Learning Linux internals & drivers
See how the kernel, device tree and drivers fit together. What matters: strong docs, a serial console header and GPIO pins to drive real hardware.
Free guide - for learners
Boards compared by what you want to build, not by spec sheets. Six use cases, the capabilities that matter for each, and the four checks to run before you spend money.
Get the checklist
Start with the project
The wrong question is which board is best. The right question is what you want to build. Find your project below; each names the capabilities that matter and the two boards that fit.
See how the kernel, device tree and drivers fit together. What matters: strong docs, a serial console header and GPIO pins to drive real hardware.
A board that runs a service all day. What matters: 4 GB+ RAM, wired network, and reliable NVMe/USB storage with ready-made images.
Capture and process images. What matters: a working CSI camera path and enough compute for the models you plan to run.
Talk to motors and sensors, sometimes with hard timing. What matters: plenty of GPIO, PWM, I²C, SPI and real-time-capable hardware.
Build your own image or run a kernel you compiled. What matters: mainline kernel support and a maintained Yocto layer.
Run vision models and small ML on the board itself. What matters: an NPU or GPU with a real, supported software stack — not just a TOPS number.
The shortlist
Every board here is readily available in 2026 and beginner-friendly. Prices are approximate and move around — check before you buy.
| Board | Best for | Approx price | Community | Standout | Watch out for |
|---|---|---|---|---|---|
| Raspberry Pi 5 | The default first board, most projects | ~$50–80 | Highest | Ecosystem, 40-pin GPIO, PCIe via HAT | Street prices volatile |
| Raspberry Pi Pico 2 | Real-time companion / first MCU | ~$5 | High | $5 microcontroller for timing-critical I/O | Not a Linux computer |
| BeagleBone Black | Drivers, robotics, Yocto learning | ~$50–70 | High (learning) | 2× PRUs for hard real-time; open hardware | 2013-era SoC, 512 MB RAM |
| BeagleY-AI | Modern open edge-AI + real-time | ~$70–86 | Growing | 4 TOPS + Cortex-R5, Wi-Fi 6, PoE, PCIe3 | Younger ecosystem than Pi |
| Radxa ROCK 5B | Home server / NAS / media / desktop | ~$150+ | Medium | RK3588, 2.5 GbE, PCIe 3.0 NVMe, 6 TOPS | Vendor kernel; more forum work |
| Jetson Orin Nano Super | Serious vision & edge AI | ~$249 | Medium | 67 TOPS, CUDA, Isaac ROS | Steeper setup; not a desktop swap |
| Raspberry Pi AI HAT+ | Add edge AI to a Pi 5 | ~$70 / ~$110 | High | Hailo NPU (13 / 26 TOPS) in the Pi camera stack | Needs a Pi 5 host |
Before you pay
Run these in order for any board on your shortlist. They take less than an hour and prevent the most common first-board regrets.
Search the exact board name together with the thing you want to build. If results are thin, pick the board with the larger community — answers you can find beat performance you cannot use.
1–2 GB RAM is enough headless; aim for 4–8 GB for a desktop or heavier services. Check how it boots — SD, eMMC, USB — and prefer a board that boots from more than one.
Most mystery crashes on a first board are the power supply. Use the one the maker specifies, and budget for case, cooling and cables — accessories often add a third to the price.
Prefer a board with a serial console header. When the screen stays black, the serial console shows what actually happened — and learning to use it is the first professional skill this hobby teaches.
Enter your email and we send the complete first-board checklist — the six use-case picks, the four pre-buy checks and the day-one kit — laid out to print and keep.
Questions