Testing a power supply without guessing

Why this matters

Lesson 36 established that a failing supply produces symptoms that look like memory, storage, graphics and operating system faults, and that it produces them intermittently. Lesson 42 provided the instrument. This lesson is where the two meet, and it converts "I think the power supply is the problem" into a reading with a tolerance beside it.

That conversion matters commercially as well as technically. A supply replaced on suspicion is a guess billed to a customer; a supply replaced because the 12 V rail measured below specification under load is a repair with evidence.

The safety rule from lesson 2 governs everything here and is worth stating once more before the detail: the supply is not opened. Every measurement in this lesson is taken at the output connectors, from outside the unit.

The lesson

What a supply is meant to deliver on each rail, and the tolerance it is allowed

The ATX specification permits ±5 per cent on the main rails. That gives:

The power good signal is the other one to know: the supply asserts it once the rails are stable, and the board will not start until it does. A supply producing correct voltages but not asserting power good gives a machine that does nothing — which looks identical to a dead supply from the outside and is the reason a tester that checks this signal is more useful than a meter alone.

Wire colours are standardised and worth memorising, because they are how you find the rail at a connector: yellow is +12 V, red is +5 V, orange is +3.3 V, black is ground, purple is +5 V standby, green is power on, grey is power good.

Two habits that make readings meaningful. Take the reading with the black probe on a ground (black) wire at the same connector, because voltage is a difference and a distant ground introduces error. And note whether the machine is idle or loaded, because a supply's behaviour under those two conditions is the whole point of the next two sections.

Power supply testers against a multimeter, and what each one misses

A power supply tester is a small device the connectors plug into. It applies a token load, starts the supply, and displays each rail plus the power good signal.

• Fast, cheap, and it checks power good, which a meter does not.
• Requires no probing, so it is safer and easier.
• Its load is minimal. A supply that fails only under real load will pass, and that is the most common failure mode of an ageing supply. This is the limitation to know.

A multimeter measures one point at a time in a running machine.

• Measures under real load, with the actual components attached.
• Can be used while the fault is occurring.
• Does not check power good directly, though the machine starting is indirect evidence of it.
• Requires care: back-probing a live connector means touching the right contact without shorting to the one beside it.

The two are complementary, and the examinable point is precisely that: a tester answers "does this supply produce the rails at all", a meter answers "does this supply hold the rails under the load this machine places on it". A supply that passes a tester and fails in the machine is not a contradiction; it is the two tools answering their own questions.

Neither tool measures ripple — the alternating component riding on a DC rail — which is a real failure mode of ageing supplies and requires an oscilloscope. Knowing that this exists, and that a supply can be within voltage tolerance and still be unfit, is the honest limit of what these tools prove.

Testing under load, since a supply that passes idle can still fail in use

The method for measuring in a running machine:

1. Machine assembled and running, case open, everything connected.
2. Meter on DC voltage, range covering 20 V.
3. Black probe on a ground at the 24-pin connector — a black wire's contact from the back of the connector.
4. Red probe on the rail under test, back-probed alongside the wire.
5. Record the reading at idle.
6. Apply load — a processor stress test, a graphics stress test, or both — and record again after the machine has been loaded for several minutes.
7. Compare both against the table above.

What the results mean:

• Within tolerance at idle and under load. The supply is delivering correctly. Look elsewhere.
• Within tolerance at idle, out of tolerance under load. This is the classic ageing supply. Capacitors degrade, the rail sags when current is drawn, and the machine becomes unstable exactly when it is working. Replace it.
• Out of tolerance at idle. Clearly faulty.
• Fluctuating markedly. Also faulty, and it is the pattern that matches an intermittent from lesson 36.

Software monitoring that reports rail voltages reads the board's own sensors. Those are convenient, they log over time, and they are only as accurate as the board's calibration — useful as a trend while the fault is occurring, and not a substitute for a meter when a decision depends on it.

The paperclip test: what it proves, what it does not, and the risks it carries

The jump-start test: with the supply disconnected from the board, bridge the green wire (power on) to any black wire (ground) at the 24-pin connector. That is what the board does when the power button is pressed. The supply should start and its fan should spin.

What it proves: the supply will start. That is all.

What it does not prove:

• That the rails are at correct voltages — the fan spinning tells you a rail is producing something, not that it is producing 12.0 V.
• That the supply holds up under load, since there is almost none.
• That power good is asserted correctly.
• That the supply is not the cause of an intermittent fault.

The risks, which is why this is the last resort rather than the first test:

• Starting a supply with no load at all can damage some designs, so leaving a drive connected as a token load is the safer version.
• A slip with the wire or paperclip shorts a rail, which can damage the supply.
• It is easy to bridge the wrong pins if the connector is not read carefully, and pin numbering is not obvious.

A power supply tester does the same job safely and adds the power good check for a modest cost, which is why lesson 45 gives it a place in the kit. The paperclip test is what you do when you have neither a tester nor a spare supply, and it answers one narrow question.

Deciding replacement from measurement rather than from suspicion

The decision, stated as evidence:

Replace when:

• A rail is outside tolerance at idle, or under load.
• A rail fluctuates significantly under steady conditions.
• The supply does not start, and the board and front-panel wiring have been eliminated.
• Power good is not asserted, on a tester that checks it.
• The supply is physically damaged, smells burnt, or has visibly swollen capacitors at the output side.
• Substituting a known-good supply resolves the fault — which remains the most definitive test of all.

Do not replace on:

• A guess, because several components appear faulty at once. That is a strong reason to test, from lesson 28, and not a result.
• Age alone. Supplies do degrade, and a measurement is still cheaper than a part.
• A fan that has become noisy, which is a fan fault and may be all that is wrong.

When replacing, size it by lesson 28's method rather than matching the old one's label, because the machine's requirements may have changed since — and because the old supply's wattage is what somebody else chose, possibly wrongly, possibly for a different set of components.

And record the reading that justified it. "Replaced supply: 12 V rail measured 10.9 V under load, specification minimum 11.4 V" is a repair a customer can see the reason for, and a colleague can verify after the change by taking the same measurement again.