Lesson 10 of 31 beginner 5 min read

When you need more than a Pi

Key takeaways Upgrade on evidence, not vibes: the real signs are sustained CPU saturation (load average pinned above core count), RAM pressure (heavy swapping), and I/O bottlenecks (USB or storage can’t keep up) — measured, not guessed. The next rung is a mini PC: 3–10× the compute, real SSD storage, ample USB, still small and quiet, at 2–4× the price and power draw. Everything this module teaches — headless Linux, systemd, monitoring — transfers unchanged, because a mini PC is just a bigger appliance. Wideband multi-system decoding is the classic GopherTrunk reason to climb.

Unit 2 closes with the honest question every SBC fan eventually faces: is this board still the right tool? The answer should come from measurements you already know how to take — and the comforting news is that “moving up” abandons none of your skills.

What are the real signs you’ve outgrown the board?

Three measurable symptoms, in the order they usually appear:

Sustained CPU saturation. The load average (from uptime or htop) sits at or above the number of CPU cores for long stretches — the machine has more runnable work than cores. For a real-time decoder this shows up as the daemon logging that decoding can’t keep up, dropped samples, or broken audio during busy periods. Occasional peaks are fine; pinned is the signal.
RAM pressure. free -h shows almost no available memory and swap in constant use; the system feels sticky and the storage (busy swapping) wears faster.
I/O bottlenecks. The workload needs more USB streaming bandwidth than the board has (several wideband SDRs at once), or storage throughput the SD interface can’t deliver.

$ uptime
 21:14:03 up 12 days,  4:11,  1 user,  load average: 4.12, 4.05, 3.98
# on a 4-core board: saturated, no headroom

Before shopping, spend an hour on the cheaper fixes: Unit 6’s Tuning for small CPUs — sensible sample rates, fewer simultaneous channels, no wasted services — often buys back the margin. Upgrade when tuned load still saturates.

What does the next rung actually look like?

The step up from an SBC is usually a mini PC — a small, quiet x86 box (the NUC-style form factor, and the many cheap efficient-CPU models around it):

	Full-size SBC	Mini PC	Small server / desktop
CPU	4 efficient ARM cores	4–8 faster x86 cores	Many fast cores
RAM	Soldered, 1–8 GB	Socketed, 8–64 GB	32 GB+
Storage	SD card / one SSD	Real NVMe SSD(s)	Many drives, RAID
Power draw	3–15 W	10–40 W	40 W+
Noise	Silent (passive)	Near-silent	Audible
Price	$35–100	$150–400	$400+
GPIO header	Yes	No	No

Note what you give up: the GPIO header (Unit 4’s direct electronics — a mini PC talks to hardware only over USB) and the last word in power draw. Note also what you don’t give up: every skill. A mini PC runs the same Debian-family Linux, the same systemd services, the same SSH workflow, the same monitoring. It is a bigger appliance, administered identically — and on x86 you’d simply grab GopherTrunk’s x86-64 build instead of the ARM one.

What’s the classic GopherTrunk reason to climb?

One trunked system at modest sample rates is comfortable SBC territory. The climb usually starts when ambition compounds: multiple systems at once, a wideband capture that lets one SDR cover many channels simultaneously (more samples per second in means proportionally more DSP), always-on recording of everything, plus a growing database and busier web console. Each is CPU and I/O; together they cross the line. The deployment module picks up the story of running GopherTrunk on bigger hosts.

Could you use several small boards instead?

Sometimes the right answer is sideways, not up: two SBCs, each dedicated to one radio system, can beat one big box — they fail independently, sit close to different antennas, and stay cheap. The cost is administering two machines (Unit 5’s monitoring and backups, twice). This “one appliance, one job” pattern is very much in the embedded spirit; choose it when the workloads are naturally separate, and a bigger single box when they share data.

Rule of thumb: measure first, tune second, upgrade third. A pinned load average after honest tuning is the green light — and buy the next size up from what today’s numbers need.

Quick check: what most clearly says a project has outgrown its SBC?

Recap

Upgrade on measured evidence: pinned load average, real RAM pressure, or I/O bandwidth the board physically lacks.
Tune before you buy — sample rates and channel counts often buy the margin back.
The next rung is a mini PC: several times the compute, real SSDs, same Linux, same skills — a bigger appliance, minus the GPIO header.
The classic climb trigger for GopherTrunk is wideband, multi-system, record-everything ambition.
Consider several small boards (“one appliance, one job”) when workloads are naturally separate.

Next up: Flashing an OS image.