Mini PCs and embedded computers: what they are for

A computer the size of a deck of cards, or tucked inside a keyboard, that stays on day and night without making a sound. Or screwed onto a DIN rail in an electrical panel, talking to meters and machines. "Mini PC" and "embedded computer" are broad terms covering very different things. The easiest way to find your way around is to start from what you want to do with one.

A small desktop computer

The Raspberry Pi 500 is a complete computer inside a keyboard, with an Italian layout in our catalogue: a quad-core Arm Cortex-A76 processor at 2.4 GHz, 8 GB of RAM, two micro HDMI outputs up to 4K at 60 Hz and the 40-pin GPIO header on the back for electronics projects. Plug in a monitor, a mouse and the power supply and you have a PC for school, for learning to code, for browsing and writing. The desktop kit also includes the power supply, a mouse and an HDMI cable.

The Raspberry Pi 5 has the same processor on a bare board: Raspberry Pi says it is up to three times faster than the Pi 4. It has two 4K HDMI outputs, Gigabit Ethernet and a PCIe interface for connecting an SSD. One detail matters on power: Raspberry Pi recommends a 5 V, 5 A USB-C power supply, such as its own 27 W unit. With that, the USB ports can deliver 1.6 A in total, useful for an SSD or an external drive; with a 3 A supply the limit drops to 600 mA, which suits a keyboard, a mouse and other low-power peripherals.

Boards to get started with are in Development Boards.

A small server that is always on

This is the most common use at home. Home Assistant, the free home automation software, lists the Raspberry Pi 5 and 4 among the systems to install it on, with at least 2 GB of RAM, an A2-class microSD card of at least 32 GB, and an Ethernet cable, which is required for installation: afterwards Home Assistant can also work over Wi-Fi, but it highly recommends Ethernet as more reliable. The same Raspberry Pi can run an MQTT broker such as Mosquitto, to which ESP32 boards send their readings: we explain it in the guide MQTT with the ESP32.

For a server that runs all year round it is better to boot the system from an SSD rather than the microSD card. With the M.2 HAT+ board you write the system to the NVMe SSD with Raspberry Pi Imager and, with no microSD card inserted, the Raspberry Pi 5 boots from it on its own, as we describe in M.2 SSDs: NVMe or SATA, enclosures and adapters.

If you need an x86 processor, for Windows software for instance, there is the LattePanda Mu: a compute module with an Intel N100, 8 GB of RAM and 64 GB of eMMC that mounts on a carrier board. LattePanda states it works with Windows and Linux, with a thermal design power (TDP) adjustable from 6 to 35 W, and among its example uses lists a software router, a NAS and Proxmox clusters.

Machine vision and artificial intelligence

Recognising people, cars or parts on a conveyor belt takes a lot of computing, and this is where accelerators come in. Their power is measured in TOPS: one TOPS is a trillion operations per second.

The simplest route starts with the Raspberry Pi 5 and the AI HAT+, at 13 TOPS with the Hailo-8L chip or 26 TOPS with the Hailo-8. Raspberry Pi has built it into its camera software: object detection, image segmentation and pose estimation run on the accelerator. The Seeed reComputer AI R2140-12 combines a Raspberry Pi 5 and a Hailo-8 in a case, ready to go.

A step up are the computers based on NVIDIA Jetson. For the reComputer J3011, with an 8 GB Jetson Orin Nano, Seeed states up to 40 TOPS with the software it ships with; with NVIDIA's "Super" mode, which requires JetPack 6.2, the figure NVIDIA states for the module rises to 67 TOPS. All the AI hardware is in Artificial Intelligence.

In electrical panels

Here the computer has no monitor or keyboard: it sits on a DIN rail and talks to machines through industrial ports. Seeed's EdgeBox RPi 200 has RS485, RS232, digital inputs and outputs, and works with Modbus, OPC UA and IEC 61131-3 programming software such as Codesys: in practice it acts as a PLC and as a gateway to the network. It runs on low voltage, but the wiring in the electrical panel must be done by a qualified person.

Touch panels: the machine's display

Seeed's reTerminal is a Compute Module 4 with a 5-inch 1280 × 720 touch screen, dual-band Wi-Fi and Bluetooth 5.0: Seeed suggests it for IoT and home automation projects. The reTerminal DM is its industrial sibling, with a 10.1-inch screen, an IP65 front panel, built-in Node-RED and RS485 and CAN ports; it mounts in a panel, on a VESA mount or on a DIN rail (the panel and DIN rail kits are sold separately). The whole family is in Mini PCs and embedded computers.

Windows on a board

The LattePanda V1 is an x86 computer on an 88 × 70 mm board: Intel Z8350 (Cherry Trail), 4 GB of RAM, 64 GB of storage, and it runs full Windows 10. It also has an ATmega32u4 microcontroller, the same as on the Arduino Leonardo, to read sensors and drive actuators. DFRobot suggests it for portable instruments, gateways, robotics and IoT. Note that the version in our catalogue comes with Windows 10 Home not activated, so the licence has to be bought separately.

Three questions before you choose

Which operating system do you need: Linux (Raspberry Pi OS and other distributions) or Windows on x86? Where will it go: on a desk, in a cupboard at home or in a panel with a 24 V supply? And what does it have to do: a light service that is always on, machine vision, an operator panel? For Home Assistant or an MQTT broker, a Raspberry Pi 5 with an SSD. To recognise objects from a camera, a Raspberry Pi 5 with the AI HAT+ or a Jetson. To read Modbus meters in a panel, the EdgeBox RPi 200.

Got a project in mind?

Tell us what the computer has to do, where you will install it and which devices it needs to talk to: we will point you to the right model and the accessories (power supply, SSD, enclosure, antennas).