 wasm apps can be installed
  • Up to two apps can run at the same time
  • Footprint: 50KB to 200KB per app
• Akiraz runtime – Custom WASM runtime
  • App Manager
  • UI Framework with 32 widgets
  • Shell/console
  • 18 API modules
  • WebAssembly Micro Runtime (WAMR) – Two options: Interpreter or Ahead-Of-Time (AOT) compilation with 10 to 50x higher performance
• RTOS – Zephyr RTOS
  • Scheduler
  • Network stack
    • HTTP for OTA updates
    • Bluetooth LE for AkiraMesh
  • Drivers
  • LittleFS file system
• Benefits
  • Update apps in the field without a firmware flash cycle
  • No recompilation for the apps – One binary runs on ESP32-S3, nRF5x, or STM32 — no recompile
  • Device stays up even if a bad app crashes
  • Every app gets only the hardware access it explicitly requested

AkiraOS is supported on the following hardware targets:

• Tier 1 support (best)
  • Espressif Systems ESP32 series
    • ESP32-S3 (LX7) and ESP32-H2/ESP32-C6 (RISC-V)
    • ESP32-S3-DevKitM-1 recommended, and upcoming open-source hardware AkiraConsole V3 on Crowd Supply (we’ll check it once launched with full info including price)
  • Native_sim for fast iteration on x86-64 machines with no MCU hardware needed
• Tier 2 support
  • Nordic nRF54L15 Arm Cortex-M33 MCU with BLE 5.4
  • STM32 Arm Cortex-M microcontrollers
    • B-U585I-IOT02A Discovery kit (STM32U585AI)
    • STEVAL-STWINBX1 SensorTile Wireless Industrial Node Development Kit (STM32U585AI)
    • STMH753/H723 platforms

You’ll find the source and instructions to get started on GitHub, and a separate repo contains an SDK to develop apps for AkiraOS. More details can be found in the documentation on the project’s website, where I notice a mobile app is being developed to retrieve device info & status, manage apps, trigger an OTA firmware update, access the shell/terminal, and browse files on the target device. There’s also a management web interface accessible through WiFi or USB.

The post Docker for Microcontrollers? AkiraOS combines Zephyr RTOS with WebAssembly (WASM) applications appeared first on CNX Software - Embedded Systems News.

Thundercomm has launched the TurboX C7790 development kit, a compact edge AI platform built around the Qualcomm Dragonwing Q-7790 (CQ7790S) processor. The kit offers up to 24 TOPS of AI performance and support for both Android and Linux operating systems.

The development kit pairs a TurboX C7790 system-on-module equipped with 12GB LPDDR5X RAM and 128GB flash with a carrier board featuring a wide range of interfaces from dual GbE to HDMI, USB-C, and MIPI DSI display interfaces. Target applications include AI cameras, video conferencing, robotics, and edge AI gateways.

TurboX C7790 Development Kit specifications:

• System-on-Module – TurboX C7790
  • SoC – Qualcomm Dragonwing Q-7790 (CQ7790S) 4nm processor
    • CPU – Octa-core Kryo processor up to 2.8 GHz
      • 1x Gold+ core @ 2.8 GHz
      • 4x Gold cores @ 2.4 GHz
      • 3x Silver cores @ 1.8 GHz
    • GPU – Qualcomm Adreno 722 GPU up to 1.15 GHz
    • VPU
      • Video Decode – 4K120 H.264/H.265/AV1
      • Video Encode – 4K60 formats: H.264/H.265
    • Audio DSP (LPASS) – Low-power idle mode with an embedded NPU (eNPU5); always-on subsystem
    • AI Acceleration – 24 TOPS (INT8), NSP3, 8K HMX, 4x HVX
  • Memory – 12GB LPDDR5X
  • Storage 
    • 128GB flash storage (UFS/eMMC/NAND constraints apply)
    • SDC (SD3.0) for SD cards
• Carrier board
  • Storage
    • M.2 M-Key slot for NVMe SSD
    • MicroSD card slot
  • Display
    • 2x HDMI port
    • DisplayPort over USB Type-C
    • MIPI DSI connector (hardware muxed; cannot be used concurrently with HDMI OUT1)
  • Camera
    • 3x 4-lane D-PHY camera connectors (supports triple ISP, 21MP+21MP+21MP)
    • HDMI input port
  •  Audio
    • 3.5mm headphone jack
    • 4x onboard Digital Microphones (DMIC)
    • Digital (I2S) audio via HDMI output
    • Digital (SoC native) audio via HDMI output
    • Digital (I2S) audio via HDMI input
  • Networking
    • 2x 1000M Gigabit Ethernet ports (RJ45)
    • Optional WiFi and Bluetooth via M.2 E-Key socket
  • USB
    • 4x USB 3.0 Type-A connectors
    • USB 3.1 Type-C port
  • Debugging – USB Type-C connector with UART support for debugging
  • Expansion
    • M.2 M-Key socket for NVMe SSD
    • M.2 E-Key socket for Wi-Fi/BT module
    • 40-pin expansion header
  • Misc
    • Fan connector
    • RTC connector
    • FORCE USB Boot button
    • Power On, Volume Up/Down tactile switches
    • Accelerometer + Gyro (A+G sensor)
    • E-Compass
• Power – 12V via DC barrel Jack (5.5 x 2.5mm)
• Dimensions – 160 x 124 x 20.5 mm
• Operating Temperature – -30°C to +75°C

Thundercomm provides Android and Linux BSPs with OTA update capability. The platform also supports AI frameworks, as well as Thundercomm’s Claw AI solutions for running AI models locally.

Full BSP downloads for the TurboX C7790 Development Kit are not publicly available; you must create a Thundercomm account and obtain approval to access them. The public pages only offer basic documents like the Product Brief and Quick Start Guide. But once approved, you can download the Android and Linux BSPs using the TurboX SDK Manager, which also supports building and flashing images. OTA updates are included, but the detailed instructions are in the private SDK documentation. The company also mentions that you need to sign an NDA and other license agreements on Thundercomm’s portal to access 3D models or schematics, while a Qualcomm PKLA agreement is required for the core source code.

The TurboX C7790 Development Kit is available for purchase at $1,499. In the box, you’ll get the main board assembly (with the SOM already mounted), one Sony IMX577 camera module with ribbon cable, two external speakers, a USB 3.0 Type-A to Type-C cable, and a 12V power adapter (note: not included for Japan due to local regulations). More information can be found on the product page and in the press release.

The post Thundercomm TurboX C7790 Android and Linux development kit features Qualcomm Dragonwing Q-7790 Edge AI SoC appeared first on CNX Software - Embedded Systems News.

Elecrow’s ThinkNode G4 is a Wi-Fi HaLow gateway with Ethernet and 2.4 GHz WiFi that allows you to connect compatible devices or extend the range of existing devices when used with two gateways or in an 802.11s Mesh configuration with multiple gateways.

It’s based on a (legacy) Mediatek MT7628N MIPS processor paired with a Quectel FGH100M featuring Morse Micro MM6108 Wi-Fi HaLow SoC introduced in 2022 with a data rate of up to 32.5 Mbps and a range of up to one kilometer with line-of-sight.

ThinkNode G4 specifications:

• Core module – HLK-7628N Module
  • SoC – Mediatek MT7628N MIPS processor @ up to 580 MHz
  • System Memory – 128MB DDR2 RAM
  • Storage – 32 MB SPI flash
  • Connectivity – 2.4 GHz WiFi 4 2×2 MIMO 802.11b/g/n up to 300 Mbps combined
• WiFi HaLow
  • Module – Quectel FGH100M based on Morse Micro MM6108 chipset
  • WiFi protocol – IEEE 802.11ah
  • Frequency band – 850~950 MHz
  • Modulation methods – OFDM, BPSK, QPSK, 16QAM, 64QAM
  • Encryption modes – AES, SHA-256, SHA-384, SHA-512, WPA3
  • Operation modes – AP/ STA / Mesh (802.11s)
  • Physical layer maximum rate – 32.5 Mbps
  • Tx power – 21dBm
  • Range up to 1 to 2 km
• Ethernet – 10/100 Mbps Ethernet RJ45 port
• USB – 1x USB Type-C port
• Misc
  • CFG Button to enter configuration mode, restart gateway, or restore factory settings
  • Power, HaLow, and LINK LEDs
  • 3x holes for wall mounting
  • 2.4 GHz WiFi antenna
  • External WiFi HaLow antenna (tuned for 902-928 MHz)
• Power Supply – 5V/1A via USB-C port
• Dimensions – 75 x 75 x 30mm exclusing antenna
• Weight – 80 grams
• Temperature Range – Operating: -20~70°C; storage: -20~70°C
• Relative Humidity – 10%~90%, non-condensing

The ThinkNode G4 Wi-Fi HaLow gateway ships with a Wi-Fi HaLow antenna (915 MHz), a 2.4GHz antenna for regular 2.4 GHz WiFi, and a USB-C cable for power. Since the antenna is tuned for the 902-928 MHz band, suitable for the Americas, Southeast Asia, Japan, Australia, and a few other countries, but not the 868 MHz band notably used in Europe, you may have to replace it depending on your country.

We’re not told anything about the software, but I saw a screenshot with Linux 5.15 and “OpenWrt 1.1”, meaning it’s probably a fork rather than OpenWrt with Linux 6.12, despite the Mediatek MT7628N being supported in upstream OpenWrt. I suspect they just use the OpenWrt fork from Morse Micro like most MM WiFi HaLow gateways. We also have more information about the LuCi-based user interface through the user manual. The gateway can be configured as a client (STA) or a gateway (AP) in standard Wi-Fi HaLow mode, or as a Mesh Gate/Point when using the 802.11s standard.

Target applications include remote camera monitoring, industrial automation control, asset management and tracking, Smart Home, Smart City, Wi-Fi/Ethernet/Wi-Fi-HaLow extension and bridging, and rural/remote internet access.

This type of solution is not exactly new since we’ve covered products such as a WiFi HaLow extension kit and AsiaRF AP7688-WHM WiFi HaLow IoT gateways at least since 2021. More recently, we noted the HalowLink 2 with more powerful hardware (Mediatek MT7621 dual-core @ 880MHz, 256MB DDR3,  32MB NAND Flash, MM8108 chipset), and the main advantage of the ThinkNode G4 is its affordability.

Elecrow sells the ThinkNode G4 for $54 plus shipping and taxes on its online store, and it may soon become available on the company’s AliExpress store.

The post $54 ThinkNode G4 Wi-Fi HaLow gateway extends the range of Ethernet or 2.4 GHz WiFi 4 devices appeared first on CNX Software - Embedded Systems News.

Erqos EQSP32CE is a DIN rail-mountable industrial IoT PLC based on an ESP32-S3 WiFi and Bluetooth SoC and offering Ethernet, RS-485, RS-232, and CAN bus industrial communication interfaces.

The IIoT logic controller also features several protected digital (16x) and analog (8x) inputs, eight current inputs, eight “special mode” analog inputs, and sixteen digital outputs with PWM support. A USB-C port is used for firmware flashing and monitoring, and the PLC takes a wide 7V – 26V DC input voltage and outputs 5V/1A for I/O expansion modules.

Erqos EQSP32CE specifications:

• SoC – Espressif ESP32-S3 dual-core LX7 processor @ 240 MHz with 8MB Flash, 512kB RAM, wireless connectivity (so probably ESP32-S3FN8)
• Communication interfaces
  • 10/100 Mbps Ethernet RJ45 port
  • Wi-Fi + Bluetooth Low Energy (on ESP32-S3) with internal antenna
  • RS232 (protected) and RS485 half-duplex (protected) with support for Modbus RTU, DMX512, and custom serial protocols
  • CAN Bus (protected)
• USB – 1x USB-C power for programming and monitoring
• I/O interfaces (software defined)
  • Digital Inputs – Up to 16x protected, configurable 24V-tolerant digital inputs (all terminals supported)
  • Analog Inputs – Up to 8x protected, configurable analog inputs (0–10V) on terminals 1–8
  • Current Inputs – Up to 8x software-selectable 4–20mA inputs on terminals 1–8
  • Special Analog Modes – Relative (RAIN) and Temperature (TIN) analog input modes on terminals 1–8
  • Outputs – Up to 16x protected, configurable solid-state pull-down switches (PWM-capable)
• Misc
  • Power, Status, WiFi, Bluetooth, and Ethernet LEDs
  • Buzzer for system alerts
• Supply Voltage
  • Input – 7V – 26V DC
  • Output – 5V DC @ 1A (regulated auxiliary output)
• Dimensions – 90 x 71 x 57 mm
• Weight – 175 grams
• Temperature Range – -20°C to +55°C
• IP Rating – IP20
• Certification and Compliance (CE Compliance under Directive 2014/53/EU (RED):
  • Safety (Electrical) – EN IEC 62368-1:2020 + A11:2020
  • Outdoor Safety – EN 62368-1 Annex Y
  • RF Exposure / Health – EN IEC 62311:2020, EN 50566:2017, EN 62479:2010, EN 50663:2017, EN 50665
  • EMC (General/IT) – EN 55032, EN 55024 (or EN 55035)
  • EMC (Wireless) – ETSI EN 301 489-1 V2.2.3, ETSI EN 301 489-17 V3.2.4
  • Industrial EMC – ETSI EN 303 446-1 V1.2.1 (references EN 61000-6-2, EN 61000-6-4, and IT EMC standards)
  • Radio (WiFi/BLE) – ETSI EN 300 328 V2.2.2 (Full + Spurious only), ETSI EN 300 440 V2.2.1 (Full + Spurious only), ETSI EN 301 893 V2.1.1 (Spurious only)
  • Environmental – Directive 2011/65/EU (RoHS)

The EQSP32CE PLC can be connected to a range of expansion modules from Erqos, including digital I/O, analog input, analog output, thermocouple, PT100/RTD, pH, relay, and
GSM/GPRS cellular modules. You’ll find more technical details on the documentation website for the ESQP32 PLC family

Like other members of the EQSP32 family, the EQSP32CE offers a software-defined I/O architecture, and all 16 terminals can operate as digital inputs or solid-state outputs, while the first eight also support analog modes. The device is designed to be programmed in C/C++ via the Arduino IDE or VS Code using the EQSP32 Arduino library, and advanced users can also rely on the ESP-IDF framework. The company also introduced the EQ-AI generative AI development system, powered by ChatGPT, enabling engineers to describe behavior in plain language or simple step-based logic to automatically produce C++ code for the EQSP32CE, handling MQTT topics, JSON payloads, network setup, reconnection logic, diagnostics, and REST endpoints.

While it’s the first time we cover an ESP32-S3 PLC from Erqos, this type of ESP32-based IIoT controller has been around for years from companies such as TECHBASE, NORVI/Iconic Devices, and Industrial Shields. The EQSP32CE just adds another option.

The EQSP32CE ESP32-S3 industrial IoT PLC is available now for $185/155 Euros on the Erqos online store in single-unit quantities, and volume pricing is available for OEM customers. The company is based in Greece, but they also have warehouses in the US and Hong Kong.

The post Erqos EQSP32CE – An industrial IoT ESP32-S3 PLC with Ethernet, RS232, RS485, CAN Bus, DIN Rail support appeared first on CNX Software - Embedded Systems News.

Designed by StillFixing in Normandy (France), the EKOS is a local-first, low-power ePaper dashboard built around an ESP32-S3 SoC. It operates without any cloud dependency, subscriptions, or external accounts, offering full privacy, faster response times, and direct local control.

The device comes in two variants: the EKOS Pure is a minimalist, non-touch version with two physical buttons for basic control, and the EKOS Sense adds a capacitive touch layer for smart home control, such as toggling devices, triggering scenes, or managing tasks. Both models feature a repair-friendly design with no adhesives, using four screws for assembly and a user-replaceable lithium-polymer battery. Additionally, the dashboard can be updated entirely over your home network without needing the internet. You can push data to the screen using a local API from a phone or PC, link it straight to Home Assistant, or use it to control other ESPHome devices around your house.

EKOS specifications:

• Wireless Module – ESP32-S3-WROOM-1
  • SoC – Espressif Systems ESP32-S3
    • CPU – Dual-core Tensilica LX7 up to 240 MHz with vector extension for AI/ML workloads
    • RAM – 512KB SRAM; 8MB PSRAM
    • Storage – 16MB flash
    • Wireless – WiFi 4 and Bluetooth LE 5
  • Antenna – PCB antenna
• Storage Expansion – Onboard MicroSD card slot for holding localized UI assets, fonts, telemetry logs, and optional backup firmware partitions.
• Display – 7.5-inch reflective ePaper Display with an 800×480 resolution; optional capacitive touch layer (EKOS Sense variant only)
• Sensor – Bosch Sensortec BME280 sensor for ambient temperature, relative humidity, and barometric pressure
• Expansion – 16-pin exposed Pogo pin connector on the back for future modular hardware accessories and custom maker extensions
• Power
  • 5V via USB-C port
  • 4,000 mAh Lithium-Polymer (Li-Po) cell connected via JST connector
  • Internal charging IC management circuitry
  • System-level light-sleep and deep-sleep scheduling to maximize battery autonomy up to several weeks per charge cycle
• Dimensions – TBD
• Enclosure
  • CNC-machined solid Normandy Oak wood chassis treated with natural VOC-free oils
  • 3mm structural matte-anodized aluminum front/rear faceplates

In terms of software support, the firmware driving EKOS is built on the official ESP-IDF v5. x framework rather than Arduino, for better power optimization and low-level peripheral control. The user interface and widget rendering are handled by the LVGL graphics library, optimized for clean 1-bit and grayscale output on the ePaper display.

The system follows a fully local, non-cloud architecture, using an asynchronous event loop to manage sleep cycles, power states, and display updates. It also runs an internal HTTP server exposing a documented local REST API (OpenAPI/Swagger), allowing developers to send updates, modify configurations, or trigger partial screen refreshes via simple curl commands without requiring a mobile app.

Widgets are structured using JSON templates that define layout, parameters, and update intervals, and are rendered via C-based implementations in LVGL. For home automation, built-in widgets can communicate directly with platforms like Home Assistant via local HTTP or webhook calls. A companion mobile app for iOS and Android is in development for visual widget management, but remains entirely optional.

The project is fully open-source, and all the firmware source code, schematics (PDF), Gerber files, and BoM have been released in the firmware repository on GitHub under the GNU General Public License v3.0 (GPL-3.0), and the KiCAD schematics and PCB layout, 3D CAD models, and STEP files will be released under the CERN-OHL-S-2.0 license in the future. Documentation, API references, and user guides are published separately under the Creative Commons CC-BY-SA 4.0 license, ensuring full transparency and allowing users to study, modify, and redistribute every aspect of the project.

Over the years, we have covered many low-power ePaper displays that work perfectly as smart home dashboards. If you are looking for alternative options, you can check out the paper.ink 4.2-inch ESP32-based e-Paper display, the large Inkplate 13SPECTRA, or the M5Paper ESP32 IoT development kit. More recent alternatives include the M5Stack PaperColor ESP32-S3 devkit and the reTerminal E1001/E1002, for which we have made a detailed review. However, the  EKOS’s touchscreen function is harder to find due to the long refresh time of some e-Paper displays.

The EKOS project is live on Kickstarter with a funding goal of $17,410, running until early July 2026. Early-bird “First Batch” pricing starts at €169 (~$197) for the non-touch EKOS Pure and €189 (~$220) for the touch-enabled EKOS Sense, increasing to €199 and €219 after those tiers sell out. Multi-unit bundles are also available, including the Harmony Set (two EKOS Pure units for €339), the Synergy Set (one Pure and one Sense for €359), and the Ecosystem Set (two EKOS Sense units for €379). Manufacturing and assembly will take place in Normandy, France, with the first units expected to ship in December 2026. Final shipping costs and taxes will be calculated later through a pledge manager.

The post EKOS – An ESP32-S3 ePaper dashboard housed in an oak-aluminum enclosure (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

YPlasma has showcased its ultrathin solid-state cooling module based on dielectric barrier discharge (DBD) plasma actuators on NVIDIA Jetson Orin Nano at Computex 2026.

Ultrathin solid-state cooling solutions replacing thicker and noisier mechanical fans have been demonstrated on consumer hardware in the past, with solutions such as xMEMS µCooling fan-on-a-chip for SSDs and Frore Systems Airjet Mini and Airjet Pro used in laptops and mini PCs. Both solutions generate tiny vibrations to create an airflow, but YPlasma’s DBD plasma cooling module generates ionic wind to cool the device. This forces convection without moving parts, audible noise, vibration, or dust ingestion paths. We’ll explain more about the technology below.

The company validated the following on the NVIDIA Jetson Orin Nano:

• Thermal range – 7 W to 25 W, the full operating range of the Jetson Orin Nano family, including Super Mode (25W), with a steady state reached in 10 minutes.
• Form factor – 200 micrometer flexible actuator with an 87 × 60 × 2 mm conductive plate, fitting within a 6 mm Z-height, below standard USB port height.
• Drive conditions: 16 kVpp at 50 Hz; actuator draws below 1 W. Production-grade portable driver targets IEC 62368-1 compliant operation under 2 W end to end.
• Acoustics – Solid-state operation, enabling sub-20 dBA operation
• Mechanical robustness – Zero moving parts, no measurable vibration, no dust intake path. Built for sealed enclosure thermal management in IP-rated industrial, automotive, and outdoor edge deployments.

What is ionic wind solid-state cooling exactly? YPlasma explains it in detail on its website. Here’s a summary.

Ionic wind, also known as electrohydrodynamic (EHD) flow or the corona wind, is the bulk movement of neutral air induced by collisions with charged particles accelerated through an electric field.  So instead of moving air with spinning blades or vibrating membranes, an ionic wind cooling module moves air with electric fields at high voltage (but low current, so it’s safe). The processor unfolding in three steps:

• Corona discharge and ionization: a sharp emitter electrode (a wire, needle, or thin exposed strip) is held at a high voltage, typically between 3 and 15 kV, against a grounded collector, and generates a cloud of positive (or negative) ions.
• Ion drift and momentum transfer: these ions accelerate and eventually collide with the vast majority of molecules — which remain neutral — transferring momentum on every collision.
• Bulk flow and convective heat transfer: the cumulative momentum of trillions of collisions per second drags the surrounding neutral air into a coherent jet that sweeps across nearby surfaces

The force per unit volume in EHD flow can be calculated with the formula F = ρ_q × E, where ρ_q is the local space-charge density and E is the electric field.

Three types of ionic wind devices have been implemented over time: Wire-to-Plate (Corona Wind), Needle-to-Ring, and DBD Plasma Actuators. YPlasma focuses on the latter, also the most advanced of the three. Here’s a short description:

Two electrodes are separated by a thin dielectric layer (typically Kapton, ceramic, or glass), with one electrode exposed to the air and the other buried beneath the dielectric. When driven by a high-voltage AC waveform, the exposed electrode ignites a stable, low-temperature surface discharge along the dielectric, generating a wall-jet of ionic wind tangent to the surface. DBD actuators are less than 1 mm thick, consume only 1–5 W, eliminate the spark-over risk of bare-electrode designs thanks to the dielectric barrier, and can be printed onto flexible films and conformed to almost any surface.

The table below compares ionic wind DBD actuators to mechanical fans.

What’s probably missing is the price, and while heatsinks and mechanical fans are cheap, the solid-state cooling solutions we’ve seen so far are not. Previous solutions appear to focus on consumer devices, while YPlasma targets embedded and robotics applications with the NVIDIA Jetson Orin Nano, where there may be more flexibility with regards to pricing, especially once space constraints are taken into account. Having said that, the company also mentions consumer electronics, data centers, automotive and power electronics, medical devices, and aerospace and defense as other potential applications.

Additional information may be found in the announcement.

Thanks to TLS for the tip.

The post NVIDIA Jetson gets fanless solid-state cooling module appeared first on CNX Software - Embedded Systems News.

Brisk4t’s “Tossed The TV — Kept The Remote” (TTVKTR) is an open-source firmware project for Raspberry Pi RP2040 USB boards that aims to reduce electronics waste by converting old IR remote controls into presentation clickers.

Most Raspberry Pi RP2040 boards with USB ports should work, but the project highlights the Waveshare RP2040-Zero combined with a standard 38 kHz infrared receiver due to its small size and low price ($4-5). The project also relies on the built-in RGB LED for layer color feedback.

That’s about it for the hardware. It just required some basic soldering of the IR receiver to GPIO 28 (OUT), 5V or 3.3V, and GND pins. Nothing too hard. The WS2812 RGB LED is already connected to GPIO 16. I tried to look for RP2040 USB boards with a built-in IR receiver, but I could not find any.

The firmware receives IR codes from a standard 38 kHz receiver and translates them to USB HID reports based on a JSON configuration stored on the device’s filesystem. But users don’t need to edit the JSON themselves, since a browser-based config tool communicates over Web Serial to let users map buttons, learn IR codes, and arrange layouts in a way similar to VIA or QMK for mechanical keyboards and macropads. It’s what differentiates TTVKTR firmware from other similar attempts such as Adafruit pIRkey and CH32V003-USB-IR-Receiver, which require some manual coding for each remote.

The user interface offers the following features:

• Custom Keybindings UI in a browser for assigning remote buttons to keys, media controls, or custom actions.
• Multiple layers for one remote with the board’s RGB LED showing the layer selection
• Support for multi-step inputs, including modifier combinations, repeated presses, and chained actions.
• Layouts enable a distinct way of telling different IR remotes apart
• Manual configuration editing with JSON editor

Example of JSON file stored in /settings.json on the board’s partition:

{
  "ir": {
    "modeChangeCode": "0xC40387EE",
    "modeCount": 2,
    "receivePin": 28,
    "handleRepeat": true,
    "repeatInitialDelayReports": 5
  },
  "led": {
    "pin": 16,
    "modeColors": ["0xFF0040", "0x0080FF"],
    "brightnessPercent": 10
  },
  "modes": [
    {
      "name": "Layer 1",
      "slots": [
        { "irCode": "0xC40387EE", "type": "consumer", "key": "0xCD" },
        { "irCode": "0x...",      "type": "keyboard",  "key": "0x28" },
        { "irCode": "0x...",      "type": "keyboard",  "key": "0x1D", "mods": "0x01" },
        { "irCode": "0x...",      "type": "mode_switch" },
        { "irCode": "0x...",      "type": "text",  "value": "hello world" },
        { "irCode": "0x...",      "type": "combo", "steps": [
          { "type": "keyboard", "key": "0x04", "mods": "0x01" },
          { "type": "keyboard", "key": "0x4C" }
        ]}
      ]
    }
  ],
  "layouts": [
    {
      "name": "Default Layout",
      "buttons": [
        { "irCode": "0x...", "x": 0, "y": 0 },
        { "irCode": "0x...", "x": 1, "y": 0 }
      ]
    }
  ]
}

Note that you’ll need Chrome or another browser that supports Web Serial, and for instance, Firefox is not supported [Update: recent versions of Firefox now support WebSerial, see comments section].

The source code for the firmware (Arduino/PlatformIO) and web interface (HTML+JavaScript), as well as instructions to build the code and get started, are available on GitHub.

Via Hackster.io

The post Convert old IR remote controls into presentation clickers using an RP2040 USB board and open-source TTVKTR firmware appeared first on CNX Software - Embedded Systems News.

We noticed the Rockchip RK3572 mid-range HMI SoC a couple of months ago, and Forlinx has launched the first system-on-module (FET3572-C SoM) based on the processor, along with a development board (OK3572-C) and BSP (Board Support Package) with a fairly recent Linux 6.12 kernel.

The octa-core Cortex-A73/A53 processor features a 4 TOPS NPU (the same as in the RK3588) and targets HMI applications leveraging Edge AI for consumer electronics, industrial control, edge computing, smart security, and in-vehicle terminals.

Forlinx FET3572-C Rockchip RK3572 system-on-module

Specifications:

• SoC – Rockchip RK3572 or RK3572J
  • Octa-core CPU – 2x Arm Cortex-A73 @ up to 2.2 GHz+ 2x Arm Cortex-A53 @ up to 2.1 GHz + 4x Arm Cortex-A53 @ up to 2.1 GHz
  • GPU – Arm Mali-G310V2 MC1 with support for OpenGL ES 1.1/2.0/3.2, OpenCL 3.0, and Vulkan 1.4
  • VPU
    • Hardware Encoding -H.264, H265, 4K @ 60fps
    • Hardware Decoding – H.264, H.265, VP9, AV1, AVS2, 8K @ 30fps or 4K @ 120fps
  • AI accelerator – 4 TOPS (INT8) NPU with support for INT4/INT8/INT16/FP4/FP8/FP16/BF16
  • Process – 8nm
• System Memory – 2GB, 4GB, or 8GB LPDDR5
• Storage – Up to 64GB eMMC flash
• 4x 100-pin board-to-board connectors (0.4mm pitch, 1.5mm height)
  • Display
    • HDMI 2.1 or eDP 1.3 Tx up to 4Kp60
    • 4-lane MIPI DSI 1.2 up to 2560 × 1600 @ 60 Hz
    • 8-bit RGB parallel LCD up to 1080p60
    • E0Ink EPD support up to 1872 x 1404
  • Camera – 2x MIPI CSI-2, 8/10/12/16-bit DVP
  • Audio
    • Up to 5x SAI, up to 192 kHz sample rate, 16-bit to 32-bit audio resolution
    • Up to 2x S/PDIF Tx
    • 1x S/PDIF Rx
    • 1x PDM up to 8 channels, up to 192 kHz sample rate, 16-bit to 24-bit audio resolution
  • Networking – 2x Gigabit Ethernet interfaces (RGMII/RMII)
  • PCIe/SATA/USB multiplexed interfaces
    • 2x single-lane combo interfaces (PCIe 2.1 / SATA 3.1 / USB 3.0)
    • 1x single-lane combo interface (PCIe 2.1 / SATA 3.1)
    • 2x USB 2.0 OTG
  • DSMC – Double Data Rate Serial Memory Controller for connection to PSRAM, FPGA…
  • Other I/Os
    • 2x SDIO 3.0
    • Up to 5x SPI, 9x I2C, 1x I3C
    • Up to 12x UART with RS-485 mode support
    • Up to 4x CAN and CAN FD
    • Up to 16x  PWM
    • Up to 8x 12-bit ADC (1 Mbps max)
    • DSMC
• Supply Voltage – 5V
• Dimensions – 68 x 50 mm
• Temperature Range
  • FET3572-C SoM: 0°C~+80°C
  • FET3572J-C SoM: -40°C~+85°C

Forlinx provides support for Linux 6.12-based operating systems, including Forlinx Desktop 24.04 (based on Ubuntu 24.04?), Android 16, and Debian 13. So it looks like the Rockchip SDK for the RK3572 SoC is based on a fairly recent kernel that will be supported at least until December 2028.

Hardware resources such as the datasheet, user guide, carrier board schematic and PCB (see specs below), and SoM Pinmux, as well as a software BSP (board support package) with OS images, demos, source code, and user manual, will be provided to customers. Forlinx doesn’t typically host the documentation publicly.

OK3572-C development board

Specifications

• SoM – Forlinx FET3572-C as described above
• Storage
  • MicroSD card slot
  • Optional NVMe SSD via M.2 Key-M socket
• Video Output
  • HDMI 2.1 up to 4096 × 2160 @ 60 Hz
  • MIPI DSI connector up to 2560 × 1600 @ 60 Hz; compatible with Forlinx 7-inch MIPI display (1024 × 600 @ 30 Hz).
• Camera – 4x MIPI CSI connectors: 2x 4-lane (OV13855 camera) or 4x 2-lane (OV5645 camera)
• Audio
  • On-board audio codec
  • 3.5mm headphone+mic jack
  • Speaker connector
• Networking
  • 2x Gigabit Ethernet RJ45 ports
  • Dual-band WiFi 4 and Bluetooth 5.0 via AzureWave AW-CM358SM module
  • Optional 5G/4G LTE cellular via M.2 Key-B socket + Nano SIM card slot
• USB
  • 3x USB 3.0 Type-A host ports
  • USB 2.0 Type-C OTG port for firmware flashing
• Serial – Terminal block with 2x RS485, 2x CAN FD
• Expansion
  • M.2 Key-M (PCIe 2.1 x1) socket
  • PCIe Gen 2.1 x1 slot
  • 20-pin GPIO header with 1x UART, 8x GPIO, 3.3V, 1.8V, GND
  • 6-pin analog input header (3x ADC)
• Misc
  • Power switch
  • Reset and Power buttons
  • ESC, Menu, Volume-, Volume+/Recovery buttons
  • RTC coin cell battery
• Power Supply – 12V via DC jack
• DImensions – 190 x 130 mm

Forlinx shared some details about the Rockchip RK3572, including potential applications like a multi-camera system with dual-display setup that should be useful for security and automotive applications…

… and some numbers for the NPU’s computer vision and generative AI performance. The table below doesn’t compare the results to other platforms. However, it still shows it’s suitable for CNN models such as Yolov5s and Yolov8, and large language models with a limited number of parameters, which can be useful for specialized interactions.

Forlinx didn’t provide pricing and availability information, but I suspect it might not be available right now, as Forlinx is the very first company to announce a Rockchip RK3572 module, and the supported operating systems are still marked as “R&D”. More details can be found on the product page and the announcement.

The post Forlinx launches Rockchip RK3572 system-on-module (SoM) and development board with Linux 6.12 BSP appeared first on CNX Software - Embedded Systems News.

Avalue Technology EPC-WCL is an Intel Core Series 3 Wildcat Lake-powered fanless Edge AI mini PC with up to 40 TOPS of AI performance to handle tasks like real-time inference, vision processing, and speech recognition locally.

It ships with up to 48GB DDR5-6400 memory via a single SO-DIMM slot and offers M.2 expansion for NVMe storage, Wi-Fi 7, and 4G LTE/5G cellular modules. Display outputs include HDMI, DisplayPort, and USB Type-C with DP Alt Mode, supporting up to three displays. The USB Type-C port also supports 45W Power Delivery, enabling power, data, and video over a single cable.

Avalue EPC-WCL specifications:

• SoC – Intel Core Series 3 Wildcat Lake processors with five or six cores, Intel Xe3 Graphics, and up to 40 TOPS of combined AI performance. TDP: 15W
• Memory – Up to 48GB DDR5 6400MHz via 262-pin SO-DIMM socket
• Storage – M.2 Key-M 2280 slot for NVMe SSD
• Display
  • HDMI 2.0b
  • DisplayPort 1.4a
  • USB Type-C with support for DisplayPort Alt Mode
  • Up to 3x independent displays
• Audio
  • 3.5mm Mic-In and Line-Out jacks
  • RealTek ALC888S-VD2-GR audio codec
• Connectivity
  • 2x 2.5 Gbps Ethernet RJ45 ports via Intel I226LM and Intel I226V controllers
  • Optional Wi-Fi 7 / Bluetooth 6.0 module via M.2 Key-E 2230 slot
  • Optional 4G LTE/5G cellular module via M.2 Key-B slot
  • 2x Antenna mounting holes with dust protection covers
• USB
  • USB Type-C port with 45W Power Delivery and DisplayPort Alt mode
  • USB 3.2 Gen 2 x1 Type-A port
  • 2x USB 3.2 Gen 1 x1 Type-A ports
  • 2x USB 2.0 Type-A ports on the front panel
• Serial – 2x COM ports (RS232/422/485) via front-panel DB-9 ports (selectable via BIOS)
• Expansion
  • M.2 Key-M 2280 slot (PCIe Gen4 x2) for NVMe SSD
  • M.2 Key-E 2230 slot for Wi-Fi 7 / Bluetooth 6.0 module
  • M.2 Key-B 2242/3042/3052 slot for secondary storage or LTE/WWAN/GNSS modules (utilizing PCIe x2, USB 3.2, and USB 2.0)
• Security – TPM 2.0
• Misc
  • Power Button w/LED
  • Reset Button
  • Power on/off and Storage Access LED
  • Intel iAMT support
  • Watchdog Timer (H/W Reset, 1–65535 sec/min)
  • Hardware status monitoring (CPU temp, voltages)
  • AMI uEFI BIOS, 256Mbit SPI Flash ROM
• Power
  • +12V to +24V DC wide-range input via a Phoenix connector (the company mentions a “Lockable DC Jack” option)
  • AT/ATX power modes (AT set as default)
  • 120W AC-to-DC power adapter included
• Dimensions – 177 x 126 x 57 mm (Ultra Slim Fanless Chassis)
• Temperature – Operating: -10°C to 60°C with 0.5m/s airflow; storage: -30°C to 70°C
• Humidity – Up to 95 % Relative Humidity, non-condensing @ 40°C
• Shock – 10G, IEC 60068-2-27, half-sine, 11ms (with SSD)
• Vibration – 1.5Grms, IEC 60068-2-64, random, 10~500Hz (with SSD)
• Drop – ISTA 2A, IEC-60068-2-32 Test (1 corner, 3 edges, 6 faces)
• Mounting – Table Stand (default), Wall Mounting Kit (included); DIN-Rail and VESA mounts are optional

With these specifications, the system is clearly designed for industrial use, and its fanless chassis helps improve resistance to shock and vibration. The company says it supports Windows 11 (64-bit) and Linux, but no download links were available at the time of writing.

Just last month, we wrote about the Arbor ARES-2100 as one of the first fanless industrial box PC with the Intel Core Series 3 “Wildcat Lake” processor, and the Avalue system adds another option with a slightly different feature set, and many more are coming.

Avalue has not released any pricing information for the EPC-WCL embedded system, as is typical for this type of industrial hardware. However, the product is listed as “Coming Soon,” and you can request a quote directly from Avalue. More information is available on the product page and press release.

The post Avalue EPC-WCL fanless Edge AI mini PC features Intel Core Series 3 Wildcat Lake processor appeared first on CNX Software - Embedded Systems News.

ModRetro has announced the M64, an open-source Nintendo 64-compatible console powered by an AMD Artix UltraScale+ FPGA, designed to play original cartridges using hardware-level emulation instead of software.

The M64’s reliance on an AMD Artix FPGA enables accurate and responsive gameplay, and the console supports original game cartridges and controllers, while also adding modern connectivity features such as HDMI, WiFi, Bluetooth, and USB-C.

ModRetro M64 specifications:

• FPGA – AMD Artix UltraScale+ (16nm architecture)
• Memory – PSRAM
• Storage – MicroSD card slot (for firmware updates and potential homebrew applications)
• Media Interfaces – Dedicated Nintendo 64 physical cartridge slot for original retro media preservation and play.
• Display – HDMI port (upscaling for modern TVs, supporting custom video filters like Scan Line and CRT profiles)
• Audio – Digital audio output via HDMI
• Connectivity
  • Bluetooth with advanced “SCI” wireless mode (ModRetro’s own low-latency protocol)
  • Wi-Fi support for firmware upgrade
• USB – 3x USB-C ports for power, accessories, and controllers
• Expansion – 4x front-facing Nintendo 64 controller ports supporting original wired controllers, accessories like Rumble/Transfer Paks, and modern third-party add-ons
• Misc
  • Power Button
  • Menu Dial
  • Eject Button for cartridges
  • LED lighting for console port and cartridges
  • Compatible with original N64 controllers, NSO N64 controllers, 8BitDo gamepads, and the included M64 Pro controller
• Power – Dedicated USB-C port on the back specifically for power
• Dimensions – TBD
• Enclosure –  Translucent shell without adhesives (designed for easy repairability)

ModRetro mentions that they use four fast PSRAM chips instead of DDR memory to get more consistent, low-latency performance, which is important for accurate N64 emulation, even though DDR can be faster in peak speed. The company is also developing a CRT AV adapter that will automatically detect 240p/480i signals and output composite, component, and S-Video for use with vintage CRT monitors.

The company also offers the M64 Pro Controller, which takes on the classic N64 “Trident” design and connects via Bluetooth or wired USB. The company also upgraded the joystick with a modern TMR (Thumbstick Magnetic Resonance) analog one to solve the drifting issues of the original N64 gamepad. It also features a ceramic-coated aluminum backshell and swap-in AA or rechargeable battery options.

The core logic running on the AMD FPGA is built upon the open-source N64 MiSTer core, originally developed by Robert Peip (FPGAzumSpass). According to ModRetro CEO Torin Herndon, the company is actively adapting and improving this code to fully utilize the faster fabric and 6-input LUT architecture of the AMD UltraScale+ platform. The company has promised to share its improved FPGA core code with the community after launch to help improve accuracy and preserve the original hardware.

Previously, we have written about FPGA-based retro gaming consoles, including the SuperStation ONE home console built around an Intel Cyclone V SoC, and the Roshambo Retro Gaming Console DIY Kit based on the Rock64 or RockPro64 Board. We have also written about handhelds like the open-source Game Bub handheld, which relies on an AMD Xilinx Artix-7 chip to interface with original Nintendo cartridges.

The ModRetro M64 will launch on July 28, 2026, priced at $229.99 with one controller included, while extra controllers cost $89.99. Alongside the console, ModRetro will also release new indie games on physical N64 cartridges, including Xeno Crisis, Extreme-G: Turbo Fusion, and Buck Bumble. You can express your interest by signing up on the ModRetro website, and find more details on AMD’s blog.

The post ModRetro M64 – An AMD Artix UltraScale+ FPGA based open-source Nintendo 64-compatible console with original cartridge support appeared first on CNX Software - Embedded Systems News.

GL.iNet Comet Q (GL-RMQ1) is a USB-C KVM device designed to remotely control smartphones, tablets, and laptops via a single USB Type-C cable and web browser.

The hardware is based on a dual-core Arm processor paired with 512MB of RAM and a 512MB NAND flash, integrates a 1.8-inch touchscreen LCD for control and information display (e.g., IP address), a short USB-C cable to connect to the target, and a USB-C port for optionally charging the target while it is being controlled.

Comet Q specifications:

• SoC – Unnamed dual-core Arm Cortex-A53 processor
• System Memory – 512MB LPDDR4
• Storage – 512 MB NAND flash
• Display – 1.8-inch touchscreen color LCD
• Video
  • Resolution and frame rate – 2K QHD @ 60 FPS
  • Latency – About 80 ms
• Wireless
  • Wi-Fi 6 IEEE 802.11 a/b/g/n/ac/ax
    • 5 GHz: 286 Mbps
    • 2.4 GHz: 286 Mbps
  • Internal Wi-Fi antenna
• Interface
  • USB Type-C port with power passthrough
  • USB Type-C cable with support for DisplayPort Alt Mode
• Power Input – 5V/3A via USB-C PD
• Dimensions – 70 mm Φ x 22 mm
• Weight – TBD

The system runs a Linux operating system built with Buildroot. There’s no need for drivers, as users can simply control a target (smartphone, tablet, or laptop) with USB-C DP support through a web browser on the host over WiFi. Alternatively, you can also install the GLKVM app on an Android or iOS device, as well as Windows and macOS computers for easier remote access over the Internet. Tailscale & ZeroTier are also supported for secure control over VPN through the web browser without opening firewall ports. You can check out our review of the GLKVM app with the GL.iNet Comet device to get a feel for the user interface and features.

Supported target devices include:

• iPhones: iPhone 15 and newer (excluding iPhone 16e, 17e and iPhone Air)
• iPads: All iPad Pro (2018 and newer), iPad Air (4th generation and newer), iPad mini (6th generation and newer), and iPad (10th generation and newer)
• Android phones (much longer list, some Samsung models provided as examples): Samsung Galaxy S8 and newer, Galaxy Note 8 and newer, Galaxy Z Fold/Flip series, and Galaxy Tab S series
• Laptops & mini PCs: Most models with a USB-C port that supports video output, including MacBooks, Mac mini, and most modern Windows laptops

If you can’t quite think of use cases for controlling a smartphone over KVM, GL.iNet has some ideas:

• Developers can remotely manage test devices, run builds, and debug without being at their desk
• IT teams can maintain and troubleshoot a fleet of mobile devices from a single interface
• Content creators can control a dedicated streaming or recording device from anywhere in the room
• Power users can manage a secondary device without ever picking it up
• Children living away from home can remotely help their parents with everyday tech frustrations

Some use cases could certainly be handled through scrcpy software in combination with VNC or Raspberry Pi Connect remote access software, but it’s not quite as easy as asking someone to connect the “remote access clock” to the USB-C port of their phone or tablet. Besides simplicity, a hardware solution enables debugging of boot issues or is OS agnostic, something not possible with remote access software.

GL.iNet has launched the Comet Q on Kickstarter with a $10,000 funding goal that’s already been easily surpassed. Rewards start at $79 for an Early Bird pledge for the GL-RMQ1 USB-C KVM device, and we’re told the MSRP will be $129.99 after the crowdfunding campaign is over. Note you’ll need to add $15 for worldwide shipping. Rewards are scheduled to start shipping by August 2026.

The post Comet Q USB-C KVM device is made for smartphones, tablets, and laptops (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

Reolink has launched the new OMVI Series of triple-lens AI security cameras, featuring a 4K PTZ motorized camera and a fixed dual-lens system that offers a 24/7 180° panoramic view. The camera also supports AI detection and video search, designed to eliminate the typical blind spots of a standard Pan-Tilt-Zoom (PTZ) camera.

The series includes three models: the OMVI 3i PoE, OMVI 3i WiFi, and OMVI X16 PoE. All of them use the same basic design, combining a fixed dual-lens camera for a 180° panoramic view with a separate PTZ camera in a single housing. This setup allows the camera to monitor a wide area continuously while the PTZ lens focuses on and tracks specific objects.

Reolink OMVI triple-lens security cameras specifications:

The OMNI cameras are supported by the Reolink mobile app for iOS 15.1+ and Android 8.0+, the Reolink Client for Windows 8.1+ and macOS 10.15+, Reolink NVR, and third-party software thanks to ONVIF support.  Data is stored on a microSD card and a subscription is not required, although users can also select cloud backup to the Reolink Cloud.

The OMVI series uses on-device AI to link the panoramic camera with the PTZ module. When the wide-angle lens detects a person, vehicle, or pet, the system automatically triggers the PTZ camera to lock onto and track the subject, adjusting framing in real time (SyncTrack). It also supports manual “Pinpoint” control, where you can click anywhere on the 180° panoramic view and the PTZ camera instantly moves to that location. The AI also enables various event triggers such as line crossing, zone intrusion, and loitering alerts, while reducing false positives from general motion. All video analysis runs locally, including AI-based search using text filters (e.g., by person, vehicle, or attributes), without cloud processing.

The OMVI series is launching in phases. The OMVI 3i PoE is available now for $239.99 on Amazon (after clicking on the 20% discount there) and $299.99 on the official Reolink store. The dual-band OMVI 3i WiFi version will be released on June 17, 2026, priced at $319.99. The higher-end OMVI X16 PoE is expected in Q3 2026, with pricing to be announced later.

The post Reolink OMVI triple-lens AI security cameras combine 4K PTZ camera with dual-lens system with fixed 180° panoramic view appeared first on CNX Software - Embedded Systems News.

Morse Micro MM8108-M20 is a high-power Wi-Fi HaLow module based on the MM8108 SoC and a high-power amplifier delivering up to 28.5 dBm transmit output power, alongside a surface acoustic wave (SAW) filter tuned for the 902-928 MHz band of the North American market.

Other MM8108 modules, such as the Quectel FGH200M Wi-Fi HaLow module, are limited to 26 dBm Tx power, but US and Canada regulations enable higher power transmissions, and by extension longer range. The obvious downside of the MM8108-M20 is that it might be outright illegal or may require a special license in other countries.

Morse Micro MM8108-M20 specifications:

• SoC – Morse Micro MM8108 32-bit RISC-V Host Applications Processor (HAP)
  • Standard – IEEE802.11ah Wi-Fi HaLow
  • Frequency band – 850-950 MHz (Worldwide sub-1 GHz license-exempt range)
  • Operating Modes – Access Point (AP) and Station (STA)
  • Channel Width – 1/2/4/8 MHz
  • Modulation – BPSK, QPSK, 16QAM, 64QAM, 256QAM.
  • Data rate – Up to 43.3 Mbps using 256-QAM modulation at an 8 MHz bandwidth
• Module-specific RF specs
  • High-power amplifier delivering up to 28.5 dBm transmit output power
  • Surface acoustic wave (SAW) filter tuned for the 902-928 MHz band
• Host interfaces – USB 2.0, SDIO 2.0, or SPI
• Security
  • AES
  • SHA-256 / SHA-384 / SHA-512
  • WPA3
  • OWE (Opportunistic Wireless Encryption)
• Supply Voltage – TBD
• Dimensions – 18.5 x 14 mm (LGA package)
• Weight – TBD
• Certifications – FCC (America) and IC (Canada)

Morse Micro will usually let OEMs design modules based on their WiFi HaLow chip, and the MM8108-M20 appears to be the first module from the company itself.

There’s very limited information about it, since the Australian company only published a press release and has yet to provide a product page for the new high-power WiFi HaLow module. Going from 26 dBm to 28.5 dBm should extend the range by about a third. The company expects the module to be used in surveillance cameras (the most common application for WiFi HaLow) and access points.

Thanks to TLS for the tip.

The post Morse Micro MM8108-M20 high-power Wi-Fi HaLow module delivers up to 28.5 dBm Tx output power appeared first on CNX Software - Embedded Systems News.

Wireless-Tag has launched a Kickstarter campaign for the ESP32P4C61-TINY, a compact open-source AIoT development board based on their WT01P461-S1 module, which combines ESP32-P4 (general-purpose) and ESP32-C61 (wireless) RISC-V SoCs.

Like other ESP32-P4 boards, including the ESP32P4C5 Core Board, the M5Stack Stamp-P4, or ESP32-P4-Pi-VIEWE, this one also uses a separate SoC (ESP32-C61) for wireless connectivity. Additionally, it features built-in MIPI CSI and DSI for camera and display support, along with a microSD card slot for storage, making it suitable for AIoT and edge computing applications.

ESP32P4C61-TINY Specifications

• Wireless Module – Wireless-Tag WT01P461-S1 (25 x 25mm)
  • Microcontroller – ESP32-P4
    • MCU
      • Dual-core RISC-V microcontroller @ 360/400 MHz with AI instructions extension and single-precision FPU
      • Single-RISC-V LP (Low-power) MCU core @ up to 40 MHz
    • GPU – 2D Pixel Processing Accelerator (PPA)
    • VPU – H.264 and JPEG codecs support
    • Memory – 768 KB HP L2MEM, 32 KB LP SRAM, 8 KB TCM, 16/32MB PSRAM
    • Storage – 128 KB HP ROM, 16 KB LP ROM
  • Wireless – 2.4 GHz WiFi 6 and Bluetooth 5 via ESP32-C61 SoC
  • Storage – 16MB NOR flash
  • IPEX antenna connector
• Storage Expansion – MicroSD card slot (bottom side)
• Display – MIPI DSI connector (2.8-inch 640×480 display comes with the kit)
• Camera – MIPI CSI connector (camera comes with  the kit)
• USB
  • USB Type-C port for USB/JTAG debugging
  • USB Type-C port for Fast Speed USB
• Expansion – 2x 27-pin headers exposing 36x GPIOs (ESP32-P4) and 10x GPIOs (ESP32-C61)
• Misc – RGB LED, Boot button, Reset button
• Power Supply – 5V/2A via USB-C
• Dimensions – 69 x 33 mm

The company mentions the project is fully open source, but like most Kickstarter campaigns, no specific GitHub repository or detailed documentation is available, and it will likely be released once the campaign ends. So far, the page only provides a blurry schematic, and the board will most likely rely on Espressif’s standard ESP-IDF framework for development. All information should eventually show up on the company’s GitHub account and documentation website.

The ESP32-C61 is supposed to be a cost-down version of the ESP32-C6 without an 802.15.4 radio, a single RISC-V core, and less memory (320KB SRAM), but it’s unclear if the lower price is reflected in this specific board…

The ESP32P4C61-TINY board is live on Kickstarter as part of a kit, with the “Super Early Bird” tier priced at around $40, a 51% discount from the $81.50 MSRP. The package will include the board itself, a 2.8-inch MIPI display, a MIPI camera, an antenna, headers, and a USB-C cable. Other tiers include an “Early Bird” at ~$66 and a “Double Discount” offering two kits for ~$76.  The company states that shipping is expected to start in July 2026, and that taxes and shipping costs are extra.

The post Wireless-Tag ESP32P4C61-TINY board combines ESP32-P4 and ESP32-C61 SoCs (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

Nordic Semiconductor has added AI-assisted development to its wireless IoT microcontroller, with workflows covering the full life cycle from the first prototype to a deployed fleet.

Many developers copy/paste information from LLMs trained on generic data. However, Nordic’s AI solution is specifically trained on the nRF Connect SDK documentation and nRF Cloud data and integrates with a developer’s favorite IDE. It also connects to Claude Code, Cursor, GitHub Copilot, or any other LLM at a much lower token cost thanks to the specialized model.

The company says it’s based on an implementation of the Model Context Protocol (MCP), where the Nordic MCP servers give AI assistants access to validated sources from Nordic, including SDK documentation, API references, device configurations, and the customer’s field data from nRF Cloud.

Highlights of Nordic’s AI-assisted development

• Connected to nRF Connect SDK documentation and nRF Cloud data
• Integrates with AI assistants such as Claude Code, Cursor, GitHub Copilot, or any other
• Designed to assist, not replace, developers
• Covers the full development lifecycle, from prototyping to fleet management

Nordic explains the AI agent can be especially useful to automate tedious tasks, speed up prototyping, and ease debugging, for instance, when migrating between SDK versions, for custom board bring-up, or diagnosing a crash on a deployed device.

Several video examples are provided, including AI-assisted migration, finding faulty devices in your fleet, keeping AI costs down, troubleshooting user-reported errors, DeviceTree and Kconfig generation (see video below), validating release readiness, and adding shell commands.

The important part is to be specific and review the code, since the agent can make mistakes. For example, in the video above, the AI agent added some random peripherals (a button and extra LEDs), and this had to be corrected manually. It’s also clear you have to be an engineer to use these tools, since the prompt needs to be quite specific and technical, or in other words, you can’t just “vibe code” your way to a board bring-up. I still view AI agents as interns or drunk/high senior software engineers; in either case, you can’t just let them do their own thing, and they need supervision…

AI is often seen as a threat to the livelihood of software engineers, but so far, it’s mostly a tool. While some companies have fired software engineers due to AI, I’ve seen several headlines about rehiring due to prohibitive AI costs. Some also argue that the number of software engineers may increase rather than decrease, since cheaper software, and in this case firmware, development costs may eventually lead to higher demand for software. Time will tell.

You can check the examples and learn how to get started on the Nordic’s website.

Thanks to TLS for the tip.

The post Nordic adds AI-assisted development to the nRF Connect SDK and nRF Cloud appeared first on CNX Software - Embedded Systems News.

ASUS Ascent QN10 is the first mini PC powered by a Qualcomm Snapdragon X2 Elite 18-core Armv9 processor with 80 TOPS of AI performance for Windows CoPilot+ PCs.

The powerful Arm computer ships with up to 32GB LPDDR5 memory and a 512GB to 2TB NVMe SSD. It supports up to four video outputs through one HDMI and three USB4 (40 Gbps) ports, and offers 2.5GbE and WiFi 7 networking, an audio jack, and a few extra USB 3.2/2.0 ports.

ASUS Ascent QN10 specifications:

•  SoC – Qualcomm Snapdragon X2 Elite (Glymur 8480B / X2E-88-100)
  • CPU – 18x Armv9 cores with 12 Prime cores up to 4.7 GHz (single/dual core) / 4.0 GHz (multicore), and 6 Performance cores up to 3.4 GHz
  • GPU – Adreno X2-90 @ 1.70 GHz with support for DirectX 12.2 Ultimate, Vulkan 1.4, OpenCL 3.0
  • VPU
    • Encode: HEVC, AVC: Dual 8K UHD @ 30 FPS, AV1: 8K UHD @ 15 FPS, UHD @ 60 FPS
    • Decode: AV1, HEVC, AVC: Dual 8K @ 60 FPS
    • Concurrent: 8K UHD @ 30 FPS Encode + 8K @ 60 FPS Decode
  • AI accelerators
    • 80 TOPS (INT8) NPU
    • Dual Micro NPU on Qualcomm Sensing Hub
  • Process – 3 nm
  • TDP – 65W
• System Memory – Up to 32GB LPDDR5X-8533 or LPDDR5X-9600
• Storage
  • 512GB, 1TB, or 2TB M.2 NVMe (PCIe 4.0) SSD
  • M.2 2280 PCIe Gen4 NVMe SSD socket (I assume occupied by SSD above)
  • M.2 2280 PCIe Gen5 NVMe SSD socket
• Video Output
  • HDMI 2.1 port
  • DP 2.1 via 3x USB4 port
  • Up to 4x independent 4K monitors
• Audio
  • 3.5mm audio (headphone+mic) jack
  • Qualcomm WCD9378C codec
• Networking
  • 2.5GbE RJ45 port via Realtek controller
  • WiFi 7 and Bluetooth 5.4 via Foxconn NCM820A module
• USB
  • 3x USB 4 Gen2 40Gbps Type-C ports with DP 2.1 and USB PD ( 5V/3A)
  • 1x USB 2.0 Type-A port
  • 3x USB 3.2 Gen2 10Gbps Type-A ports
• Security
  • fTPM 2.0
  • Qualcomm Secure Processing Unit (SPU) that supports Microsoft Pluton.
• Misc – Power button
• Power Supply – 19.5V/9.23A (180W) power adapter
• Dimensions – 130 x 130 x 39.96 mm
• Weight – 720 grams

The ASUS Ascent QN10 comes pre-loaded with Windows 11 Pro Copilot+PC or Windows 11 Home Copilot+PC, and the company mentions access to the Qualcomm AI Hub machine learning repository. The system ships with a 180W power adapter, a power cord, a Safety/Caution/Regulatory insert, and a warranty card.

The price has not been announced, but considering the ASUS Zenbook A16 laptop with a Qualcomm Snapdragon X2 Elite Extreme SoC is selling for around $1,600, I’d assume the Ascent QN10 should go for well over $1,000. The company mostly targets tech enthusiasts/prosumers, AIoT developers & integrators, small and medium businesses, retail, education & research, and smart factories & manufacturing. A few more details may be found on the product page.

Via Qualcomm press release.

The post ASUS Ascent QN10 mini PC is powered by Qualcomm Snapdragon X2 Elite 18-core Armv9 processor appeared first on CNX Software - Embedded Systems News.

NVIDIA just announced the Isaac GR00T Reference Humanoid Robot in a half-baked press release for Computex 2026, with an “available-soon reference workflow” and availability sometime by the end of the year with the Unitree H2 humanoid chassis.

One component of the kit that appears to be available now, albeit in limited quantity, is the Sharpa Wave high-end dexterous robotic hand with 22 degrees of freedom (DoF) and a dynamic tactile array (DTA) on each finger, enabling it to feel objects as lightweight as a butterfly.

Sharpa Wave robotic hand highlights:

• 1:1 scale human form – The palm width to hand length ratio is approximately 0.618, enabling the Wave to manipulate the same tools as humans.
• 22 active Degrees of Freedom – Isomorphic design mirroring the human hand
• Dynamic Tactile Array (DTA) powered by a proprietary neural network-based algorithm and sensing modules to enable detection of tight objects (like a butterfly)
• Maximum active fingertip force – 20 N
• Minimum grasp diameter – 10 mm
• Fingertip position repeatability – ±1 mm
• Control frequency – 500 Hz
• Movement speed across all gestures  – > 4 Hz
• 5x fingers each with a tactile sensor and a torque sensor
• Host interface – Gigabit Ethernet
• Supply Voltage – 18-28V DC
• Power Consumption – 15W (static), up to 180W (transient peak)
• Dimensions – 208 × 90 × 50 mm
• Weight – 1.3 kg
• Durability and protection
  • Press test – Validated 2.5 million times
  • Friction test of fingertips – Verified over 4,000 km
  • Impact test – Hundreds of impacts without damage
  • Shock – 3200 cycles at 30 g acceleration
  • Automatic protective clench within 0.10 seconds
  • 1,000+ hours of continuous operation in a waving finger type of test.

The Sharpa Wave hand is designed for embodied AI research and for integration into robots, and the company provides ROS2 packages, MuJoCo URDF/MJCF simulation assets, and C++ and Python programming languages are supported on macOS and Linux hosts. You’ll find more technical details and instructions to get started on the documentation website.

The best way to understand the performance of the SharpaWave hands is probably to watch the demo below using teleoperation to put a garbage bag, peel an egg, and take a photo with a smartphone.

If instead, you’d like a more technical presentation about the hand and technology, you’ll want to watch the 28-minute video below.

The product page has no price information and instead, interested companies are asked to contact the sales team. However, 36kr Europe interviewed the company and mentioned a $50,000 price tag for the hand, and at the end of 2025, it was only possible to test the hand by visiting their office in Shanghai…  This type of hand is made of tiny gears, which reminds me of the design of expensive Swiss watches, but we can only hope prices will come down, because with two hands in the Isaac GR00T Reference Humanoid Robot, we would expect a full kit for $150,000 to $200,000… At least I now understand why a $699 robotic hand is considered mid-range…

The post Sharpa Wave is a high-end dexterous robotic hand with 22 DoF, high-sensitivity dynamic tactile array appeared first on CNX Software - Embedded Systems News.

Sixfab has launched the AI HAT+ for Raspberry Pi 5, a PCIe HAT+ based on the DEEPX DX-M1 AI accelerator, which we also found in the DEEPX DX-AIPlayer, Mini DX-M1 SoM, and ALPON X5.

Unlike the M.2 module used in the ALPON X5, the AI HAT+ has the accelerator soldered directly to the board. It connects to the Pi 5 via the PCIe FFC cable and draws power from the 40-pin header. The board is also available in 13 TOPS and 25 TOPS versions and is designed to run vision AI tasks such as object detection and segmentation locally on the Pi 5.

Sixfab AI HAT+ specifications:

• Supported SBC – Raspberry Pi 5
• AI Accelerator (one or the other):
  • DEEPX DX-M1M with up to 25 TOPS (INT8), 1 GB LPDDR4X NPU memory
  • DEEPX DX-M1ML with up to 13 TOPS (INT8), 512 MB LPDDR4X NPU memory
• Host Interface – PCIe Gen 3 x1 via 16-pin FFC cable
• Misc – Passive cooling by default; included 2-pin JST fan connector
• Power Supply
  • Input – 5V / 3A via Pi 5 40-pin GPIO header (no auxiliary connector required); Note: 27W PSU required, and the 15W PSU is insufficient
  • Consumption
    • NPU peak – 2.5–3 W under full inference load
    • NPU idle – ~0.5–1 W
    • Combined Pi 5 + HAT+ – 13–15 W (27W or more PSU recommended)
• Dimensions – 65 x 56.5 mm (Raspberry Pi HAT+ compliant), 6.56 mm height
• Temperature Range – 0 to 70°C (Commercial)
• Certifications – CE, FCC, UKCA, RoHS, REACH (currently in progress)

The board works with Raspberry Pi OS (Trixie). It uses the HAT+ EEPROM for auto-configuration, and setup only requires installing the dxrt-runtime package from Sixfab’s APT repository, which includes the driver and runtime.

You can either use pre-compiled AI models from the Sixfab Model Zoo, such as YOLOv8, MobileNet, and ResNet, or run your own models. Custom models can be exported to ONNX and compiled into DXNN format using the DX-COM tool. The runtime supports both Python and C++. You’ll find more details in the documentation.

In terms of AI power, the Sixfab AI HAT+ is similar to the Raspberry Pi AI HAT+ (Hailo-8), as both are designed for object detection and image processing. But the Sixfab board is not designed for generative AI and cannot run LLMs as it lacks transformer decoder support and has limited on-board memory. The main differences between the two are in their architecture, software stack, and pricing. In comparison, the newer AI HAT+ 2 (Hailo-10H) is designed for both computer vision and generative AI workloads, with higher performance (up to 40 TOPS) and 8GB of dedicated memory to support LLM and VLM applications. SifFab says that “LLMs are on the DEEPX silicon roadmap” and the company will support them as the silicon enables, but no dates were provided.

The Sixfab AI HAT+ for Raspberry Pi 5 with DX-M1 AI accelerator is available now on the Sixfab store: the 13 TOPS (DX-M1ML) variant is priced at $63, while the 25 TOPS (DX-M1M) variant is priced at $90. The company is also working on the Edge AI Expansion Board for Raspberry Pi 5, which combines AI acceleration with NVMe SSD storage and LTE/5G cellular connectivity on a single board, but details are limited at this stage.

The post Sixfab AI HAT+ for Raspberry Pi 5 integrates DEEPX DX-M1 AI accelerator appeared first on CNX Software - Embedded Systems News.

Back in November last year, we covered the launch of the NuMicro M55M1 MCU from Nuvoton, which combines an Arm Cortex-M55 core with an Ethos-U55 microNPU for on-device AI and gesture control. Now, they have released the NuMaker-GestureAI-M55M1, a development module based on that MCU for AI vision-related applications.

This new board integrates the M55M1 MCU with a GC0308 CMOS image sensor, a digital microphone, and a microSD card slot for storing AI models. It is designed for applications such as gesture control, basic vision systems, and touchless interfaces.

NuMaker-GestureAI-M55M1 specifications:

• MCU – Nuvoton NuMicro M55M1R2LJAE
  • CPU – Cortex-M55 MCU @ 220 MHz
  • Memory – 1.5 MB SRAM
  • Storage – 2 MB Flash
  • AI Accelerator – Arm Ethos-U55 micro-NPU @ 220 MHz
• Storage – MicroSD card slot (located on the back) for storing AI model files
• Camera – Integrated VGA GC0308 CMOS image sensor
• Audio – On-board digital microphone (DMIC)
• USB – High-Speed USB (HSUSB) Type-C port
• Expansion and Debugging
  • ICE Interface header for debugging
  • UART5 header (used for outputting gesture/detection data)
  • I2C1 header
• Misc – Reset button, user button, user LED
• Power Supply – 5V via USB-C port
• Temperature – -40°C to +105°C (industrial grade)

According to NuVoton’s products page, the NuMaker-GestureAI-M55M1 board ships with firmware for out-of-the-box gesture and person recognition. The pre-loaded firmware can track human presence and recognize over 10 specific hand gestures, including: Call, First, Like, Mute, OK, One, Palm, Peace, Stop, and Three.

The board outputs this inference data via the UART5 interface using a simple hexadecimal-and-decimal packet structure. The payload includes a start marker, the tracked object/gesture ID, the bounding box coordinates, and an end marker.

UART Output Format Example: [AA][ID][x y w h][55][CC]

• [AA]: Start marker (0xAA)
• [ID]: Hexadecimal tracking ID. Person IDs range from 0x00 to 0x7F, while gesture IDs map from 0x80 to 0x8A.
• [x y w h]: Bounding box coordinates (4 decimal values from 0 to 9999).
• [55]: Intermediate marker (0x55)
• [CC]: End marker (0xCC)

The NuMicro M55M1 AI gesture control board also supports Nuvoton’s AI toolchain, allowing developers to run custom machine learning models using frameworks like TensorFlow Lite for Microcontrollers. You’ll find instructions to get started on the product’s page, and the firmware and source code can be found on GitHub.

The NuMaker-GestureAI-M55M1 is available on the Nuvoton Direct Store for $30.00, but at the time of writing, it’s out of stock. More information can be found in the press release.

The post Nuvoton NuMaker-GestureAI-M55M1 module combines Cortex-M55 MCU with GC0308 camera for AI gesture control appeared first on CNX Software - Embedded Systems News.

Waveshare RP2350B-Plus-W is a development board that follows the Raspberry Pi Pico 2 W form factor, but offers 41 GPIOs thanks to the RP2350B MCU, integrates 16 MB of flash, and includes a USB-C port.

So, in several ways, it’s an upgrade over the RP2350A-based official board, which offers only 26 GPIOs via two 13-pin GPIO headers, 4MB of flash, and a micro USB port. Since it’s the same size, where do the extra GPIOs come from? Answer: The company simply added 15 pads to the bottom of the board, not quite as convenient since it’s requires soldering, but it does the job.

Waveshare RP2350B-Plus-W specifications:

• SoC – Raspberry Pi RP2350B
  • CPU
    • Dual-core Arm Cortex-M33 @ 150 MHz with Arm Trustzone
    • Dual-core RISC-V Hazard3 @ 150 MHz
    • Only two cores can be used at any given time
  • Memory – 520 KB on-chip SRAM
  • Package – QFN-80
• Memory – Footprint for PSRAM chip
• Storage – 16 MB on-board QSPI flash
• Wireless – 2.4GHz 802.11n WiFi 4 and Bluetooth 5.2 via Raspberry Pi RM2
• USB – USB 1.1 Type-C host/device connector for power and programming
• Expansion
  • 2x 20-pin GPIO headers with
    • 26x GPIOs
    • 2x UART
    • 2x SPI controllers
    • 2x I2C controllers
    • 22x PWM channels
    • 4x ADC
    • 3x PIO blocks, 12x PIO (Programmable IO) state machines
    • HSTX interface
  • 15 solder pads with up to 15x GPIO, 2x SPI, 2x I2C, 3x ADC, 2x UART (many shared with the 20-pin headers)
• Debugging – SWD debug interface
• Misc
  • Reset and Boot buttons
  • 2x user LEDs
• Power Supply – 5V DC via ME6217C33M5G regulator
• Dimensions – 51 x 21 mm excluding the PCB antenna (same size as Raspberry Pi Pico 2 W). With PCB antenna: 55.92 x 21 mm
• Temperature Range – -20°C to +70°C

Other changes include a footprint to solder a PSRAM chip, a Reset button, and one more LED. The only downside from a technical perspective is the removal of the SWD debug port. The Waveshare RP2350B-Plus-W is also a little longer than the Raspberry Pi Pico 2 W once we take the PCB antenna from the RM2 module into account. It’s programmable with the Pico C/C++ SDK, MicroPython, and more, like other RP2350 boards, and Waveshare provides more technical details and instructions to get started in the wiki.

Somehow, it’s the first board that combines Raspberry Pi Pico 2 W form factor with RP2350B, and the closest competitor is probably the Olimex RP2350-PICO2, although it’s fairly longer. If you don’t care about the form factor, you’ll find many more RP2350B boards for all sorts of applications.

The board was introduced nearly a year ago, and when I first came across it, I skipped it since I assumed it was just a clone with a USB-C port. However, it was brought back to my attention when I thought I had found a new Made-in-Thailand board with the INEX Pico-2350W. It quickly became clear that the Thai edtech company just sourced the board from Waveshare. For better or worse, it’s a trend I’ve seen in Thailand, and which likely occurs in other countries. Some engineers and/or companies have the capabilities to develop their own board, but it takes time and effort to develop boards, and it’s just faster and usually cheaper to get a board from China instead.  So it makes perfect sense to source the hardware from China and develop educational materials for the local market. The long-term repercussions can always be discussed…

If you live in Thailand, you can consider getting the board from INEX, but for everybody else, you’ll find the RP2350B-Plus-W board for $11.42 on AliExpress, $18.23 on Amazon, and $10.99 on the Waveshare shop.

The post Waveshare RP2350B-Plus-W – A Raspberry Pi Pico 2 W-sized board with 41 GPIOs, 16MB flash, USB-C port appeared first on CNX Software - Embedded Systems News.

Radxa started its partnership with Qualcomm last with the Dragon Q6A SBC, but it turns out it was just the start, and the company showcased more Qualcomm SBCs and NAS systems at a Radxa + Qualcomm developer day on May 30, 2026.

The Radxa Q8B SBC will be based on a Qualcomm Snapdragon 8cx Gen3 octa-core SoC and the Q5E SBC on a Dragonwing QCS6690 octa-core Kryo SoC. The company also teased DragonStation and DragonBay NAS systems, and a 2026 roadmap features a total of 22 Qualcomm systems made by Radxa.

Radxa Dragon Q8B

Dragon Q8B specifications:

• SoC – Qualcomm Snapdragon 8cx Gen 3 compute platform
  • Octa-core CPU – 4x 3.0 GHz Kryo Prime cores, 4x 2.4 GHz Kryo Efficiency Cores
  • GPU – Adreno 690 GPU with DirectX 12 (DX12) API support
  • DSP – Qualcomm Hexagon Processor, Qualcomm Sensing Hub
  • AI – Qualcomm Neural Processing Engine SDK support for AI (up to 29+ TOPS)
• System Memory – Up to 32GB LPDDR4x RAM up to 4266 MT/s
• Storage
  • MicroSD card slot
  • UFS 3.1 module connector
  • Up to 2x NVMe SSDs via M.2 Key-M PCIe Gen3 sockets (See Expansion section)
• Video Output
  • HDMI 2.1 port
  • 2x DisplayPort 1.4b via USB-C ports up to 4Kp120
• Audio
  • 3.5mm audio jack
  • Microphone connector
• Connectivity
  • 2x 2.5GbE RJ45 ports
  • Optional WiFi and Bluetooth via M.2 Key-E socket
• USB
  • 2x USB 3.2 Gen 2 Type-C ports with DP Alt. mode
  • 2x USB 3.2 Gen 2 Type-A ports
  • 2x USB 2.0 Type-A ports
• Expansion
  • M.2 Key-M 2280 (PCIe Gen3 x4) socket
  • M.2 Key-M 2280 (PCIe Gen3 x2) socket
  • M.2 Key-E 2230 socket for WiFi/Bluetooth
  • 40-pin color-coded GPIO header with UART, I2C, SPI…
  • 16-pin PCIe Gen3.0 x1 FFC connector compatible with Raspberry Pi FFC connector
• Misc
  • Power button
  • RTC connector
• Power Supply – USB-C PD connector
• Dimensions – 100 x 75 mm

We’re not told module as OS support, except Linux with “open-source toolchains” will be the focus here. The board itself targets Edge AI, Arm NAS, and intelligent gateways.

Radxa Dragon Q5E SBC

Dragon Q5E specifications:

• SoC – Qualcomm Dragonwing QCS6690
  • CPU – Octa-core Qualcomm Kryo CPU 7-series (1x Kryo Prime core @ 2.0 GHz, 3x Kryo Gold @ 2.0 GHz, 4x Kryo Silver @ 1.8GHz)
  • GPU – Qualcomm Adreno GPU 7-series
  • DSP – Qualcomm Hexagon DSP with HVX and HMX
  • ISP – Qualcomm Spectra ISP 665 image processing
  • VPU
    • Video Encode – Up to 4K @ 60 fps for H.264/H.265
    • Video Decode – Up to 4K @ 120 fps for H.264/HEVC and up to 4K @ 30 fps for VP9
  • AI performance – 6 TOPS
  • Process – 4nm
• System Memory – Up to 16GB LPDDR5
• Storage
  • MicroSD card slot
  • UFS module interface
• Video Interfaces
  • HDMI port up to 1080p90
  • 4-lane MIPI DSI connector
• Camera – 4-lane MIPI CSI connector up to 32MP camera
• Networking – 2x 2.5GbE RJ45 ports, one with optional PoE
• USB – USB 3.1 Gen1 Type-A port
• Expansion – 40-pin GPIO header with UART, SPI, I2C, I3C,  etc.
• Misc
  • Power and EDL buttons
  • RTC battery connector
  • PWM fan connector
• Power Supply – 5V via USB-C port
• Dimensions – 65 x 56 mm

The Dragon Q5E will support Debian-based Radxa OS and Ubuntu, and documentation will eventually be published on the Radxa documentation website.

DragonStation and DragonBayNAS systems

Two Qualcomm-based NAS systems by Radxa are also in the works in collaboration with FeiNiu, enabling support for fnOS. The DragonStation will be a 6-bay M.2 NVMe SSD NAS with support for 10GbE networking, and we’re told an AI accelerator card will enable support for 120B local models and operation of Agents such as OpenClaw and Hermes.

The DragonBay will be a 4-bay mainstream NAS built on an unnamed Qualcomm mobile platform, and offer high-capacity storage, media libraries, photo archiving, data backup, and multi-user file collaboration.

Radxa hasn’t released further technical details about the two NAS so far.

Roadmap and pricing

The company will be quite busy with Qualcomm projects this year, as the 2026 roadmap below (source: SBCWiki on X) shows a total of 22 products, from the existing Dragon Q6B to the high-end RoboX Q1000 robotic platform and rCore-Q1000 system-on-module, both based on the upcoming Dragonwing IQ10-series SoCs. There will also be Raspberry Pi Compute Module-compatible modules and a cluster system based on the Dragonwing IQ-9075 SoC, also found in the company’s Fogwise AIRbox Q900.

“Pleasantly surprising” prices for the Dragon Q8B and Q5E were supposed to be announced at the developer event on May 30, but I was unable to find public reports and get answers from Radxa. Most information above was gathered from Radxa’s WeChat account.

The post Radxa’s 2026 Qualcomm hardware: Dragon Q8B and Q5E SBCs, DragonStation and DragonBay NAS systems appeared first on CNX Software - Embedded Systems News.

I received a kit for review last week with an ODROID-H5 SBC, a Type1 case, an M.2 10GbE module, and other accessories. In the first part of the review, I went through an unboxing, assembled the kit, and tested whether it could boot an M.2 NVMe SSD with Ubuntu 24.04 and Windows 11.

I’ve now updated the system to Ubuntu 26.04, run a few benchmarks, tested the two 10GbE RJ45 ports, as well as other features. I’ll report my experience about all that today.

Upgrading from Ubuntu 24.04 to Ubuntu 26.04

The 512GB M.2 NVMe SSD I use in the ODROID-H5 comes from a laptop, which I upgraded with a 2TB SSD. That means the operating systems were not upgraded for a few months. While Ubuntu 24.04 could boot, it would not show the two 10GbE interfaces.

That’s because the required drivers were not available in the version of Ubuntu 24.04 HWE on the SSD. I thought, no problem, I’ll just use a USB-C dock with built-in Ethernet. The only problem is that the ODROID-H5 doesn’t come with a USB-C port. I switched the USB-C to USB-C cable with a USB-C to USB-A (power and data) cable, but the USB-C dock (MINIX 480GB SSD) was not recognized with that cable. I ended up using the same USB cable with my Android smartphone and enabled USB tethering to update Ubuntu 24.04 to a more recent kernel (Linux 6.17) with RTL8127 drivers.

After that, both RTL8127-based 10Gbps Ethernet interfaces were recognized. Sp I disconnected the Android smartphone and upgraded from Ubuntu 24.04 to Ubuntu 26.04 over one of the Ethernet ports without major issues. That’s probably a unique case, and anybody installing Ubuntu 26.04 or the latest version of Ubuntu 24.04 should have both RJ45 ports working out of the box.

ODROID-H5 System Info on Ubuntu 26.04

Let’s check the system information in Settings->About.

We are shown a HARDKERNEL ODROID-H5 hardware model powered by an Intel Core i3-N300 octa-core processor paired with 16GB RAM and 512.1 GB storage running Ubuntu 26.04 LTS. The device name is CNX-LAPTOP-5 since it came from a laptop, and I’ll change that to CNX-ODROID-H5 shortly…

Ubuntu 26.04 LTS 64-bit ships with  Linux 7.0 and defaults to Wayland windowing system as expected…

Let’s find out more details about the system using inxi utility:

jaufranc@CNX-ODROID-H5:~$ sudo inxi -Fc0
System:
  Host: CNX-ODROID-H5 Kernel: 7.0.0-15-generic arch: x86_64 bits: 64
  Console: pty pts/1 Distro: Ubuntu 26.04 (Resolute Raccoon)
Machine:
  Type: Desktop Mobo: HARDKERNEL model: ODROID-H5 v: 1.0 serial: N/A
    Firmware: UEFI vendor: American Megatrends LLC. v: 1.2 date: 04/09/2026
CPU:
  Info: 8-core model: Intel Core i3-N300 bits: 64 type: MCP cache: L2: 4 MiB
  Speed (MHz): avg: 700 min/max: 700/3800 cores: 1: 700 2: 700 3: 700 4: 700
    5: 700 6: 700 7: 700 8: 700
Graphics:
  Device-1: Intel Alder Lake-N [UHD Graphics] driver: i915 v: kernel
  Display: unspecified server: X.org v: 1.21.1.22 with: Xwayland v: 24.1.10
    driver: X: loaded: modesetting unloaded: fbdev,vesa dri: iris gpu: i915
    tty: 80x24 resolution: 1920x1080
  API: EGL v: 1.5 drivers: iris,swrast platforms: gbm,surfaceless,device
  API: OpenGL v: 4.6 compat-v: 4.5 vendor: mesa v: 26.0.3-1ubuntu1
    note: console (EGL sourced) renderer: Mesa Intel Graphics (ADL-N), llvmpipe
    (LLVM 21.1.8 256 bits)
  Info: Tools: api: eglinfo,glxinfo wl: swaymsg,wlr-randr x11: xdriinfo,
    xdpyinfo, xprop, xrandr
Audio:
  Device-1: Intel Alder Lake-N PCH High Definition Audio driver: snd_hda_intel
  API: ALSA v: k7.0.0-15-generic status: kernel-api
Network:
  Device-1: Realtek RTL8127 10GbE driver: r8169
  IF: enp1s0 state: down mac: 00:1e:06:45:e4:cc
  Device-2: Realtek RTL8127 10GbE driver: r8169
  IF: enp2s0 state: up speed: 2500 Mbps duplex: full mac: 00:1e:06:45:dc:2b
  IF-ID-1: docker0 state: down mac: 46:3e:46:34:b1:c2
Drives:
  Local Storage: total: 476.94 GiB used: 332.89 GiB (69.8%)
  ID-1: /dev/nvme0n1 vendor: Intel model: SSDPEKNU512GZ size: 476.94 GiB
Partition:
  ID-1: / size: 369.97 GiB used: 332.79 GiB (90.0%) fs: ext4
    dev: /dev/nvme0n1p5
  ID-2: /boot/efi size: 256 MiB used: 103.7 MiB (40.5%) fs: vfat
    dev: /dev/nvme0n1p1
Swap:
  ID-1: swap-1 type: file size: 8 GiB used: 0 KiB (0.0%) file: /swapfile
  ID-2: swap-2 type: zram size: 7.43 GiB used: 0 KiB (0.0%) dev: /dev/zram0
Sensors:
  System Temperatures: cpu: 79.5 C mobo: N/A
  Fan Speeds (rpm): N/A
Info:
  Memory: total: 16 GiB available: 14.86 GiB used: 3.97 GiB (26.7%)
    igpu: 60 MiB
  Processes: 284 Uptime: 2m Init: systemd Shell: Sudo inxi: 3.3.40

The Core i3-N300 features eight cores clocked at 700 to 3800 MHz, the two RTL8127 10GbE interfaces are detected once connected at 2.5 Gbps (expecting since the mini PC is connected to a 2.5GbE switch at this stage), and we’ve got 16GB RAM and a 476.94 GiB NVMe SSD. Everything looks good, except that a CPU temperature of 79.5°C is a bit worrying after updating a few packages in Ubuntu 26.04.

ODROID-H5 benchmarks

Let’s find out how the SBC performs with some benchmarks, all done at a room temperature of around 30 to 32°C.

I’ll start with the usual sbc-bench.sh script:

jaufranc@CNX-ODROID-H5:~$ sudo ./sbc-bench.sh -r
Starting to examine hardware/software for review purposes...

sbc-bench v0.9.72

Installing needed tools: apt-get -f -qq -y install powercap-utils links mmc-utils smartmontools stress-ng, p7zip 16.02, tinymembench, ramlat, mhz, cpufetch, cpuminer. Done.
Checking cpufreq OPP. Done.
Executing tinymembench. Done.
Executing RAM latency tester. Done.
Executing OpenSSL benchmark. Done.
Executing 7-zip benchmark. Done.
Throttling test: heating up the device, 5 more minutes to wait. Done.
Checking cpufreq OPP again. Done (12 minutes elapsed).

Results validation:

  * Measured clockspeed not lower than advertised max CPU clockspeed
  * No swapping
  * Background activity (%system) OK
  * Too much other background activity: 0% avg, 4% max -> https://tinyurl.com/mr2wy5uv
  * Powercap detected. Details: "sudo powercap-info -p intel-rapl" -> https://tinyurl.com/4jh9nevj

# HARDKERNEL ODROID-H5 1.0 / i3-N300

Tested with sbc-bench v0.9.72 on Sat, 30 May 2026 14:55:58 +0700.

### General information:

    Information courtesy of cpufetch:
    
    Name:                Intel(R) Core(TM) i3-N300
    Microarchitecture:   Alder Lake
    Technology:          10nm
    Max Frequency:       3.800 GHz
    Cores:               8 cores
    AVX:                 AVX,AVX2
    FMA:                 FMA3
    L1i Size:            64KB (512KB Total)
    L1d Size:            32KB (256KB Total)
    L2 Size:             2MB (4MB Total)
    L3 Size:             6MB
    
    i3-N300, Kernel: x86_64, Userland: amd64
    
    CPU sysfs topology (clusters, cpufreq members, clockspeeds)
                     cpufreq   min    max
     CPU    cluster  policy   speed  speed   core type
      0        0        0      700    3800       -
      1        0        1      700    3800       -
      2        0        2      700    3800       -
      3        0        3      700    3800       -
      4        0        4      700    3800       -
      5        0        5      700    3800       -
      6        0        6      700    3800       -
      7        0        7      700    3800       -

15215 KB available RAM

### Policies (performance vs. idle consumption):

Status of performance related policies found below /sys:

    /sys/module/pcie_aspm/parameters/policy: default [performance] powersave powersupersave

### Clockspeeds (idle vs. heated up):

Before at 92.0°C:

    cpu0: OPP: 3800, Measured: 3786 

After at 94.0°C (throttled):

    cpu0: OPP: 3800, Measured: 3638      (-4.3%)

### Performance baseline

  * memcpy: 10382.9 MB/s, memchr: 16875.8 MB/s, memset: 19661.3 MB/s
  * 16M latency: 104.2 91.02 104.0 91.47 103.8 87.02 85.24 91.69 
  * 128M latency: 111.0 107.6 111.4 108.1 111.3 104.2 103.7 107.3 
  * 7-zip MIPS (3 consecutive runs): 13754, 12460, 11069 (12430 avg), single-threaded: 3797
  * `aes-256-cbc     636313.90k   942809.62k  1019438.34k  1045540.18k  1062128.30k  1053212.67k`
  * `aes-256-cbc     778920.92k  1068944.68k  1137494.19k  1157777.41k  1166417.92k  1164170.58k`

### PCIe and storage devices:

  * Intel Alder Lake-N [UHD Graphics] (Onboard - Video): driver in use: i915
  * Intel Alder Lake-N Processor USB 3.2 xHCI (Onboard - Other): driver in use: xhci_hcd
  * Intel Alder Lake-N PCH USB 3.2 Gen 2x1 (10 Gb/s) xHCI Host (Onboard - Other): driver in use: xhci_hcd
  * Intel Alder Lake-N eMMC (Onboard - Other): driver in use: sdhci-pci
  * Intel Alder Lake-N PCH High Definition Audio (Onboard - Sound): driver in use: snd_hda_intel
  * Realtek RTL8127 10GbE: Speed 8GT/s, Width x2, driver in use: r8169, 
  * Realtek RTL8127 10GbE: Speed 8GT/s, Width x2, driver in use: r8169, 
  * 476.9GB "INTEL SSDPEKNU512GZ" SSD as /dev/nvme0: Speed 8GT/s, Width x2 (downgraded), 4% worn out, drive temp: 60°C, ASPM Disabled
  * Winbond W25Q128JV 16MB SPI NOR flash, drivers in use: spi-nor/intel-spi

### Challenging filesystems:

The following partitions are NTFS: nvme0n1p3,nvme0n1p4 -> https://tinyurl.com/mv7wvzct

### Swap configuration:

  * /swapfile on /dev/nvme0n1p5: 8.0G (0K used)
  * /dev/zram0: 7.4G (0K used, lzo-rle, 4K streams, 74B data, 12K compressed,  total)

### Software versions:

  * Ubuntu 26.04 LTS (resolute)
  * Compiler: /usr/bin/gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0 / x86_64-linux-gnu
  * OpenSSL 3.5.5, built on 27 Jan 2026 (Library: OpenSSL 3.5.5 27 Jan 2026)    

### Kernel info:

  * `/proc/cmdline: BOOT_IMAGE=/boot/vmlinuz-7.0.0-15-generic root=UUID=39c940e2-b543-4242-a63b-816552412786 ro quiet splash crashkernel=2G-4G:320M,4G-32G:512M,32G-64G:1024M,64G-128G:2048M,128G-:4096M`
  * Vulnerability Reg file data sampling:    Mitigation; Clear Register File
  * Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
  * Vulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization
  * Vulnerability Vmscape:                   Mitigation; IBPB before exit to userspace
  * Kernel 7.0.0-15-generic / CONFIG_HZ=1000

Waiting for the device to cool down............... 94.0°C^C

CPU throttling does occur, and the CPU temperature is ultra-high at all times.  We can check a few data points from the full log.

Tinymembench memory benchmark is a single-thread benchmark and normally not challenging for the system, but the temperature was 94-95°C:

System health while running tinymembench:

Time        CPU    load %cpu %sys %usr %nice %io %irq   Temp
14:44:12: 3433MHz  1.76  12%   0%   4%   6%   0%   0%  95.0°C  
14:44:22: 3450MHz  1.64  12%   0%  12%   0%   0%   0%  95.0°C  
14:44:33: 3253MHz  1.54  12%   0%  12%   0%   0%   0%  95.0°C  
14:44:43: 3648MHz  1.46  12%   0%  12%   0%   0%   0%  94.0°C  
14:44:53: 3157MHz  1.36  12%   0%  12%   0%   0%   0%  95.0°C  
14:45:03: 3166MHz  1.30  16%   0%  12%   3%   0%   0%  95.0°C  
14:45:13: 3208MHz  1.25  12%   0%  12%   0%   0%   0%  95.0°C  
14:45:23: 3468MHz  1.21  12%   0%  12%   0%   0%   0%  94.0°C  
14:45:33: 3662MHz  1.18  12%   0%  12%   0%   0%   0%  94.0°C

cpuminer is much more demanding, and performance collapses with the CPU running as low as 800 MHz here:

System health while running cpuminer:

Time        CPU    load %cpu %sys %usr %nice %io %irq   Temp
14:51:01: 1100MHz  8.18  28%   0%  23%   3%   0%   0%  94.0°C  
14:51:45:  800MHz  8.09 100%   0%  99%   0%   0%   0%  94.0°C  
14:52:29: 1100MHz  8.12 100%   0%  99%   0%   0%   0%  96.0°C  
14:53:12: 1148MHz  8.10 100%   0%  99%   0%   0%   0%  94.0°C  
14:53:56: 1001MHz  8.13 100%   0%  99%   0%   0%   0%  94.0°C  
14:54:40: 1045MHz  8.13 100%   0%  99%   0%   0%   0%  94.0°C  
14:55:24:  800MHz  8.10 100%   0%  99%   0%   0%   0%  94.0°C

Let’s check the power limits:

jaufranc@CNX-ODROID-H5:~$ sudo powercap-info -p intel-rapl
enabled: 1
Zone 0
  name: package-0
  enabled: 1
  max_energy_range_uj: 262143328850
  energy_uj: 15967828873
  Constraint 0
    name: long_term
    power_limit_uw: 17000000
    time_window_us: 27983872
    max_power_uw: 7000000
  Constraint 1
    name: short_term
    power_limit_uw: 20000000
    time_window_us: 2440
    max_power_uw: 0
  Constraint 2
    name: peak_power
    power_limit_uw: 78000000
    max_power_uw: 0
  Zone 0:0
    name: core
    enabled: 0
    max_energy_range_uj: 262143328850
    energy_uj: 4003726933
    Constraint 0
      name: long_term
      power_limit_uw: 0
      time_window_us: 976
  Zone 0:1
    name: uncore
    enabled: 0
    max_energy_range_uj: 262143328850
    energy_uj: 1285824
    Constraint 0
      name: long_term
      power_limit_uw: 0
      time_window_us: 976

PL1 and PL2 power limits were set to 17 and 20 Watts, respectively, fairly high values, and in Hardkernel’s “unlimited performance” mode.

That’s a problem, so we either have to lower the power limits or install a fan. Hardkernel does mention a fan is recommended on the product page:

While the generous onboard heatsink makes fanless operation technically feasible, we strongly advocate for the installation of an active cooling solution to preserve the peak performance of the H5 board’s 8-core architecture during sustained loads.

However, by sending the kit without a fan, they sent the wrong signal, and I assumed fanless operation would be fine. A quick email exchange confirmed I needed to add a fan. Luckily, I still have the 92mm fan from the ODROID-H4 Type 3 case I reviewed in 2024. So let’s turn off the system and install the PWM fan.

The fan is not particularly noisy (in a room with an air conditioner), and the idle CPU temperature is now much lower:

Sensors:
  System Temperatures: cpu: 43.5 C mobo: N/A
  Fan Speeds (rpm): N/A

Let’s run sbc-bench.sh again:

jaufranc@CNX-ODROID-H5:~$ sudo ./sbc-bench.sh -r
Starting to examine hardware/software for review purposes...

sbc-bench v0.9.72

Installing needed tools: distro packages already installed. Done.
Checking cpufreq OPP. Done.
Executing tinymembench. Done.
Executing RAM latency tester. Done.
Executing OpenSSL benchmark. Done.
Executing 7-zip benchmark. Done.
Throttling test: heating up the device, 5 more minutes to wait. Done.
Checking cpufreq OPP again. Done (10 minutes elapsed).

Results validation:

  * Measured clockspeed not lower than advertised max CPU clockspeed
  * No swapping
  * Background activity (%system) OK
  * Too much other background activity: 0% avg, 9% max -> https://tinyurl.com/mr2wy5uv
  * Powercap detected. Details: "sudo powercap-info -p intel-rapl" -> https://tinyurl.com/4jh9nevj

# HARDKERNEL ODROID-H5 1.0 / i3-N300

Tested with sbc-bench v0.9.72 on Sat, 30 May 2026 16:25:51 +0700.

### General information:

    Information courtesy of cpufetch:
    
    Name:                Intel(R) Core(TM) i3-N300
    Microarchitecture:   Alder Lake
    Technology:          10nm
    Max Frequency:       3.800 GHz
    Cores:               8 cores
    AVX:                 AVX,AVX2
    FMA:                 FMA3
    L1i Size:            64KB (512KB Total)
    L1d Size:            32KB (256KB Total)
    L2 Size:             2MB (4MB Total)
    L3 Size:             6MB
    
    i3-N300, Kernel: x86_64, Userland: amd64
    
    CPU sysfs topology (clusters, cpufreq members, clockspeeds)
                     cpufreq   min    max
     CPU    cluster  policy   speed  speed   core type
      0        0        0      700    3800   Alder Lake
      1        0        1      700    3800   Alder Lake
      2        0        2      700    3800   Alder Lake
      3        0        3      700    3800   Alder Lake
      4        0        4      700    3800   Alder Lake
      5        0        5      700    3800   Alder Lake
      6        0        6      700    3800   Alder Lake
      7        0        7      700    3800   Alder Lake

15215 KB available RAM

### Policies (performance vs. idle consumption):

Status of performance related policies found below /sys:

    /sys/module/pcie_aspm/parameters/policy: default [performance] powersave powersupersave

### Clockspeeds (idle vs. heated up):

Before at 44.0°C:

    cpu0: OPP: 3800, Measured: 3786 

After at 59.0°C:

    cpu0: OPP: 3800, Measured: 3786 

### Performance baseline

  * memcpy: 12034.0 MB/s, memchr: 19681.9 MB/s, memset: 21967.5 MB/s
  * 16M latency: 103.9 91.07 104.3 91.01 103.6 86.84 84.47 89.69 
  * 128M latency: 110.2 107.1 110.3 107.5 109.6 103.1 102.4 103.4 
  * 7-zip MIPS (3 consecutive runs): 23742, 23860, 23872 (23820 avg), single-threaded: 4308
  * `aes-256-cbc    1009051.44k  1316697.66k  1361145.00k  1372802.05k  1376116.74k  1375857.32k`
  * `aes-256-cbc    1026534.15k  1316796.59k  1361263.87k  1372716.71k  1375636.14k  1376447.15k`

### PCIe and storage devices:

  * Intel Alder Lake-N [UHD Graphics] (Onboard - Video): driver in use: i915
  * Intel Alder Lake-N Processor USB 3.2 xHCI (Onboard - Other): driver in use: xhci_hcd
  * Intel Alder Lake-N PCH USB 3.2 Gen 2x1 (10 Gb/s) xHCI Host (Onboard - Other): driver in use: xhci_hcd
  * Intel Alder Lake-N eMMC (Onboard - Other): driver in use: sdhci-pci
  * Intel Alder Lake-N PCH High Definition Audio (Onboard - Sound): driver in use: snd_hda_intel
  * Realtek RTL8127 10GbE: Speed 8GT/s, Width x2, driver in use: r8169, 
  * Realtek RTL8127 10GbE: Speed 8GT/s, Width x2, driver in use: r8169, 
  * 476.9GB "INTEL SSDPEKNU512GZ" SSD as /dev/nvme0: Speed 8GT/s, Width x2 (downgraded), 4% worn out, drive temp: 36°C, ASPM Disabled
  * Winbond W25Q128JV 16MB SPI NOR flash, drivers in use: spi-nor/intel-spi

### Challenging filesystems:

The following partitions are NTFS: nvme0n1p3,nvme0n1p4 -> https://tinyurl.com/mv7wvzct

### Swap configuration:

  * /swapfile on /dev/nvme0n1p5: 8.0G (0K used)
  * /dev/zram0: 7.4G (0K used, lzo-rle, 4K streams, 74B data, 12K compressed,  total)

### Software versions:

  * Ubuntu 26.04 LTS (resolute)
  * Compiler: /usr/bin/gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0 / x86_64-linux-gnu
  * OpenSSL 3.5.5, built on 27 Jan 2026 (Library: OpenSSL 3.5.5 27 Jan 2026)    

### Kernel info:

  * `/proc/cmdline: BOOT_IMAGE=/boot/vmlinuz-7.0.0-15-generic root=UUID=39c940e2-b543-4242-a63b-816552412786 ro quiet splash crashkernel=2G-4G:320M,4G-32G:512M,32G-64G:1024M,64G-128G:2048M,128G-:4096M`
  * Vulnerability Reg file data sampling:    Mitigation; Clear Register File
  * Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
  * Vulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization
  * Vulnerability Vmscape:                   Mitigation; IBPB before exit to userspace
  * Kernel 7.0.0-15-generic / CONFIG_HZ=1000

Waiting for the device to cool down...................................... 49.0°C

It’s night and day. The frequency is stable (2600 MHz), and the CPU temperature maxes out at 67°C with cpuminer:

System health while running cpuminer:

Time        CPU    load %cpu %sys %usr %nice %io %irq   Temp
16:20:48: 2600MHz  7.20  31%   1%  29%   0%   0%   0%  65.0°C  
16:21:30: 2600MHz  7.59 100%   0%  99%   0%   0%   0%  66.0°C  
16:22:11: 2600MHz  7.85 100%   0%  99%   0%   0%   0%  67.0°C  
16:22:53: 2600MHz  7.92 100%   0%  99%   0%   0%   0%  67.0°C  
16:23:34: 2600MHz  7.96 100%   0%  99%   0%   0%   0%  67.0°C  
16:24:16: 2600MHz  7.98 100%   0%  99%   0%   0%   0%  67.0°C  
16:24:57: 2600MHz  7.99 100%   0%  99%   0%   0%   0%  67.0°C  
16:25:39: 2600MHz  8.00 100%   0%  99%   0%   0%   0%  67.0°C

You can check the full log for reference. Needless to say, the rest of the review was done with the cooling fan. We’ll compare benchmark results further below.

Geekbench 6.7.1 benchmark was used to check single-core and multi-core performance.

The octa-core Core i3-N300 SBC scored 1,349 points (single core) and 5,178 points (multi-core).

Let’s start testing the built-in GPU with Unigine Heaven Benchmark 4.0. Our (actively-cooled) ODROID-H5 rendered the benchmark scene at 18.9 FPS on average with 475 points at the standard 1920×1080 resolution.

We further tested the internal GPU (iGPU) by playing some YouTube videos starting in Firefox at 4K and 8K resolutions.

A VP9 video played smoothly at 4K 30FPS with only 3 frames dropped out of 3486 after playing the short video for about 2 minutes.

The same video at 8K 30FPS was more challenging. There were bursts of dropped frames from time to time, and at other times, the video would play smoothly. In any case, it was not watchable.

Another VP9 video played at 4K 60 FPS was watchable, although the number of dropped frames (1,795 out of 18,058) is worrying. It’s possible this video would have played more smoothly at a lower room temperature.

As one would have expected, the same video at 8K 60 FPS was a disaster with close to 100% dropped frames. I just played it for a few seconds…

I also tried to play the 8K 30 FPS video on Chrome, but the result was about the same as in Firefox. In summary, YouTube 4K video streaming is possible, but forget about 8K, which is not that important anyway since the video outputs are limited to 4K…

The WebGL Aquarium demo performed relatively well in Firefox, with 60 FPS rendering using 500/1000 fish, 30 FPS with 15,000 fish, and 25 FPS with 20,000 fish (see screenshot above).

Speedometer 2.0 can be used to estimate web browsing performance.

On Firefox, the board managed 174 runs per minutes.

On Chrome, it was 267 runs per minute. I used the older, deprecated Speedometer 2.0 instead of Speedometer 3.0 for the comparison with previous platforms. But let’s still run Speedometer 3.0 on Firefox for future reviews.

ODROID-H5 Ubuntu 26.04 benchmarks comparison against other Alder Lake-N systems

Now that we have benchmark results for the ODROID-H5 on Ubuntu 26.04, we can compare it to the previous-generation ODROID-H4+ (Intel N97) and the Weibu N10 (Intel Core i3-N305) mini PC.

Here are the basic specifications of the three systems:

And now the benchmark results.

The first conclusion is that the fan is definitely a necessity for the ODROID-H5, especially for CNX Software’s “tropical reviews” at relatively high ambient temperature, and for instance, the 7-zip performance doubled with the fan.

The ODROID-H5 offers a nice upgrade to the ODROID-H4+ if you care about multi-core performance. Compared to the Weibu N10, Geekbench results should be ignored since GB 5 and 6 scores can’t be compared, but overall the ODROID-H5 performs better, mostly because the cooling solution for the mini PC was not optimal, as shown in the 7-zip average vs top result fields. If you own an ODROID-H4 Ultra (we don’t) with an Intel Core i3-N305, you may get slightly better performance than the ODROID-H5, but it’s not available anymore due to supply issues.

Storage and USB testing

Note that the ODROID-H5 board does not come with any storage by default, so what I’ll do here is mostly test the NVMe interface (PCIe Gen3 x2) with the drive I installed. The theory is 16 GT/s… Let’s see what iozone3 reports:

jaufranc@CNX-ODROID-H5:~$ sudo iozone -e -I -a -s 1000M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
	Iozone: Performance Test of File I/O
	        Version $Revision: 3.508 $
		Compiled for 64 bit mode.
		Build: linux-AMD64 
                                                                    random    random      bkwd     record     stride                                        
              kB  reclen    write    rewrite      read    reread      read     write      read    rewrite       read    fwrite  frewrite     fread   freread
         1024000       4    131128    219380    204214    201769     52454     47841                                                                
         1024000      16     41801    333729    280733    283841    164924    321783                                                                
         1024000     512   1029101    989163    917811    945881   1001884    912225                                                                
         1024000    1024    665333     31050   1000038   1031773    960903     41382                                                                
         1024000   16384   1292577   1419928   1379885   1418714   1415326     83051                                                                

iozone test complete.

Sequential read is 1.37 GB/s and sequential write is 1.29 GB/s. It’s acceptable for the Intel SSD used, although I would have accepted something closer to 1.5-1.6 GB/s.

The ODROID-H5 comes with three USB 2.0 ports and one USB 3.2 (10Gbps) port. I tested all four with lsusb and iozone using storage devices with EXT-4 file system. From top left to bottom right:

• USB 2.0 – 480 Mbps – Read: 42 MB/s
• USB 2.0 – 480 Mbps – Read: 42 MB/s
• USB 2.0 – 480 Mbps – Read: 42 MB/s
• USB 3.0 –  10,000 Mbps – Read: 948 MB/s

All work as advertised. I would have wished for one more USB 3.0 port, especially a full-function USB-C port, but that’s probably the cost of getting four M.2 PCIe sockets and a 10GbE RTL8127 chip on the board.

10GbE networking

Besides the four M.2 PCIe sockets, the most important features of the ODROID-H5 is it’s built-in 10GbE port implemented through a Realtek RTL8127 chipset. My system even has two because I was sent an M.2 10GbE module as part of the kit.

I’ll use the iKOOLCORE R2 Max mini PC running QWRT (OpenWrt fork) and equipped with four Ethernet ports (2x 10GbE and 2x 2.5GbE) on the other side.

A first quick check shows that both Ethernet interfaces are connected at 10,000 Mbps.

I’ll test each individually, starting with the built-in 10GbE port.

• Upload

jaufranc@CNX-ODROID-H5:~$ iperf3 -t 60 -c 192.168.4.1 -i 10
Connecting to host 192.168.4.1, port 5201
[  5] local 192.168.4.169 port 52144 connected to 192.168.4.1 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-10.01  sec  11.0 GBytes  9.41 Gbits/sec    0   1.95 MBytes       
[  5]  10.01-20.01  sec  11.0 GBytes  9.41 Gbits/sec    0   2.56 MBytes       
[  5]  20.01-30.01  sec  11.0 GBytes  9.41 Gbits/sec    0   2.56 MBytes       
[  5]  30.01-40.00  sec  11.0 GBytes  9.41 Gbits/sec    0   2.56 MBytes       
[  5]  40.00-50.01  sec  11.0 GBytes  9.41 Gbits/sec    0   2.56 MBytes       
[  5]  50.01-60.00  sec  11.0 GBytes  9.41 Gbits/sec    0   2.56 MBytes       
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-60.00  sec  65.8 GBytes  9.41 Gbits/sec    0            sender
[  5]   0.00-60.00  sec  65.8 GBytes  9.41 Gbits/sec                  receiver

iperf Done.

• Download

jaufranc@CNX-ODROID-H5:~$ iperf3 -t 60 -c 192.168.4.1 -i 10 -R
Connecting to host 192.168.4.1, port 5201
Reverse mode, remote host 192.168.4.1 is sending
[  5] local 192.168.4.169 port 58094 connected to 192.168.4.1 port 5201
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.01  sec  11.0 GBytes  9.41 Gbits/sec                  
[  5]  10.01-20.01  sec  11.0 GBytes  9.41 Gbits/sec                  
[  5]  20.01-30.01  sec  11.0 GBytes  9.42 Gbits/sec                  
[  5]  30.01-40.00  sec  11.0 GBytes  9.41 Gbits/sec                  
[  5]  40.00-50.01  sec  11.0 GBytes  9.42 Gbits/sec                  
[  5]  50.01-60.01  sec  11.0 GBytes  9.41 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-60.01  sec  65.8 GBytes  9.41 Gbits/sec    0            sender
[  5]   0.00-60.01  sec  65.8 GBytes  9.41 Gbits/sec                  receiver

iperf Done.

• Full-duplex/bidirectional:

jaufranc@CNX-ODROID-H5:~$ iperf3 -t 60 -c 192.168.4.1 -i 10 --bidir
Connecting to host 192.168.4.1, port 5201
[  5] local 192.168.4.169 port 38018 connected to 192.168.4.1 port 5201
[  7] local 192.168.4.169 port 38034 connected to 192.168.4.1 port 5201
[ ID][Role] Interval           Transfer     Bitrate         Retr  Cwnd
[  5][TX-C]   0.00-10.01  sec  10.9 GBytes  9.36 Gbits/sec    0   2.71 MBytes       
[  7][RX-C]   0.00-10.01  sec  10.8 GBytes  9.27 Gbits/sec                  
[  5][TX-C]  10.01-20.01  sec  10.9 GBytes  9.35 Gbits/sec    0   2.71 MBytes       
[  7][RX-C]  10.01-20.01  sec  10.7 GBytes  9.19 Gbits/sec                  
[  5][TX-C]  20.01-30.01  sec  10.9 GBytes  9.37 Gbits/sec    0   2.71 MBytes       
[  7][RX-C]  20.01-30.01  sec  10.8 GBytes  9.30 Gbits/sec                  
[  5][TX-C]  30.01-40.01  sec  10.9 GBytes  9.38 Gbits/sec    0   2.71 MBytes       
[  7][RX-C]  30.01-40.01  sec  10.8 GBytes  9.28 Gbits/sec                  
[  5][TX-C]  40.01-50.01  sec  10.9 GBytes  9.36 Gbits/sec    0   2.71 MBytes       
[  7][RX-C]  40.01-50.01  sec  10.7 GBytes  9.21 Gbits/sec                  
[  5][TX-C]  50.01-60.01  sec  10.9 GBytes  9.36 Gbits/sec    0   2.71 MBytes       
[  7][RX-C]  50.01-60.01  sec  10.6 GBytes  9.14 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID][Role] Interval           Transfer     Bitrate         Retr
[  5][TX-C]   0.00-60.01  sec  65.4 GBytes  9.36 Gbits/sec    0            sender
[  5][TX-C]   0.00-60.01  sec  65.4 GBytes  9.36 Gbits/sec                  receiver
[  7][RX-C]   0.00-60.01  sec  64.5 GBytes  9.23 Gbits/sec  289            sender
[  7][RX-C]   0.00-60.01  sec  64.5 GBytes  9.23 Gbits/sec                  receiver

iperf Done.

Upload and download tests are both maxed out at 9.41 Gbps. The full-duplex test is still pretty good, but slightly below the theoretical limit. In the review of the iKOOLCORE R2 Max (Intel N100), we noted that the bottleneck can be the CPU.

So I can iperf3 full-duplex again on with two parallel streams to use more than one core:

jaufranc@CNX-ODROID-H5:~$ iperf3 -t 60 -c 192.168.4.1 -i 10 --bidir -P 2
Connecting to host 192.168.4.1, port 5201
[  5] local 192.168.4.169 port 50932 connected to 192.168.4.1 port 5201
[  7] local 192.168.4.169 port 50946 connected to 192.168.4.1 port 5201
[  9] local 192.168.4.169 port 50956 connected to 192.168.4.1 port 5201
[ 11] local 192.168.4.169 port 50970 connected to 192.168.4.1 port 5201
[ ID][Role] Interval           Transfer     Bitrate         Retr  Cwnd
[  5][TX-C]   0.00-10.01  sec  5.47 GBytes  4.70 Gbits/sec    0   1.55 MBytes       
[  7][TX-C]   0.00-10.01  sec  5.48 GBytes  4.70 Gbits/sec    0   1.54 MBytes       
[SUM][TX-C]   0.00-10.01  sec  10.9 GBytes  9.40 Gbits/sec    0             
[  9][RX-C]   0.00-10.01  sec  5.46 GBytes  4.69 Gbits/sec                  
[ 11][RX-C]   0.00-10.01  sec  5.46 GBytes  4.69 Gbits/sec                  
[SUM][RX-C]   0.00-10.01  sec  10.9 GBytes  9.38 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[  5][TX-C]  10.01-20.01  sec  5.47 GBytes  4.70 Gbits/sec    0   1.64 MBytes       
[  7][TX-C]  10.01-20.01  sec  5.47 GBytes  4.70 Gbits/sec    0   1.63 MBytes       
[SUM][TX-C]  10.01-20.01  sec  10.9 GBytes  9.40 Gbits/sec    0             
[  9][RX-C]  10.01-20.01  sec  5.47 GBytes  4.70 Gbits/sec                  
[ 11][RX-C]  10.01-20.01  sec  5.46 GBytes  4.69 Gbits/sec                  
[SUM][RX-C]  10.01-20.01  sec  10.9 GBytes  9.39 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[  5][TX-C]  20.01-30.01  sec  5.47 GBytes  4.70 Gbits/sec    0   1.64 MBytes       
[  7][TX-C]  20.01-30.01  sec  5.47 GBytes  4.70 Gbits/sec    0   1.63 MBytes       
[SUM][TX-C]  20.01-30.01  sec  10.9 GBytes  9.40 Gbits/sec    0             
[  9][RX-C]  20.01-30.01  sec  5.46 GBytes  4.69 Gbits/sec                  
[ 11][RX-C]  20.01-30.01  sec  5.47 GBytes  4.70 Gbits/sec                  
[SUM][RX-C]  20.01-30.01  sec  10.9 GBytes  9.39 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[  5][TX-C]  30.01-40.01  sec  5.47 GBytes  4.70 Gbits/sec    0   1.64 MBytes       
[  7][TX-C]  30.01-40.01  sec  5.47 GBytes  4.70 Gbits/sec    0   1.63 MBytes       
[SUM][TX-C]  30.01-40.01  sec  10.9 GBytes  9.40 Gbits/sec    0             
[  9][RX-C]  30.01-40.01  sec  5.47 GBytes  4.70 Gbits/sec                  
[ 11][RX-C]  30.01-40.01  sec  5.47 GBytes  4.70 Gbits/sec                  
[SUM][RX-C]  30.01-40.01  sec  10.9 GBytes  9.39 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[  5][TX-C]  40.01-50.01  sec  5.47 GBytes  4.70 Gbits/sec    0   2.53 MBytes       
[  7][TX-C]  40.01-50.01  sec  5.47 GBytes  4.70 Gbits/sec    0   3.22 MBytes       
[SUM][TX-C]  40.01-50.01  sec  10.9 GBytes  9.40 Gbits/sec    0             
[  9][RX-C]  40.01-50.01  sec  5.47 GBytes  4.70 Gbits/sec                  
[ 11][RX-C]  40.01-50.01  sec  5.46 GBytes  4.69 Gbits/sec                  
[SUM][RX-C]  40.01-50.01  sec  10.9 GBytes  9.39 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[  5][TX-C]  50.01-60.01  sec  5.47 GBytes  4.70 Gbits/sec    0   2.53 MBytes       
[  7][TX-C]  50.01-60.01  sec  5.47 GBytes  4.70 Gbits/sec    0   3.22 MBytes       
[SUM][TX-C]  50.01-60.01  sec  10.9 GBytes  9.40 Gbits/sec    0             
[  9][RX-C]  50.01-60.01  sec  5.46 GBytes  4.69 Gbits/sec                  
[ 11][RX-C]  50.01-60.01  sec  5.48 GBytes  4.70 Gbits/sec                  
[SUM][RX-C]  50.01-60.01  sec  10.9 GBytes  9.39 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID][Role] Interval           Transfer     Bitrate         Retr
[  5][TX-C]   0.00-60.01  sec  32.8 GBytes  4.70 Gbits/sec    0            sender
[  5][TX-C]   0.00-60.01  sec  32.8 GBytes  4.70 Gbits/sec                  receiver
[  7][TX-C]   0.00-60.01  sec  32.8 GBytes  4.70 Gbits/sec    0            sender
[  7][TX-C]   0.00-60.01  sec  32.8 GBytes  4.70 Gbits/sec                  receiver
[SUM][TX-C]   0.00-60.01  sec  65.6 GBytes  9.40 Gbits/sec    0             sender
[SUM][TX-C]   0.00-60.01  sec  65.6 GBytes  9.40 Gbits/sec                  receiver
[  9][RX-C]   0.00-60.01  sec  32.8 GBytes  4.69 Gbits/sec  190            sender
[  9][RX-C]   0.00-60.01  sec  32.8 GBytes  4.69 Gbits/sec                  receiver
[ 11][RX-C]   0.00-60.01  sec  32.8 GBytes  4.69 Gbits/sec  199            sender
[ 11][RX-C]   0.00-60.01  sec  32.8 GBytes  4.69 Gbits/sec                  receiver
[SUM][RX-C]   0.00-60.01  sec  65.6 GBytes  9.39 Gbits/sec  389             sender
[SUM][RX-C]   0.00-60.01  sec  65.6 GBytes  9.39 Gbits/sec                  receiver

iperf Done.

That would be 9.40 Gbps Tx and 9.39 Rx, close to perfect.

Let’s now focus our attention on the M.2 10GbE module

• Upload:

jaufranc@CNX-ODROID-H5:~$ iperf3 -t 60 -c 192.168.4.1 -i 10
Connecting to host 192.168.4.1, port 5201
[  5] local 192.168.4.122 port 47024 connected to 192.168.4.1 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-10.01  sec  11.0 GBytes  9.41 Gbits/sec    0   2.10 MBytes       
[  5]  10.01-20.01  sec  11.0 GBytes  9.41 Gbits/sec    0   2.10 MBytes       
[  5]  20.01-30.00  sec  11.0 GBytes  9.41 Gbits/sec    0   3.86 MBytes       
[  5]  30.00-40.01  sec  11.0 GBytes  9.41 Gbits/sec    0   3.86 MBytes       
[  5]  40.01-50.01  sec  11.0 GBytes  9.41 Gbits/sec    0   3.86 MBytes       
[  5]  50.01-60.01  sec  11.0 GBytes  9.41 Gbits/sec    0   3.86 MBytes       
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-60.01  sec  65.8 GBytes  9.41 Gbits/sec    0            sender
[  5]   0.00-60.01  sec  65.8 GBytes  9.41 Gbits/sec                  receiver

iperf Done.

• Download:

jaufranc@CNX-ODROID-H5:~$ iperf3 -t 60 -c 192.168.4.1 -i 10 -R
Connecting to host 192.168.4.1, port 5201
Reverse mode, remote host 192.168.4.1 is sending
[  5] local 192.168.4.122 port 34616 connected to 192.168.4.1 port 5201
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.01  sec  11.0 GBytes  9.40 Gbits/sec                  
[  5]  10.01-20.01  sec  11.0 GBytes  9.41 Gbits/sec                  
[  5]  20.01-30.01  sec  10.9 GBytes  9.40 Gbits/sec                  
[  5]  30.01-40.01  sec  11.0 GBytes  9.41 Gbits/sec                  
[  5]  40.01-50.01  sec  11.0 GBytes  9.41 Gbits/sec                  
[  5]  50.01-60.01  sec  11.0 GBytes  9.41 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-60.01  sec  65.7 GBytes  9.41 Gbits/sec    1            sender
[  5]   0.00-60.01  sec  65.7 GBytes  9.41 Gbits/sec                  receiver

iperf Done.

• Full-duplex bidirectional:

jaufranc@CNX-ODROID-H5:~$ iperf3 -t 60 -c 192.168.4.1 -i 10 --bidir
Connecting to host 192.168.4.1, port 5201
[  5] local 192.168.4.122 port 56306 connected to 192.168.4.1 port 5201
[  7] local 192.168.4.122 port 56312 connected to 192.168.4.1 port 5201
[ ID][Role] Interval           Transfer     Bitrate         Retr  Cwnd
[  5][TX-C]   0.00-10.01  sec  10.9 GBytes  9.37 Gbits/sec    0   2.70 MBytes       
[  7][RX-C]   0.00-10.01  sec  10.8 GBytes  9.23 Gbits/sec                  
[  5][TX-C]  10.01-20.01  sec  10.9 GBytes  9.36 Gbits/sec    0   2.70 MBytes       
[  7][RX-C]  10.01-20.01  sec  10.7 GBytes  9.17 Gbits/sec                  
[  5][TX-C]  20.01-30.01  sec  10.9 GBytes  9.39 Gbits/sec    0   2.70 MBytes       
[  7][RX-C]  20.01-30.01  sec  10.9 GBytes  9.39 Gbits/sec                  
[  5][TX-C]  30.01-40.01  sec  10.9 GBytes  9.39 Gbits/sec    0   2.70 MBytes       
[  7][RX-C]  30.01-40.01  sec  10.9 GBytes  9.39 Gbits/sec                  
[  5][TX-C]  40.01-50.01  sec  10.9 GBytes  9.39 Gbits/sec    0   2.70 MBytes       
[  7][RX-C]  40.01-50.01  sec  10.9 GBytes  9.39 Gbits/sec                  
[  5][TX-C]  50.01-60.01  sec  10.9 GBytes  9.39 Gbits/sec    0   2.70 MBytes       
[  7][RX-C]  50.01-60.01  sec  10.9 GBytes  9.40 Gbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID][Role] Interval           Transfer     Bitrate         Retr
[  5][TX-C]   0.00-60.01  sec  65.6 GBytes  9.38 Gbits/sec    0            sender
[  5][TX-C]   0.00-60.01  sec  65.6 GBytes  9.38 Gbits/sec                  receiver
[  7][RX-C]   0.00-60.01  sec  65.2 GBytes  9.33 Gbits/sec  194            sender
[  7][RX-C]   0.00-60.01  sec  65.2 GBytes  9.33 Gbits/sec                  receiver

iperf Done.

Excellent even on one core, so I won’t try parallel streams here.

Triple display testing

The ODROID-H5 supports up to three displays thanks to its HDMI port and two DisplayPort connectors.

I connected the HDMI port to a 14-inch Crowview portable laptop monitor (Full HD), one of the DisplayPort to a KTC A32Q8 4K Google TV monitor using a DisplayPort cable, and the other DisplayPort connector to another KTC A32Q8 display with a DisplayPort to HDMI cable.

No problem here with one 1080p60 portable display and two 2160p60 monitors.

ODROID-H5 power consumption

I measured the ODROID-H5 power consumption with a wall power meter as follows:

• Power off – 1.0 – 1.1 Watts
• Idle
  • 1x 2.5GbE – 11.5 – 11.6 Watts
  • 1x 2.5GbE + 1x 10GbE – 12.9 – 13.2 Watts
  • 2x 10GbE – 14.2 – 14.6 Watts
• YouTube 4Kp60 (Firefox) – 32.1 – 36.2 Watts
• Stress test – 40.8 – 41.4 Watts
• iperf3 over 1x 10GbE full-duplex – 29.8 – 30.5 Watts

The ODROID-H5 was connected to an HDMI display, a USB RF dongle for a keyboard and mouse combo, and 2.5 Gbps Ethernet unless otherwise noted.

[Update: You can lower the idle power consumption somewhat by going to Chipset in the BIOS and selecting PCI Express Configuration inside PCH-IO.

Then for each available PCI Express Root Port set ASPM from disabled to Auto. Hardkernel disabled it due to some incompatible SSDs, but it increases power consumption when enabled.

Here are the updated idle power consumption numbers (and delta):

• 1x 2.5GbE – 9.1 – 9.2 Watts (down ~2.4 Watts)
• 1x 2.5GbE + 1x 10GbE – 10.5 – 10.6 Watts (down ~2.4 Watts)
• 2x 10GbE – 11.6 – 11.8 Watts (down ~2.6 Watts)

Note that the fan was active during measurements, likely because it’s hot in here. In a cooler climate, you can probably shave 0.5 Watts of the results above.

]

I also measured some relevant power consumption numbers on the iKOOLCORE R2 max (Intel N100):

• Idle – 2.5GbE (WAN) + 10GbE (WAN) – 15.3 Watss
• iperf3 over 1x 10GbE full-duplex
  • Server mode – 25.6 Watts
  • Client mode – 24.3 Watts

I was expecting the Marvell AQC113C-B1-C 10GbE controller on the R2 max to consume more than the Realtek RTL8127 controller on the ODROID-H5, but I suppose the Intel N100 vs Core i3-N300 and Ubuntu Desktop vs OpenWrt (headless) may impact results more than expected. Two other things were connected to the ODROID-H5 that were not on the R2 Max: an HDMI cable and a USB RF dongle, but those only account for a couple of Watts at most. The ODROID-H5 appears to still consume a little less at idle with 10GbE and 2.5GbE despite having a fan, but when running iperf3, the R2 Max consumes less for some reason.

Conclusion

The ODROID-H5 works great under Ubuntu 26.04 with excellent performance (once I connected a 92mm fan), and all features are working as expected, including the HDMI and DisplayPort connectors for triple display setup, the two 10GbE interfaces, and the USB ports.

The board is not really suitable for fanless operation, and most people will require a fan unless they lower the power limits and operate the device in cool temperatures. I found YouTube to struggle more than usual. Videos play fine at 4K 30 FPS, and are somewhat watchable at 4K 60 FPS (still 10% dropped frames), but 8K is very choppy. I also like the flexibility of the whole ecosystem around the ODROID-H5, as users can add network interfaces, SATA interfaces, NVMe storage, AI accelerators, and select among two cases (for now) to build what they need.

I’d like to thank Hardkernel for sending an ODROID-H5 SBC and accessories for review. Here’s the price breakdown of the kit I received:

• ODROID-H5 SBC – $250
• H5 Case Type 1 – $11
• LED Power Button – $5.99
• 15V/4A PSU (US plug) – $11
• M.2 10GbE card – $76
• 16GB DDR5 SO-DIMM – $220

The total is $573.99 before shipping and taxes, and doesn’t include the required 92mm fan (only $4 though). You can probably optimize the price a little bit by shopping around. The first three items must be purchased from Hardkernel or distributors, but you can potentially find a 10GbE M.2 card and 16GB RAM locally at lower prices. The full price doesn’t include the NVMe SSD, so the total should be quite over $600.

The post ODROID-H5 Review – Part 2: A dual 10GbE mini PC tested with Ubuntu 26.04 appeared first on CNX Software - Embedded Systems News.

Tanaka Masayuki’s PCMFlow722 library enables (half-duplex) two-way real-time HD voice over ESP-NOW on ESP32 boards with a speaker and a microphone, effectively transforming them into walkie-talkies.

The library implements a G.722 wideband codec add-on for PCMFlow lightweight audio decode and PCM flow library for Arduino, which already supports uncompressed PCM, MP3, and FLAC audio codecs. PCM and FLAC take too much bandwidth over ESP-NOW, and MP3 is not suitable for real-time audio, so the legacy G.722 audio codec was selected instead.

The keyword here is “HD voice,” since two-way audio over ESP-NOW was previously implemented in projects such as Atomic14’s esp32-walkie-talkie (5 years ago) and, more recently, the well-documented Adafruit ESP-NOW Walkie-Talkie project, but these typically rely on lower-quality G.711 audio or compressed audio.

The PCMFlowG722 library and G.722 codec enable HD voice with “7 kHz audio at 16 kHz sampling using the same 64 kbps wire budget as G.711 — same packet size, twice the audio bandwidth”, as explained by Tanaka. The table below compares G.711, G.722, and Opus codecs and libraries.

Opus offers lower bandwidth and full-band audio quality, but the G.722 audio codec is less complex and requires fewer resources (CPU, storage).  It’s also well suited to ESP-NOW, since the protocol carries up to 250-byte payloads, and a 20-ms G.722 voice frame at 16 kHz produces 160 bytes, the same as G.711, but with twice the audio bandwidth, while raw 16 kHz mono 16-bit PCM would require 640 bytes (G.722 compresses four times).

The library implements a G.722 encoder compressing 16 kHz PCM into G.722, and a decoder doing the reverse. If you want to try it on your own board, the EspNowTransceiver.ino Arduino sketch is probably what you want to try. It’s a half-duplex HD-voice transceiver over ESP-NOW, and the same firmware acts as both sender and receiver.

It’s been tested on the M5Stack Core2 boards with an ESP32 SoC, 8MB PSRAM, 16MB SPI flash, a small 2-inch display showing the EPS-NOW channel, a 1W speaker (1W-0928), and an SPM4123 microphone. While button A is held, the audio is broadcast over ESP-NOW to one or more Core2 devices, and in all other cases, the devices are set as audio receivers.

Via Adafruit

The post PCMFlow722 library enables two-way real-time HD voice over ESP-NOW with G.722 audio codec appeared first on CNX Software - Embedded Systems News.

Microchip has expanded its dsPIC33 lineup with the dsPIC33CK Value Line family, a new series of low-cost 16-bit digital signal controllers (DSCs) designed for motor control applications. The devices deliver up to 100 MIPS of deterministic performance and integrate high-resolution PWM and a 12-bit ADC, making them suitable for motor Field-Oriented Control (FOC) and precision sensing tasks.

These DSCs are designed to bridge the gap between basic microcontrollers and higher-end dsPIC33A devices. It offers flash memory from 32 KB to 256 KB, supports function consolidation to reduce component count and BOM cost, and is available in automotive-grade (AEC-Q100 Grade 1) versions with secure boot.

Microchip dsPIC33CK specifications:

• MCU Core – 16-Bit dsPIC33CK CPU operating up to 100 MHz / 100 MIPS and DSP with 40-bit wide accumulators
• Memory & Storage
  • 8 KB to 16 KB Data RAM
  • 32 KB to 256 KB Program Flash with Error Correction Code (ECC)
  • 384 Bytes of One-Time-Programmable (OTP) Memory
• Peripherals
  • Up to 53x GPIOs with high-current capability (GPIO count depends on package)
  • 3x UART (LIN, DMX, IrDA support)
  • 1x I2C (with SMBus)
  • 2x SPI/I2S
  • 1x CAN FD
  • SENT interface (Automotive industry)
  • Peripheral Pin Select (PPS)
  • 4-channel DMA controller
• Analog
  • 12-bit ADC, up to 2 Msps, 20 channels, per-channel buffers, oversampling, and digital comparators
  • 12-bit DAC with slope compensation
  • Analog comparator (~30 ns)
• Timers
  • 16-bit general-purpose timer
  • SCCP (capture/compare/PWM) with 16/32-bit operation and ~2.5 ns high-resolution PWM
  • Peripheral Trigger Generator (PTG)
• Clock
  • Internal 8 MHz FRC oscillator
  • External crystal/resonator or clock input
  • Programmable PLLs
  • Fail-Safe Clock Monitor (FSCM)
  • Reference clock output (REFCLKO)
• Security
  • Flash ECC and RAM MBIST
  • Dual Watchdog Timer (WDT)
  • Windowed Deadman Timer (DMT)
  • CodeGuard security
  • Clock monitoring and backup oscillator
  • CRC module
• Debug
  • JTAG boundary scan
  • In-circuit programming/debugging (ICSP)
  • Trace buffer and breakpoints
• Power
  • Operating voltage – 3.0V to 3.6V
  • Sleep, Idle, and Doze modes
  • Power-on Reset (POR) and Brown-out Reset (BOR)
• Packages
  • 28-pin SSOP
  • 36-pin UQFN
  • 48-pin TQFP
  • 64-pin TQFP
• Temperature – -40°C to +125°C (AEC-Q100 Grade 1)

On the software side, the dsPIC33CK Value Line DSCs are supported by the MPLAB X IDE, along with the MPLAB XC-DSC compiler and MPLAB Code Configurator, similar to other Microchip MCUs and DSCs. The company mentions that the DSCs can be used with touch/HMI, advanced sensing, consumer appliances, industrial automation, and entry-level automotive (sensors/actuators).

Microchip offers a Curiosity Nano low-cost development board designed for rapid prototyping of motor control applications with the dsPIC33CK256MC005 DSC. It features a 100 MHz DSC with 256 KB ECC Flash, 16 KB RAM, a 12-bit 2 MSPS ADC, DAC, comparator, and high-speed PWM, along with an onboard debugger for programming and debugging without external tools. The board can be expanded using Click boards, touch adapters, and other Curiosity Nano accessories for motor control, sensing, and embedded applications.

The company also offers the dsPIC33CK256MC006 Motor Control DIM, a dual-in-line module designed for dsPIC33CK Value Line DSCs for motor control applications. It features a 64-pin dsPIC33CK256MC006 and supports external inverter boards for motor current feedback, enabling development and testing with Microchip’s MCS inverter platforms such as MCLV-48V-300W and MCHV-230VAC-1.5kW. The module is designed for rapid prototyping of motor control designs, but the company mentions it is not compatible with older PIM-based inverter boards.

The dsPIC33CK Value Line DSCs, along with the Curiosity Nano Evaluation Kit and Motor Control DIM, can be purchased directly from Microchip. The DSCs start at $0.51, while the development boards are priced at $9.90 and $12.99, respectively. Interestingly, the company mentions that the “pricing is consistent regardless of order volume”. More details can be found on the product page and in the press release.

The post Microchip dsPIC33CK low-cost motor control MCU sells for $0.51 and up appeared first on CNX Software - Embedded Systems News.

XPU Labs’ Petros CH32H417M Alef is a Raspberry Pi Pico-sized board based on the WCH CH32H417M RISC-V USB 3.0 microcontroller and taking a 2MP OV2640 camera module through the MCU’s digital image interface (DVP).

The board comes with 896KB SRAM and 960KB Flash from the WCH microcontroller, two 20-pin GPIO headers following the Raspberry Pi Pico’s pinout, a 6-pin SWD and UART6 header for debugging, and a Reset button.

Petros CH32H417M Alef specifications:

• MCU – WCH CH32H417MEQ6
  • MCU
    • QingKe RISC-V5F up to 400 MHz
    • QinKe RISC-V3F up to 144 MHz
  • GPU – Graphics Processing Hardware Accelerator GPHA
  • Memory – 896KB SRAM
  • Storage – 960KB Flash
• Camera I/F – 40-pin B2B connector with DVP @ 144MHz, SPI, I2C and ADC interfaces for 2MP Phos Ayin OV2640 camera module
• USB – 1x USB 3.0 Type-A port (5Gbps, tested up to 430 MB/s) for data, power, and firmware flashing
• Debugging – 6-pin SWD and UART6 header
• Expansion – 2x 20-pin headers compatible with Raspberry Pi Pico headers
• Misc
  • Reset button
  • Power LED
• Power Supply
  • 5V via USB 3.0 port or pin header
  • 3.3V/1A LDO
• Dimensions – 52mm × 21mm

The documentation page mostly comes with hardware information, and the user guide, code samples, documents, and 3D drawing section are only placeholders at this stage. So it’s not ideal for now, and I could not find any video demo, except for a USB 3.0 transfer test (430 MB/s) without the camera module. Hopefully, it will show as a standard USB UVC (USB Video Class) webcam when connected to a host once the code is ready. The SDK for the WCH CH32H417 development board comes with DVP and USB 3.0 code samples for the MounRiver SDK, and I also noticed a USB SS0 (SuperSpeed) UVC DVP sample, which could work on the Petros board, possibly with a few modifications.

Contrary to the Muselab nanoCH32H417 board with USB 3.0 and Fast Ethernet ports (but no DVP camera interface), the Petros CH32H417M Alef lacks an onboard Link-E debugger, and you can add it through the Amnos LinkE Alef debug board, which I assume connects to the SWD/UART6 header.

The CH32V417M USB 3.0 camera board is sold for $12.99 on AnalogLamb. However, that price is only for the RISC-V board, and most people will likely prefer to go with the $19.99 bundle (board + camera module), or the $23.99 bundle adding a debug board.

The post Petros CH32H417M Alef – A Raspberry Pi Pico-sized RISC-V USB 3.0 camera board appeared first on CNX Software - Embedded Systems News.

u-blox has recently announced the ALMA-B2 standalone BLE 6.0 and 802.15.4 module family built around the Nordic Semi nRF54LM20 Cortex-M33 wireless microcontroller with a dedicated NPU for low-latency Edge AI applications.

There are four specific product variants in the u-blox ALMA-B2 series: ALMA-B201, ALMA-B206, ALMA-B211, and ALMA-B216, all of which support Bluetooth 6.0 and Bluetooth Channel Sounding for distance measurement. They also support IEEE 802.15.4, including Thread, Zigbee, and Matter, as well as Nordic’s proprietary 2.4 GHz protocol and NFC. The company also mentions that the nRF54LM20B-based ALMA-B211 and ALMA-B216 variants include an Axon NPU that performs machine learning tasks up to 15 times faster and with greater energy efficiency than running the same tasks on the main processor alone.

u-blox ALMA-B2 Series Specifications:

• Wireless SoC – Nordic Semiconductor nRF54LM20A (for ALMA-B201 and B206) or nRF54LM20B (for ALMA-B211 and B216)
  • CPU
    • Arm Cortex-M33 application processor clocked at up to 128 MHz
    • RISC-V coprocessor running at 128 MHz
  • Memory – 512 kB of RAM
  • Storage – 2 MB of Non-Volatile Memory (NVM)
  • AI accelerator – Built-in Axon NPU (only for ALMA-B211 and ALMA-B216)
  • Wireless
    • Bluetooth LE 6.0 with PHY rates of 125 kbit/s, 500 kbit/s, 1 Mbit/s, and 2 Mbit/s
    • Bluetooth Channel Sounding and Bluetooth LE long range (coded PHY)
    • IEEE 802.15.4 supporting Thread, Zigbee, and Matter
    • Nordic Proprietary 2.4 GHz protocol and NFC
    • +10 dBm maximum radiated output power (EIRP)
• Antenna – External antenna connector ALMA-B201 and ALMA-B211 or PCB antenna for ALMA-B206 and ALMA-B216
• I/Os via LGA pads
  • Up to 66 GPIO
  • 1x USB High Speed
  • 1x HS-SPI + 6x SPI
  • 6x TWI/I2C
  • 1x HS-UART + 6x UART
  • 3x PWM
  • SAADC (14-bit)
  • PDM
  • QDEC
  • 7x timers
• Security – Arm TrustZone, secure boot, secure key management, secure FOTA, hardware crypto-accelerator, physical tamper detection, and debug access port protection
• Power
  • Supply – 1.71 to 3.60 V
  • Consumption – 4.8 mA in Active TX (@ 0dBm), 4 uA in Standby, and 700 nA in Sleep mode
• Dimensions
  • 10.4 x 11.2 x 1.9 mm (Antenna pin variants)
  • 10.4 x 14.3 x 1.9 mm (Internal PCB antenna variants)
• Weight – < 1.0 gram
• Operating Temperature – -40°C to +85°C

The ALMA-B2 modules use an open CPU architecture. This allows users to write and run their own applications directly on the module using the built-in Arm Cortex-M33 processor. So, there is no need for an external host MCU, which saves BOM cost and PCB space, and development is done with Nordic’s nRF Connect SDK.

The company also introduced two EVK variants in their announcement – the EVK-ALMA-B211 (external antenna) and EVK-ALMA-B216 (internal PCB antenna), but at the time of writing, dedicated product pages or user guides for the boards are not yet published on the u-blox website (they usually appear a bit later when samples become available).

Early samples of the ALMA-B2 series will be available in the coming weeks. More details, block diagrams, and technical documents can be found on the u-blox product page, and some additional information is available in the press release.

The post u-blox ALMA-B2 Bluetooth 6.0 and 802.15.4 module features Nordic nRF54LM20 Edge AI wireless MCU appeared first on CNX Software - Embedded Systems News.

The Qualcomm Snapdragon C (Compute) Arm SoC is designed for entry-level Arm Windows laptops, or “mobile PCs” in marketing speak, with prices starting at $300, which should compete against entry-level laptops using Intel Alder Lake-N or Twin Lake processors.

The company didn’t provide many technical details, but we do know the Snapdragon C comes with an NPU for AI acceleration, and said it’s based on Kryo mobile cores rather than the more recent Oryon cores at a media event. Qualcomm still promises all-day battery life, smooth web browsing, video streaming, and productivity for students, families, and small businesses.

I’m not sure what kind of announcement that is, as Qualcomm didn’t provide any specifications. However, they did mention Acer, HP, and Lenovo will offer Snapdragon C laptops, and we have some specifications for the upcoming Acer Aspire Go 15 (AG15-Q31P).

• SoC – Qualcomm Snapdragon C
  • CPU – Unknown number of Kryo cores at an unknown frequency…
  • GPU – Qualcomm Adreno GPU
  • AI accelerator – Built-in NPU with unspecified performance…
• System Memory – Up to 8GB RAM
• Storage – Up to 512 GB storage
• Display – 15.6-inch Full HD (1920×1080), 16:9 aspect ratio
• Video Output – HDMI 1.4 port
• Camera – 1080p FHD webcam
• Networking – Wi-Fi 6E and Bluetooth 5.4 or above
• Audio
  • 3.5mm audio jack
  • Dual speakers (and I assume a microphone somewhere)
• USB
  • 2x full-function USB Type-C ports
  • 1x USB Type-A port
• Misc – Copilot key, AcerSense, Acer My Key
• Battery – 53 Wh battery
• Dimensions – TBC
• Weight – TBC

The Acer Aspire Go 15 will run Windows 11 Home. Those will be useful for browsing the website, watching videos, checking emails, and running office suites, but like other entry-level laptops, you won’t be able to run the latest AAA games, and it won’t be ideal for video editing or running local AI models.

This was probably a rushed announcement for Computer 2026, which explains why details are rather light. Availability is also unclear, with Qualcomm and Acer simply stating that Snapdragon C devices are expected to hit the shelves later this year.

The post Qualcomm Snapdragon C to power cheaper, entry-level Arm Windows laptops selling for $300 and up appeared first on CNX Software - Embedded Systems News.

AMD has added three new chips to its Versal Prime Series Gen 2 lineup: the Versal 2VM3454, 2VM3254, and 2VM3104. Designed for space-constrained applications like Pro AV, broadcast, and industrial IoT, these new devices deliver up to 100K DMIPS of scalar compute in packages as small as 23 x 23 mm.

AMD started shipping the first production units of the Versal Prime Gen2 Series with the 2VM3858 device late last year. The 2VM3558 has since entered full production, while the 2VM3358 is currently sampling. These new devices are designed to provide an optimized footprint and processing subsystem compared to the earlier models. Despite the reduction in core count, AMD claims these devices can deliver up to 5x the scalar compute performance compared to existing AMD adaptive SoCs.

AMD Versal Prime Series Gen 2 (2VM3104, 2VM3254, and 2VM3454) specifications:

• Processor Subsystem (PS)
  • APU – Quad-core Arm Cortex-A78AE with 64 KB L1 Cache (with parity & ECC), 512 KB L2 Cache, and 2 MB L3 Cache per cluster
  • Real-time Processor – Hexa-core (6x) Arm Cortex-R52 with 32 KB L1 Cache (with ECC) and 128 KB TCM (with ECC)
  • GPU – Integrated single-core Arm Mali-G78AE GPU
  • Video Codec Unit (VCU) – 1x hardened tile on the 2VM3254 and 2VM3454 supporting HEVC & AVC encode and decode up to 4K60, 4:4:4, 12-bit (not available on 2VM3104)
  • Memory – 1 MB On-Chip Memory with ECC
• Programmable Logic (PL) and DSP
  • System Logic Cells – Up to 564,760 (2VM3454)
  • LUTs – Up to 258,176 (2VM3454)
  • DSP Engines – Up to 1,140 (2VM3454)
• Memory
  • DDR5 memory controllers up to 6400 Mb/s and LPDDR5X up to 8533 Mb/s; up to 102 GB/s maximum bandwidth
  • Total PL Memory – Up to 45.4 Mbit
• Storage – UFS 3.1 support
• Display – DisplayPort 1.4
• Networking
  • Up to 2x 100Gbps Multirate Ethernet MAC on the 2VM3454 (1x on 2VM3104 and 2VM3254)
  • 10Gbps Ethernet and 1Gbps Ethernet
• USB – USB 3.2 and USB 2.0 support
• Other I/O – Up to 16x GTYP transceivers (PL-Only) up to 32 Gbps (4x on 2VM3104; 8x on 2VM3254)
• Expansion – PL-based PCIe Gen5x4 controller on the 2VM3254 and 2VM3454 (not available on 2VM3104), plus PS-based PCIe Gen5 x4 support
• Security
  • Application Security Unit and general security enhancements
  • High-speed crypto engine (HSC) available on the 2VM3454
• Packages
  • 23×23 mm – 2VM3104, 2VM3254
  • 29×29 mm – 2VM3454

The combination of Cortex-A78AE and Cortex-R52 cores helps these SoCs handle complex tasks and real-time control more easily. The programmable logic also enables developers to process sensor data quickly and in real time. AMD has also added various safety features (ASIL D/SIL 3), eliminating the need for a separate safety controller and reducing system size. Developers can still update algorithms or video processing later without changing hardware.

The 2VM3654 SKU was released earlier, but it also shares the same footprint, so the new chips can be installed in older PCBs very easily. AMD also notes that the new small devices offer more programmable logic per square millimeter than comparably sized 8-core devices.

On the software side, AMD provides Vivado for RTL design (Verilog/VHDL) and Vitis for building software on the Arm Cortex-A78/R52 processor cores. AMD also maintains the Embedded Development Framework (EDF) based on Yocto, and including ready Linux images, drivers for video, GPU, and fast memory. You can also run simple bare-metal programs on the real-time cores.

Early Access design tools are already available for the Versal 2VM3654, with the 2VM3454 expected to begin sampling later in 2026. The more compact 2VM3254 and 2VM3104 are expected to become available in 2027. Full production tool support will follow accordingly. Additional information can be found on the Versal Prime Series Gen 2 product page and in the press release.

The post AMD launches Versal Prime Gen 2 2VM3454, 2VM3254, and 2VM3104 adaptive SoCs in compact 23x23mm packages appeared first on CNX Software - Embedded Systems News.

AAEON MIX-PTLWV1 is an industrial mini-ITX motherboard based on Intel Core Ultra Series 3 “Panther Lake” processors with up to two 10GbE and two 2.5GbE RJ45 jacks, and four 4K-capable DisplayPort video outputs.

The motherboard also supports up to 128GB DDR5 C/SO-DIMM memory, M.2 NVMe SSD storage, optional WiFi and Bluetooth, and/or 4G LTE/5G cellular via M.2 modules, a PCIe Gen4 x8 slot for expansion, six serial interfaces, 12V-24V wide DC input, and more.

AAEON MIX-PTLWV1 specifications:

• SoC – Intel Panther Lake processor (PTL-H404/PTL-H484 families) with Intel Xe LPG graphics; TDP: 15 to 25W
• System Memory – Up to 128GB via 2x 262-pin DDR5 C/SO-DIMM sockets up to 6400MHz (H404 SKU) or 7200MHz (H484 SKU)
• Storage – M.2 2280 M-Key (PCIe Gen4 x4) socket for NVMe SSD
• Display Interfaces
  • 4x DisplayPort 1.4 up to 4096 x 2160 @ 60Hz
  • 40-pin LVDS/eDP co-layout connector for
    • Default: 18/24 bit dual-channel LVDS up to 1920 x 1080 @ 60Hz
    • 4-channel eDP1.4 up to 4096 x 2160 @ 120Hz (by BOM)
  • eDP/LVDS Tcon IC voltage select
  • eDP/LVDS panel backlight voltage select (3.3V/5V/12V)
  • LVDS Inverter
  • Quad independent display support
• Audio
  • Realtek ALC897 HD audio codec
  • Line-out jack
  • 4-pin header for 2W/4Ω speakers
• Networking
  • 2.5GbE RJ45 port via Intel I226-V 2.5Gbps Ethernet controller
  • 2.5GbE RJ45 port via Intel I226-LM 2.5Gbps Ethernet controller
  • SKU 4L only – 2x 10GbE via Intel E610-XAT2 10Gbps Ethernet controllers
  • Optional WiFi and Bluetooth via M.2 Key-E socket
  • Optional 4G LTE/5G cellular via M.2 Key-B socket and SIM card slot
• USB
  • 4x USB 3.2 Gen1 (5Gbps) Type-A ports
  • 2.0mm pitch 20-pin header with 2x USB 3.2 Gen 1 interfaces
• Serial
  • 1x RS-232/422/485 box header (5V/12V/RI optional)
  • 5x RS-232 box headers
• Expansion
  • PCIe Gen 5 x8 open edge slot (BoM optional on H484)
  • M.2 2230 E-Key socket (PCIe 4.0 x1/USB 2.0/CNVi)
  • M.2 3042/3052 B-Key socket (PCIe x1/USB 3.2 Gen 1) + Nano SIM Socket; default 3052 by BOM (optional)
  • 8-bit programmable DIO
• Security – Infineon SLB9672 TPM2.0 chip
• Misc
  • Watchdog Timer
  • AMI UEFI/BIOS on 256Mbit SPI Flash ROM
  • H/W Monitor – Temperature monitor on CPU/System, voltage monitor on Vcore/5V/3.3V/12V, fan monitor on chassis
  • I/O Chipset –Nuvoton NCT6126D
  • 4-pin CPU fan connector, 4-pin SYS fan connector
  • Wake-On-LAN (WoL) and PXE  support
  • Power State S4, S5
  • Lithium battery for BIOS
  • CMOS jumper
  • Front panel header
  • ME disable
• Power Supply – 12 to 24V DC input via 4-pin connector; ATX/AT mode
• Dimensions – 170 x 170 mm (mini-ITX form factor)
• Weight – TBC
• Temperature Range – Operating: 0 to 60°C; storage: -40 to 85°C
• Operating Humidity – 60°C @ 90% relative humidity, non-condensing
• Certifications – CE & FCC (Class A)

AAEON provides support for Windows 11 64-bit and Ubuntu 24.04 with Linux kernel 6.8 for the board. Note that there are two SKUs: MIX-PTLWV1-A10-4L and MIX-PTLWV1-A10-2L, with the only difference being the two 10GbE RJ45 ports on the 4L model.

While the MIX-PTLWV1 is AAEON’s first industrial motherboard powered by the Intel Core Ultra Series 3 “Panther Lake” processor family, Avalue previously introduced the EMX-PTLP thin mini-ITX motherboard with up to an Intel Core Ultra 7 358H SoC, and BCM Advanced Research is working on the MX-PTL motherboard with similar features.

AAEON did not provide price information for the MIX-PTLWV1 motherboard, and the product page is still marked as “preliminary”. It will probably launch soon, as the Taiwanese company will be demonstrating the Panther Lake board at the ITS America Conference & Expo 2026 on June 9-12, and use it or another Panther Lake platform, for two demos, one leveraging “smart intersection technology” combined with Vision Language Models (VLMs) for advanced traffic event monitoring, and the other relying on agentic AI for advanced route planning.

The post AAEON MIX-PTLWV1 Panther Lake industrial mini-ITX motherboard offers quad DisplayPort, dual 10GbE, and more appeared first on CNX Software - Embedded Systems News.

The HW1 by Gesture Platforms is a 10-degree-of-freedom (DOF) high-dexterity robotic hand and wrist built around an ESP32-S3 wireless MCU. It’s primarily designed for researchers, educators, and hobbyists; it bridges the gap between basic DIY robotic hands and expensive industrial models.

The device weighs just around 500 grams but can handle a 1kg dynamic load and a 3kg static load. It communicates via USB-C or Bluetooth 5.0 and comes with a companion desktop app as well as Python and C++ SDKs for custom development.

Gesture HW1 specifications:

• Microcontroller – Espressif Systems ESP32-S3 dual-core LX7 microprocessor with WiFi 4 and Bluetooth 5.0 LE connectivity
• Degrees of Freedom (active) – 10 DOF
  • Finger Flexion – 0° – 90° (at each joint)
  • Finger Adduction/Abduction – -10° – 40°
  • Thumb Flexion/Extension – 0° – 120°
  • Thumb Adduction/Abduction – 20° – 90°
  • Thumb Distal Flexion – 0° – 90°
  • Wrist Flexion/Extension – -50° – 50°
  • Wrist Radial/Ulnar Deviation – -10° – 40°
• Joints – 19 joints total
• Payload – 1kg dynamic / 3kg static load
• Accuracy/Repeatability – ±1mm
• Control Speed – 100Hz
• Sensors – Angle (at motor), Current, Temperature
• USB – 1x USB Type-C port for communications
• Misc
  • 4 x user-programmable RGB LEDs
  • Four-way stretch fabric cover prevents debris from entering joints
• Power – Via the XT60 connector on the side
• Dimensions – ~275mm total height (fits men-small / women-medium gloves)
• Weight – ~500 grams
• Mounting – M3x0.5 threads (depth of 4.5mm)

The HW1 uses aluminum gears at key joints like the wrist for better strength and durability, while its skeleton is made from lightweight, strong glass and carbon-fiber-filled polycarbonate, and the outer shell uses ASA plastic to handle heat and rough use. The fingertips are covered with soft silicone (Shore 30A) to improve grip. Gesture Platforms states that it will provide STEP files, so users can 3D print, modify, or replace the outer parts easily. Most probably, they will release the file after the campaign ends.

On the software side, the company provides a plug-and-play desktop application for Windows 10/11. The app also works completely offline and features four control modes, which include Text (importing .csv files), Timeline (keyframe-style editing), Controller (mapping to a standard game controller), and Stream (real-time input). For more advanced users, C++ and Python SDKs will be available to integrate the hand into custom machine learning and control algorithms.

We have previously written about a range of ESP32-based robotic arms, such as the RoArm-M3-Pro and RoArm-M3-S, the Tobor open-source modular robotic arm, and the MyCobot Robotic Arm, but we haven’t covered robotic hands as much. We did cover the Amazing Hand, an 8-DoF 3D-printable DIY open-source robotic hand, and on the other side of the spectrum, we mentioned the high-end Dex5 dexterous hand in our article about the Unitree R1-A5 and R1-A7 humanoid robot. The Gesture HW1 offers something in the middle in terms of price and functionality. 

The Gesture HW1 ESP32-S3 high-dexterity robotic hand is currently crowdfunding on Kickstarter. The “Launch Day Special” reward requires a $699 pledge (limited to 100 units), while the standard “Early Backer” tier is available for $849. A double pack for both the left and right hands is available for $1,599. Shipping is not included and will be charged after the campaign; deliveries are estimated to begin in November 2026.

The post Gesture HW1 is a 10-DOF ESP32-S3 robotic hand with high-dexterity manipulation (Crowdfunding) appeared first on CNX Software - Embedded Systems News.

Secluso is a private, open-source, DIY home security camera system built around the Raspberry Pi Zero 2 W, featuring true end-to-end encryption (E2EE) and on-device AI for human, pet, and vehicle detection. It was designed as an alternative to commercial smart home cameras that require sending raw video feeds to a proprietary cloud, a practice that often raises significant privacy concerns.

Developed by Secluso, Inc., co-founded by UC Irvine professor Ardalan Amiri Sani and John Kaczman, the project utilizes Messaging Layer Security (MLS, RFC 9420) to ensure end-to-end encryption between the camera and the user’s smartphone. Because the system uses an untrusted relay (either self-hosted on a VPS or via Secluso’s free beta relay), the server routing the footage only sees encrypted files and cannot decrypt the video or thumbnails.

Secluso hardware requirements:

• SBC – Raspberry Pi Zero 2W
• Camera – Raspberry Pi Camera Module V1 (OV5647) or V2 (IMX219)
• Audio & Sensors –  HAT with a microphone and safety temperature sensor (included in official kits)
• Enclosure – Official Raspberry Pi Zero enclosure, 3D-printable custom housing, or Secluso’s IR-pass acrylic cover

The software focuses on security and privacy. The core camera hub and server software are now written in Rust rather than in OpenSSL-based C code, making them safer and less prone to memory-related bugs. It also uses post-quantum encryption to protect data even from future attacks where hackers might try to decrypt stored data later. All parts of the system, including Secluso OS and the mobile apps, use reproducible builds, so anyone can verify that the software matches the public source code, and firmware updates are only installed if they come from signed and trusted GitHub releases. The mobile app is available on the iOS App Store and Google Play. Android users can also use Obtainium to download the app directly from GitHub’s release page. For notifications, the system doesn’t rely on Google services and uses its own relay for iOS and supports UnifiedPush on Android, so Firebase Cloud Messaging is optional.

Despite all these features, the system is designed to be deployed in just 5 minutes as the v1.0.2 release introduces Secluso OS, a minimal, Yocto-based Linux image. Users simply flash the image using the Secluso Deploy desktop utility available for Windows, macOS, and Linux. The utility automatically handles injecting unique credentials, configuring the relay over SSH, and setting up the camera without requiring the user to open a terminal. You can find the source code, build instructions, and the detailed security whitepaper on their GitHub released under a GPL license, and the “Build Your Own Guide” explains how to build your own camera. Users can fully self-host the open-source relay/server on their own VPS, so no cloud is required.

For those who don’t want to get into ordering different parts from different vendors and setting up their own server, Secluso is opening pre-orders this month (May 2026). They are offering a limited run of 100 DIY Kits for $50 (which include the night vision camera module, custom mic/temp HAT, and housing parts, and 1 year of Secluso Cloud for free) with an estimated delivery of 2–3 months. A fully assembled Plug and Play version is also available for $100, with an estimated delivery of 4–6 months and a free year of Secluso Cloud. After the included free year, Secluso Cloud is available optionally for $5/month or $50/year. Please note that pre-orders will ship to the US only initially.

Via Hackster.io

The post Privacy-focused, open-source Raspberry Pi Zero 2W DIY security camera offers end-to-end encryption, on-device AI appeared first on CNX Software - Embedded Systems News.

VIA Labs VL610/VL610D MST (Multi-Stream Transport) Hub controllers are designed for multi-display expansion, enabling USB-C adapters to drive up to three 4K UHD high-resolution HDMI or DisplayPort displays simultaneously.

The chips succeed the VL605 single-channel USB-C to HDMI 2.1 signal converter. The VL610 supports three video outputs up to 4Kp60, while the VL610D is limited to two video outputs for different docking design needs.

VIA VL610/VL610D key features and specifications:

• Display/USB Input – DisplayPort Receiver (DPRX) supporting HBR3 (High Bit Rate 3) up to 32.4 Gbps, suitable for up to 8Kp60 and D+/D- USB 2.0
• Video and Audio output
  • VL610
    • HDMI 2.1 FRL transmitter
    • Configurable port supporting DP++ or HDMI 2.1 FRL
    • Configurable port supporting DP++ or HDMI TMDS
  • VL610D – Dual display configuration (not clear details provided)
  • Single display up to 8Kp60 or 4Kp240
  • Triple display up to 4Kp60 or QHDp144.
  • Up to 6x independent audio and video streams with a single DP output supporting up to four MST streams.
  • Full support for color formats and audio
• Misc features
  • DSC 1.2a decoder for decompression to HDMI output or direct pass-through to compatible displays.
  • Variable Refresh Rate (VRR)
  • ECDSA-256 asymmetric authentication
  • Logo Bitmap display capability for brand logos, warning messages, or guidance screens
• Package – QFN100 (10x10mm) for VL610

There’s no product page yet, and VIA only published a press release for COMPUTEX 2026, and I also found an earlier presentation with the slide above. As I understand it, the VL610/VL610D chips are designed specifically for multiple-display DP 2.1 HBR3 MST Hubs, and USB data is limited to High Speed (480 Mbps), so they won’t be found in multi-function USB-C docks.

Thanks to TLS for the tip.

The post VIA Labs VL610/VL610D MST Hub controllers to enable USB-C multi-display adapters with up to three 4K display outputs appeared first on CNX Software - Embedded Systems News.