DIY Power Supply for a 10-Inch Rack Mini PC Cluster

Why a Standard ATX Power Supply Doesn't Fit This Build

A single ATX power supply is designed to feed one motherboard, one GPU, and a handful of drives from a 12V, 5V, and 3.3V rail set defined by the ATX12V specification (Wikipedia, ATX). That form factor — typically 150mm wide and at least 86mm tall — is hard to fit alongside multiple mini PCs in a 10-inch (250mm) rack enclosure, where usable depth and height are already spoken for by the compute nodes, switch, and cabling.

Most mini PCs also don't use ATX power at all. They ship with an external DC brick — commonly 19V or 12V — and draw a fraction of what a full ATX supply is built to deliver. Feeding four, eight, or twelve of them from individual wall bricks works, but it turns a compact rack into a nest of adapters and wastes the internal volume an ATX unit would otherwise occupy. The practical fix most homelab builders converge on is a single higher-wattage DC supply — often a DIN-rail or open-frame 12V unit from an industrial supplier such as Mean Well — paired with a simple distribution board that splits that one rail out to each node's input voltage.

Sizing Power for a Mini PC Cluster

Before buying or building anything, total up real draw per node rather than guessing:

1. Idle draw. Efficiency-core mini PCs built on parts like Intel's N-series (Alder Lake-N) processors are specified with TDPs as low as 6W, and idle system draw for that class of hardware commonly lands in the single digits to low teens once RAM, storage, and networking are included (Wikipedia, Alder Lake). Mobile Ryzen or Core-H mini PCs with a higher-TDP chip and dGPU will idle noticeably higher.
2. Peak/sustained draw. Look at the processor's boost TDP figure, not just the base TDP — that's the number a distribution board and wiring need to be sized against, not the idle number.
3. Headroom. Size the supply and wiring above the sum of sustained peaks, not idle totals, so a synchronized boot or benchmark run across every node doesn't brown out the rail.
4. Expansion. If the rack has open bays, size the supply and busbar for the bays you'll eventually fill, not just what's installed today — DC distribution boards are far easier to build once than to redesign around a mid-life expansion.

Treat the table as a starting worksheet, not a spec — always check the vendor's own power figures for the exact model going in the rack; the BOSGAME VTA-439 Ryzen AI 9 HX 470 and other Ryzen AI mini PCs pull meaningfully more than an N100 box under load, and that gap compounds fast once you're running a dozen nodes.

AC-Side vs. DC-Side Distribution

There are two broad ways to solve power for a rack full of mini PCs, and the right one depends on how much of the build should be electrical work versus off-the-shelf parts.

AC-side distribution keeps each mini PC's stock external power brick, and the build is really just a tidy way to feed AC mains to a rack shelf of bricks — a rack-mount or surface-mount power strip with individual, switched, surge-protected outlets. A compact 5-outlet strip with USB charging, like the QINLIANF wall charger/surge protector, illustrates the category at small scale: individually accessible outlets so a single node can be power-cycled without pulling the plug on the whole rack. At rack scale this becomes a rack-mount PDU, but the principle — switched, surge-protected AC drops per node or small group of nodes — is the same.

DC-side distribution replaces every individual brick with one higher-wattage 12V (or matching) supply and a custom busbar or PCB that splits that single DC rail out to each node's barrel or USB-C PD input. This is the more "DIY power supply" approach implied by the topic, and it's where most of the real engineering work sits:

• Supply. An open-frame or DIN-rail 12V industrial supply, sized to the sustained peak total from the worksheet above, is the standard building block homelab writeups reach for over a repurposed ATX unit — brands like Mean Well are the default reference point for this category.
• Distribution board. A simple busbar or a purpose-cut PCB with screw terminals splits the 12V rail to each node. Keep runs short and use stranded copper sized for the sustained current per branch, not just the idle current — undersized conductors are a common failure mode in DIY DC distribution, and conductor sizing tables (see American Wire Gauge on Wikipedia) exist precisely so this isn't a guess.
• Per-node conversion. Most mini PCs expect 19V or 12V from their stock brick, not raw 12V off a bus — check the label on the OEM adapter before assuming a straight 12V feed will work, and budget for a small buck/boost converter per node where the input voltage doesn't match.
• Switching. Individually switched hubs are a cheap way to prototype per-node power control before committing to a custom board — a SABRENT hub with individually switched ports (or the budget USB 2.0 version) is USB-power scale rather than a full mini PC's draw, but the same individually-switched-port concept is exactly what a DC distribution board should replicate at higher current: the ability to kill power to one node without touching the rest of the rack.

A third option worth mentioning for anyone building a cluster from scratch rather than retrofitting existing mini PCs: Power over Ethernet. IEEE 802.3bt (PoE++) delivers up to roughly 90W at the source and around 71W guaranteed at the powered device (Wikipedia, Power over Ethernet), which covers a low-power node's needs over the same cable that carries its network link — eliminating the DC distribution board entirely at the cost of needing PoE-capable switch ports and PoE splitters per node. It's a much better fit for something like the Raspberry Pi LLM cluster builds than for higher-draw x86 mini PCs, but it's worth ruling in or out before committing to a custom 12V board.

Measuring What You Actually Have

Before finalizing either approach, measure real draw rather than trusting spec-sheet TDP alone — TDP is a thermal design figure, not a live power draw number, and the two routinely diverge under real workloads. A USB in-line power meter, connected through an adapter like a USB-C converter where the node's power path runs through USB-C PD, is a low-cost way to log actual idle and sustained-load numbers per node before sizing the supply and wiring. This step is what separates a distribution board sized against a datasheet from one sized against reality — and it's the difference between headroom and a rack that browns out the moment every node boots at once.

Practicality: DIY vs. Buying a Rack PDU or Individual Bricks

There isn't a universal "DIY is always cheaper" answer here — it depends on node count, whether the builder already owns a soldering iron and a spare 12V supply, and local component pricing, which shifts too often to quote a fixed dollar figure. As a rough framework:

• Small clusters (2-4 nodes): Individual OEM bricks plus a switched AC power strip is usually the fastest and most reliable path — not enough nodes to justify designing a custom board.
• Mid-size clusters (5-10 nodes): This is where a single DC supply plus a simple busbar starts to pay off in cable management and rack space, even if the raw component cost is close to a stack of bricks.
• Larger clusters (10+ nodes): A custom distribution board becomes close to mandatory for cable management alone — a rack with a dozen individual wall warts is a heat and airflow problem before it's anything else.

Whatever the count, price the actual parts going into the specific build — supply wattage, connector count, wire gauge and length — rather than assuming a fixed DIY-vs-commercial spread; component and shipping costs move quickly enough that any number quoted here would be stale within a quarter.

Safety Fundamentals for Dense Low-Voltage Wiring

Low voltage doesn't mean low risk once current climbs into double digits across a shared rail. A few non-negotiables:

• Fuse or breaker every branch, not just the main input — a shorted node shouldn't be able to take down the whole rack, and a branch fuse rated to the branch's expected current (not the supply's total output) is what makes that true.
• Size conductors to the branch's sustained current, using a proper AWG ampacity table rather than whatever wire happens to be on hand — undersized wire under sustained load is a heat source, not just an efficiency loss.
• Insulate every splice. Heat-shrink tubing (or proper crimp connectors) over any 12V split is cheap insurance against an exposed conductor shorting against a metal rack rail.
• Don't daisy-chain grounds or opportunistically reuse a mini PC's own chassis as a return path — keep the DC return wired deliberately back to the supply.
• Check the actual input spec on each node's OEM adapter before wiring it to a shared rail — feeding 12V into a device that expects 19V (or vice versa) is a fast way to damage hardware that a distribution board was supposed to protect.

None of this replaces checking local electrical code for anything touching mains AC, or a manufacturer's datasheet for a specific supply or breaker's actual ratings — treat the above as a checklist of categories to verify, not a substitute for the datasheet.

Related Reading

For context on the mini PCs likely to end up in a build like this, see how mini PCs are actually manufactured, how an N100 box compares to a Raspberry Pi for a beginner homelab or for Home Assistant specifically, and what local LLM hardware actually needs — from Gemini's on-device requirements to a head-to-head on cloud vs. local inference tiers and, at the high end, what an RTX 5090 changes for local AI workloads.

FAQ

How much power does a typical mini PC cluster node draw?

It varies by chip and workload, but N-series efficiency-core mini PCs commonly idle in the single-digit-to-teens watt range, while mobile Ryzen/Core-H mini PCs with a discrete GPU can idle higher and peak well above that under sustained load. Measure the specific model with an in-line power meter rather than relying on TDP alone.

Can I power a mini PC cluster over PoE instead of building a DC distribution board?

For lower-power nodes, yes. IEEE 802.3bt (PoE++) delivers up to roughly 71W guaranteed at the device, enough for many efficiency-core boards, and it removes the need for a separate DC busbar since power rides the network cable. It requires PoE-capable switch ports and a splitter per node, and it's a poor fit for higher-draw mini PCs with discrete GPUs.

What wire gauge should I use for 12V DC distribution in a rack?

Size the conductor to the sustained current on that specific branch using a standard AWG ampacity table, not a fixed gauge picked in advance. A branch feeding one low-power node needs far less copper than one feeding several higher-draw nodes off the same run.

Is a DIY power distribution board actually cheaper than buying individual power bricks?

It depends heavily on node count and current component pricing. Small clusters usually come out ahead sticking with individual OEM bricks and a switched power strip; the case for a custom DC board strengthens as node count and cable-management pain both grow.

Do I need circuit breakers or fuses in a low-voltage DC rack build?

Yes. Fuse or breaker every branch off the main supply, sized to that branch's expected current, so a fault on one node can't take down the whole rack. This is separate from any AC-side breaker protecting the supply's mains input.

What's the difference between AC-side and DC-side power distribution for a mini PC cluster?

AC-side distribution keeps each node's stock power brick and just organizes how mains power reaches those bricks (a switched, surge-protected power strip or rack PDU). DC-side distribution replaces the bricks entirely with one higher-wattage supply and a custom board that splits a single DC rail out to every node.

Citations and sources

• ATX — Wikipedia
• Alder Lake — Wikipedia
• American wire gauge — Wikipedia
• Power over Ethernet — Wikipedia
• MEAN WELL — DC power supply manufacturer

This piece is editorial synthesis based on publicly available information. No independent first-party benchmarking is reported.