Before this:SBC vs microcontroller vs PC
Picking a board
Key takeaways A spec sheet answers four questions in order: CPU (how many cores, what generation — decides real-time decode capacity), RAM (soldered forever — 4 GB is the comfortable middle), USB and I/O bandwidth (the sleeper spec — an SDR streams samples continuously and some boards bottleneck all USB through one controller), and networking (wired gigabit Ethernet beats Wi-Fi for an appliance). Match the board to the job’s sustained load, and leave headroom — a board that fits exactly today is a board that’s too small next year.
Unit 2 is the shopping unit: board, storage, power, cooling. This first lesson gives you the durable skill — reading any board’s spec sheet against your actual workload — so the recommendations in the board guide make sense instead of being magic.
What job are you buying for?
Write the job down before opening a shop page. For this module’s destination build: continuously decode one or more trunked-radio channels from a USB SDR, record audio to disk, and serve a web console — 24/7. That one sentence already implies: a CPU that can run DSP in real time, storage that survives constant writing, sustained USB throughput, and a reliable network path. A different job — a network-wide ad blocker, a weather display — implies a much smaller board. The spec sheet only means something relative to the job.
How much CPU is enough?
SBC spec sheets lead with cores and clock speed, but the useful reading is generational: each Pi generation’s cores are substantially faster per core than the last, and for real-time DSP per-core speed matters as much as core count. Practical guidance:
- Real-time decoding is a sustained load, not a burst. The CPU must keep up with the radio every second; there is no “it’ll finish eventually.” Falling behind means dropped samples and broken audio.
- More channels, more CPU. Each simultaneously monitored channel adds decode work — Tuning for small CPUs turns this into a budget you can measure.
- Prefer the current generation. The jump in per-core speed between SBC generations is usually far larger than between PC generations.
How much RAM — and why is it forever?
RAM on an SBC is soldered — the number you buy is the number you keep. Rough tiers:
| RAM | What it comfortably runs |
|---|---|
| 1 GB | Bare headless Linux plus one light service |
| 2 GB | A modest decoder or network service; little headroom |
| 4 GB | The comfortable middle: GopherTrunk, recordings, OS caches, room to grow |
| 8 GB+ | Several serious workloads, big databases, containers |
Unused RAM isn’t wasted — Linux uses spare memory as disk cache, which on slow SD storage is a genuine performance feature. Buy for next year’s project, not today’s minimum.
Why is USB the sleeper spec?
Here’s the one that catches SDR builders. An RTL-SDR streams samples continuously — megabytes every second, without pause, for as long as the scanner runs. Two spec details decide whether that goes smoothly:
- USB generation. USB 2.0’s real-world sustained throughput (~30 MB/s shared) is enough for one RTL-SDR, but tight once other traffic joins. USB 3 ports give an order of magnitude of headroom.
- Controller topology. On some boards (famously, older Pis) all USB ports — sometimes the Ethernet jack too — share one internal controller, so a storage stick’s burst of writes steals bandwidth from the SDR mid-stream. Newer boards give USB 3 its own lanes. A spec sheet that says “4 × USB” without saying how they’re wired is hiding the number that matters.
Insufficient USB bandwidth shows up later as mysterious dropped samples — USB SDR gotchas is devoted to those failure modes.
What about networking, storage interfaces, and the rest?
- Ethernet. For an always-on appliance, wired gigabit Ethernet wins: no dropouts, no Wi-Fi credentials to break, and — bonus for radio work — no 2.4 GHz transmitter sitting next to your SDR. Wi-Fi is the fallback, not the plan (Networking your board).
- Storage interface. All boards take SD cards; better ones can boot from a USB SSD or (best) an NVMe drive — a major reliability upgrade the next lesson weighs.
- The GPIO header matters if you’ll attach electronics (Unit 4); all mainstream boards have one.
- Form factor and power feed the next lessons — a faster board eats more watts and needs more cooling.
Rule of thumb: size the board so your workload uses half its CPU. The other half is headroom for busy hours, OS updates, and the feature you’ll add next month.
Quick check: which under-advertised spec matters most when a board will host a streaming USB SDR?
Recap
- Start from the job, written down — a spec only means something against a workload.
- CPU: real-time decoding is a sustained load; prefer current-generation cores and budget per channel.
- RAM is soldered — 4 GB is the comfortable middle for a GopherTrunk appliance; spare RAM becomes useful disk cache.
- USB is the sleeper spec: an SDR streams continuously, and shared controllers can starve it — check topology, not just port count.
- Prefer wired gigabit Ethernet for an appliance, and size for 50% CPU headroom.
Next up: Storage & SD cards.
Frequently asked questions
How much RAM does a single-board computer need?
It depends entirely on the workload. A bare headless Linux system idles in a few hundred megabytes; a network service or decoder like GopherTrunk is comfortable in 2 GB; 4 GB gives real headroom for recordings, a database, and OS caches; 8 GB is for running several serious workloads at once. RAM is soldered on SBCs, so buy the size you’ll want in a year, not the minimum that boots today.
What spec matters most for an SDR project?
After a mid-range or better CPU, the sleeper spec is USB. A software-defined radio streams samples continuously, so the board needs enough real, sustained USB throughput — and on some boards several ports (or the Ethernet) share one internal USB controller, so the SDR competes with other traffic. A board with USB 3 ports on their own controller, plus 4 GB of RAM, is a comfortable GopherTrunk host.
Are cheap Raspberry Pi alternatives worth it?
Sometimes — the better ones offer more RAM or CPU per dollar. The price you pay is software: shorter-lived OS images, thinner documentation, and a smaller community when something breaks. For a first board or an appliance you want to trust for months, the mainstream choice usually costs less in total time.