My computer won't turn on

Before any fix, classify. “Won't turn on” is three different problems wearing one phrase: no response at all (nothing happens when the button is pressed), response but no usable screen (fans and lights, black display), and it starts but the operating system won't load (an image appears, then an error or a loop). The fix trees for these diverge immediately — the first is an electrical chain, the second is a start-up self-test or display path, the third is the operating system — so the very first step is an observation, not an action: press the button, and note exactly what happens and what doesn't. A machine with no response at all, a machine with spinning fans, and a machine showing a boot logo are three different diagnoses, and mixing up which one you have is how a ten-minute fix becomes an hour.

The second skill is knowing how a machine tells you it has power. Power doesn't announce itself with one sign; it shows up as a set of small observations — a standby LED glowing before the button press, a charging light on a laptop, a fan twitch when the button goes down, a relay click, a display waking. Each observation answers a specific question: the standby LED says “power is reaching the board,” the fan twitch says “the board is commanding the power supply to start,” the click says “something in the power path is attempting and stopping.” The method on this page is built on that vocabulary: at every stage, ask what the machine is telling you, and note which observations are present and which are absent. Absent signs are as useful as present ones — “no standby LED” localizes the fault to the power path before the board, full stop.

And the starting point, which doubles as the safety frame for everything else: the checks are ordered from outside in, because the power chain starts at the wall. Outlet first, then cable, then any surge protection, then the machine's own switch or adapter, and only then the machine itself. This ordering isn't just safe — it's statistically where the fault actually is; a surprising share of “my computer is dead” calls end at a lamp plugged into the same outlet. As you move inward, the two rules from the top of the page stay in force: unplug before opening anything, and never open a power supply. If at any point there's a burning smell, smoke, or a history of liquid, skip the rest of this section and read when a computer requires professional repair — that's the stop line, and it comes before any hardware step.

The state this section covers is specific: no LEDs, no fan movement, no sounds, no display — the machine gives no response to the power button. What that state means is narrower than “it's dead”: the machine is not receiving usable power, or the power path inside it isn't completing. The chain between the wall and the board has more links than most people think — the outlet itself, the surge protector or power bar (and its breaker), the power cable, the desktop power supply's rear switch, or the laptop's adapter and jack — and each link fails on its own, independently of the others. That's why the procedure works from the wall inward: each check removes one link from suspicion, and the checks are all cheap, safe, and reversible.

The sequence, with what each step is actually determining. The outlet: plug a known-good device — a lamp beats guessing, a phone charger works too — into the exact same socket. If it doesn't work there either, the problem is upstream of your machine (the outlet or the circuit), and the rest of this section is skipped. Try a different outlet before doing anything else, because a dead wall socket is a real and common cause. The surge protector or power bar: check its main switch, and look for a small reset button — many bars trip silently after a spike, and pressing the reset is the entire fix. The definitive test is bypassing the bar entirely and plugging straight into the wall. The cable: try a known-good power cable of the right connector; on laptops, look at the adapter's cable for fraying near the connector, which is where they fail. The power supply switch: many desktop PSUs have an I/O switch at the rear that can be switched off — and often is, after a move — with the machine otherwise perfectly healthy. It must be at the on position even though the front button is what starts the machine. Laptop adapters: check for a damage state on the brick (excessive heat, a smell, visible damage) and try a known-good charger of compatible wattage — an underpowered charger can produce exactly the “no signs of life” state on a discharged laptop.

What this tells you, and where to go from here. If power returns after one of these steps, you're done — and the step that fixed it is the note to keep (a bar that keeps tripping is telling you something about the circuit). If the full chain is proven good and the machine still gives nothing, the fault is inside the machine's power path: for a desktop, that's the power supply or the board (the power supply section covers what's safe to check and what isn't, and the motherboard section covers the standby-LED test that separates the two); for a laptop, the laptop section has the adapter and battery isolation, including the standard hard reset. One state changes everything, though: if there's any burning smell, discoloration, or a history of liquid at any point in this procedure, stop and go to the professional-repair section — the chain checks assume a machine that is merely unpowered, not one that is damaged.

Laptop power is a different chain from a desktop's, and it has two failure clusters that get reported as one: it won't charge (no indicator, or the battery level doesn't climb) and it won't power on (dead even when plugged in). The chain is charger (brick plus cable) → jack (or USB-C) → battery → board. Two honest caveats before the steps: not every laptop supports USB-C charging — many require the manufacturer's barrel connector, or USB-C at a specific power profile, so a cable that charges a phone is not evidence it will charge the laptop; and not every laptop has a removable battery — the battery steps below apply where removal is possible, which on modern thin laptops usually isn't. Where the steps say “if removable,” they mean it literally.

The sequence. The charging indicator: when you plug in, what changes? A light near the jack or on the chassis turning on (or changing color) means the charging path is alive; no change means the path is suspect. Some laptops only show charging state inside the operating system, so on a dead laptop the physical indicator — or its absence — is the whole evidence. Give it time: a deeply discharged battery can take fifteen to thirty minutes of charging before the machine shows any sign of life. Waiting and watching is a real diagnostic step here, not a waste of time — “nothing happened for two minutes” is not the same finding as “nothing after half an hour.” The charger side: a known outlet, then a known-good compatible charger. Wattage matters more than people expect — an underpowered charger may charge a resting laptop slowly or not at all, and may not start one that's under load. Check the brick for heat, smell, and visible damage. The battery, where removable: the isolation test is powering the laptop with the battery out (adapter only) versus in (battery only). Boots on adapter only → the battery or its circuit is the suspect. Boots on battery only → the charger or jack path is the suspect. A swollen battery — a bulging keyboard, a lid that no longer closes flat, a base that rocks on a level desk — is a stop: don't use it, and handle the replacement as the lithium-safety matter it is (professional section). The hard reset: the standard laptop procedure — unplug, remove the battery where possible, hold the power button for fifteen to thirty seconds, reconnect, and try. What it does is drain the residual charge and clear whatever stuck power state the board is in; it's safe, it's reversible, and it's the first thing worth doing on a laptop that looks bricked.

What this tells you. The isolation results map cleanly: adapter-only boot points at the battery or charging circuit, battery-only boot points at the charger or jack, and neither — after a known-good charger, the hard reset, and a real wait — points at the board's power path, which is professional territory (and, on many laptops, warranty territory, where opening it yourself changes the terms). One more distinction: “won't charge” with the machine otherwise working is a slower problem — the laptop will run on battery until it's gone — while “won't power on at all” is the urgent version of the same chain. If liquid is anywhere in the history, both versions skip to the professional section without the intermediate steps.

A power outage is a useful piece of history, but it's a cause in fewer cases than people assume — which is exactly why this section starts with the boring checks. An outage can do three things to your setup: nothing at all (the common case; the machine simply needs power again, and so does everything that shares the circuit), trip a protection (a surge protector's breaker, a circuit breaker at the panel, or even the power supply's rear switch being left off while you checked things), or, less often, cause actual damage from a surge on the way back. There's a fourth possibility that isn't damage at all: the machine powered down uncleanly, so even once it starts, Windows may want to run its first-boot diagnostics before anything looks normal. Each of these has a different next step, and they're checked in order of likelihood.

The sequence. The house level first: is the outlet alive (the lamp test from the no-power section)? Check the circuit panel for a tripped breaker — resetting a breaker is an ordinary homeowner task; a breaker that trips again the moment you reset it is an electrician problem, not a computer problem, and the computer should stay unplugged while that's sorted. The surge protector: many bars have a small reset button that trips after a spike, silently, with the main switch still on — press it, or bypass the bar and plug straight into the wall. The machine's switch: the PSU's rear switch, still at on. Try to start. If it starts and Windows misbehaves — a “diagnosing” screen, a restore prompt, missing desktop state — that's the unclean-shutdown consequence, and the Windows boot section covers letting the automatic repair do its job without interfering. If it's still dead: don't assume the surge caused it. Run the full no-power chain first; outages and failures co-ocall constantly, and the “surge fried it” conclusion should be earned, not assumed. The one pattern that does earn it early: multiple devices dead on the same outlet or circuit at the same moment. That points at the circuit or the protector, and the appropriate expertise shifts to the electrical side.

What this tells you, and the stop line. One machine dead after an outage, everything else on the circuit fine, no smell, all safe checks clean: the machine took a hit internally, and the honest next step is the power supply checks (safe observations only) followed by professional assessment — a surge-damaged supply can fail in ways that make repeated power-on attempts counterproductive. Multiple machines dead on one circuit: stop at the circuit, keep everything unplugged, and let the electrical side get sorted before the computer side resumes. And the preventive note, because outages repeat: this is the case where a surge-protected outlet or an uninterruptible power supply earns its keep — a UPS carries a machine through the outage cleanly and is the one purchase this whole page quietly recommends.

The distinction that saves the most time on this page: “turns on” — fans, lights, a warm machine — means the power path and the core of the board are working. A black screen on top of that is a display path problem, and the display path has four links, each with its own failure: the graphics output (the GPU, or the board's integrated graphics), the cable, the monitor's input, and the monitor itself (including, on laptops, the panel and its backlight). That's why this section's first question is about the monitor, not the computer: a significant share of “my PC is dead” reports end at a monitor that's on the wrong input or not powered at all, and the computer in those cases was booting fine into a room where nobody could see it. The external-monitor test at the end of the section is what converts “black screen” into a named component.

The sequence, cheapest first. Monitor power: is the monitor's own power light on? Monitors have their own power cable, and an off monitor looks exactly like a dead PC from the chair. Input selection: auto-input detection fails often enough to be worth doing by hand — press the monitor's input/source button and select the input your cable is actually in (HDMI 1 versus HDMI 2 versus DisplayPort are different inputs, and “the right one” is a physical fact, not a setting). The cable and its ends: reseat both ends, then try a different cable if you have one — a failed cable is a real cause, not a theory. Which port the cable is in: the one that surprises people. Many motherboards have video ports on the board itself and a discrete graphics card with its own ports, and the board's ports usually only work with integrated graphics. If a discrete GPU is installed, the cable belongs in the GPU's ports — a cable in the board's port with a GPU fitted is a classic no-image state. Laptop brightness: the function-key brightness row can sit at zero, which is a black screen that's actually a working screen; try the brightness keys before anything else on a laptop. The external monitor test: connect a known-good external display. If the external shows an image, the laptop's problem is its own panel, panel cable, or backlight — a repair-side matter. If nothing appears on any display, the video source is the suspect, and the fans-spin section takes over, because a board that powers on but produces no video anywhere is usually failing its self-test (memory, GPU, board) rather than its display.

What this tells you, and the honest limits. The external-monitor result is the branch point: image on external, black on the built-in panel → the panel side (laptop repair); nothing anywhere → the source side (the self-test suspects). It's worth saying plainly what a black screen does not prove: it's not proof of a dead graphics card — the display path above is longer and cheaper than the card — and a powered monitor with a black screen is not proof the monitor is fine, because a monitor can be powered and still be on the wrong input or in its own error state (some show a “no signal” message, others just stay black). If the machine also has beeps or debug lights during all of this, stop the display work and read them first — the diagnostic indicators section is the faster road when the board is talking. And if your monitor's side of things gets involved (the monitor itself misbehaving across computers), the monitor-no-signal guide covers that problem from the display's perspective.

What spinning fans actually mean, stated precisely: the power supply is delivering power, and the motherboard is at least partially alive — the machine has passed the power stage and is failing a later one. There are two later stages to separate. The first is the display path (the machine may be booting perfectly into a broken or misrouted display — the black-screen section owns that, and it comes first here because it's cheaper than any internal work). The second is the self-test: during power-on, the board runs checks on its core components — memory, graphics, processor — and a check that fails halts the boot, on many modern boards silently: fans on, no image, no beep. The classic suspects for that silent halt are RAM (mis-seated or failed module), the graphics card (mis-seated, failed, or in the wrong configuration), and — last and least common for a first-time failure — the board itself. That ordering is the ordering of the procedure below.

The safe sequence, in cost order. Prove the display path first — monitor power, input, cable, correct port, external display test (all of the black-screen section). Skipping this is the most expensive mistake in no-display troubleshooting: an hour of RAM work to discover the cable was in the board's port. Listen and look for the board's report — beeps, a speaker, debug LEDs, a two-digit code: if the board is trying to tell you what failed, read it before reseating anything (the diagnostic section). The RAM reseat — the most common fixable cause of a silent post-halt on an otherwise-powered desktop: power off, unplug, wait a moment, reseat the modules, one at a time, in the slots your board's manual recommends, testing single modules (the full procedure and the slot logic are in the RAM section). The graphics card — reseat it firmly until its latch holds, check its power connectors are fully seated, try a different cable and port; if the board has working integrated graphics, removing the card entirely and using the board's port is both a fix and a test. Clear the CMOS — removing the motherboard's coin cell battery (or using the board's documented clear jumper) for a few minutes resets the BIOS to defaults; it's safe and documented on every board, it clears stuck configuration states, and it's worth one attempt — with the honest note that it changes nothing you'll miss except the clock and custom settings. Minimize — remove every peripheral and every non-essential component (extra drives, added cards, the second memory module where practical) and attempt a boot with the smallest configuration that should work. If it boots minimized, add pieces back one at a time to find the culprit; if it doesn't, the remaining suspects are board-level.

What this tells you, and the stop line. The single-module test results read cleanly: one module boots and the other doesn't → a bad module (run on the good one; replace the other). A module boots in one slot but not another → a bad slot, which is board-side. Neither module, in any slot, with the display path proven good and the board's report (if any) read → the self-test is failing below the components you can service, and that's the motherboard section's territory with professional assessment as the likely end state. Two honest boundaries: reseating a graphics card and memory is well within an ordinary user's safe reach when done unplugged; pulling a processor, probing the board, or opening the power supply is not, and this page deliberately stops at that line. If the machine beeps or shows codes at any point in this sequence, that's the good-news branch — a report means the board is alive and specific, and the decoding section is the faster road than continuing to reseat.

This behavior — the machine starts the boot sequence and then interrupts it — is worth its own section because it's a different failure shape than “nothing” or “no display”: something is actively stopping a start that had begun. The honest framing is that this symptom has more true causes than any single one, and pretending otherwise is how people waste money on parts. The main families: a protection trip (the board or supply shutting the machine down to prevent damage — thermal protection is the common one, for instance when a CPU fan isn't connected or spinning, which many boards treat as an immediate shutdown condition); a power delivery problem (a supply that can hold voltage at idle but sags the moment the system loads); and a fault in something newly present (a recently added or loosened component drawing current it shouldn't, or shorting). The goal of this section is narrowing — identifying which family — not a single diagnosis, and the first observation does more than the second step will.

Start with the observation, because the timing of the shutdown carries information. How long does it run before dying? Fans that spin for several seconds — maybe through the first moments of a logo — before shutdown points differently from a machine that dies the instant the button is released: a few seconds of activity suggests the self-test started and hit a problem (memory, a protection trip, a component under load), while an almost-immediate death suggests the power path itself. Anything during the window? A beep or a debug LED that lights before the shutdown is the board's report, and it's worth more than the shutdown itself (the diagnostic section). Then the narrowing procedure: what changed recently? New hardware is the first suspect for a machine that started dying after a session in the case — the new-hardware section is the direct path, and a loosely seated card or a bumped connector are the usual culprits. Minimize — remove peripherals and added components, test with the smallest configuration (one memory module, integrated graphics where available), because a machine that boots minimized has a component problem, not a machine problem. Cooling and airflow — check that the CPU fan is actually spinning when you press the button (watch through the case or listen for it), and that its connector is seated on the board's CPU fan header; a missing or unplugged CPU fan is a classic cause of the spin-then-die pattern, and it's a two-second check. If the machine was recently cleaned or moved, the fan connector is exactly the thing that gets disturbed (that section). Power delivery — if the minimum configuration still dies under load and the cooling is verified, the supply becomes a real suspect: the power supply section covers what's safe to observe and the swap test that settles it.

What this tells you, and the stop line — which matters most in this section. Repeated shutdowns are the symptom where “keep trying” can actively cause damage: a machine that's tripping protection repeatedly, or a supply that's failing under load, gets stressed with every attempt. The stop criteria are specific: a burning smell at any point (stop, unplug, professional — no further attempts); a power supply that's hot to the touch or making clicks; and a pattern of repeated spin-then-die that survives the minimum configuration with cooling verified, because at that point the remaining suspects (the supply, the board) are professional-diagnosis territory and each additional power-on cycle is a small cost against whatever the fault is. Note for the record: “restart to test” is fine in small numbers while you're narrowing; it stops being fine the moment the pattern looks established. The difference is whether the attempt is gathering information or just repeating a result.

When a machine that “won't turn on” is actually beeping, or showing a debug light or a two-digit code, something important has changed: the board is alive, its self-test has run, and it has identified a specific subsystem to blame. That's the good-news reading of a “dead” machine, and it's why this section sits before the component sections — a report converts an open search into a targeted one. The mechanism, briefly: during power-on, the board tests its core components, and the firmware is built to signal failures through whatever reporting hardware exists on that board — a small speaker that produces beep patterns, a set of labeled debug LEDs (commonly CPU, DRAM, VGA, BOOT on modern boards), or a tiny two-digit hexadecimal display on enthusiast-oriented boards. Many boards have more than one of these; many have only the LEDs; and a fair number ship with no speaker at all, which is why “no beeps” on a silent machine tells you less than people think — the board may simply have no beeper to beep with.

The part this section has to be honest about: the codes are manufacturer-specific, and this page is the process, not the answer key. The same beep pattern can mean different things on different makers' boards, the same debug LED can be used for different purposes on different models, and the two-digit codes differ by firmware family. So the procedure is: identify the make and exact model of the board (or of the whole machine, for laptops and branded desktops), find the manufacturer's documentation — the support page's manual section usually has a “beep codes” or “debug LED” page — and decode the pattern there, then follow that maker's guidance for the named subsystem. What you can reliably infer without the documentation, and what makes the decoding worthwhile: a DRAM-pointing code or light sends you to the RAM procedure (reseat, one module at a time — and a memory-fault code is one of the strongest justifications for that work); a VGA/display code sends you to the video path (cable and monitor first, then the card, per the black-screen section); a CPU-pointing code is the one that mostly ends in “professional” — socket-level work carries its own damage risk and isn't a first-weekend procedure; and a BOOT-stage code, or a machine that reaches a “no bootable device” message, means the self-test actually passed and the problem is the storage or operating-system layer (that section). The table below is the honest version of the quick reference — what the observation usually points at, with the documentation doing the confirming.

Two practical notes that save people from dead ends. First, record the pattern exactly — count the beeps, note the pauses, photograph the LED state or code at the moment of failure — because patterns are position-sensitive (“two long, one short” is a different code from “one short, two long”) and your memory of it will be worse than the documentation's expectation of it. Second, some boards beep only for certain failures and are silent for others, and some change their reporting behavior with BIOS settings; absence of a report is therefore one data point in the fans-spin sequence, never a conclusion. If you do find the documentation and the named subsystem is memory or video, you already know where to go on this page — that's the point of the board telling you.

The distinction that keeps this section separate from everything else on the page: “won't turn on” (no power, no display) and “turns on fine but Windows won't load” are different problems, and the second one is the gentler of the two. If the machine shows any image — a logo, an error message, a recovery screen, even a “no bootable device” line — the hardware path has already done its job: power arrived, the self-test passed, the video is working. The failure is in the boot layer: the operating system, its recovery tools, or the storage the system lives on. That matters practically because the fix tree here involves no case opening, no components, and no electrical risk at all — it's software and one cable. The classification test for this section is the same one from the basic section: does an image appear? If yes, you're here. If no, the fans-spin and no-power sections are the real destinations and this one is a detour.

The boot-layer ladder, in the order it should be run — cheapest and safest first, each step reversible. Let it try: after an unclean shutdown (a power outage, a held power button, a forced off), Windows often runs a one-time “diagnosing your PC” or “repairing startup” pass on the next start. That pass can take a while and can look like a frozen screen; letting it finish is not waiting, it's the first step. A real restart: where the machine is at all interactive, choose Restart rather than shut down and power on — a restart reinitializes more of the system than a cold power cycle does, and it clears a share of one-off boot failures. Reach the recovery environment: the standard documented path is to interrupt the normal start a few times (power off at the logo, repeat) until Windows offers the recovery options; a recovery USB or the installation media does the same. This environment is the hub for everything below, and reaching it is progress in itself — it proves the disk is readable. Within recovery: run Startup Repair first (automatic, safe, it attempts the common fixes without touching your files), then — if the failure started after an update — the options to go back to an earlier build or restore point, which are the direct undos for the update-broke-boot case. “No bootable device” is its own branch: that message means the board is healthy and the system drive isn't being found. The first check is the cable (a SATA connector or the NVMe seat — exactly the “disturbed connector” failure from the moved/cleaned section), and if reseating doesn't bring the drive back, a failing drive is the working hypothesis, at which point data comes first and the sounds section's stop rules apply. A reinstall is the last step, not a troubleshooting step: it's the right end state in some cases, but it's a data decision — back up first (recovery's file access makes that possible), and it's the one step on this ladder you can't undo.

What this tells you. The ladder's results are diagnostic as well as curative: the recovery environment loading at all means the system disk is readable and the fault is in Windows itself; “no bootable device” after a reseat means the storage or its connection is the suspect (hardware territory, with data implications); and a machine that loops between the logo and a restart with no recovery access points at a deeper boot damage where the recovery media is the next tool. The one anti-advice this section exists to give: don't jump to a reinstall from the first error screen. The automatic repair, the rollback options, and the drive-cable check clear a large share of “Windows won't load” cases without touching the installation — and a reinstall done before those steps is the difference between a fixed machine and a wiped one.

Why memory is the first component to touch in a silent no-boot: the board tests its memory during the power-on self-test, and a memory that fails that test can halt the boot with no visible sign on many modern boards — fans running, no image, and on speakerless boards, no beep. The symptom profile that points here is specific: the machine worked recently, nothing was changed (or a module was moved), and it now powers on into nothing, sometimes only some of the time. Intermittency is itself a clue — a marginal seat or a failing contact often fails “most of the time,” which is why a machine that boots occasionally is still a memory suspect and not a machine that's “getting better.” Dual-module configurations add one more version: one bad module in a two-module set can take down the whole set's test, which is why the procedure below works module-by-module rather than assuming the installed configuration is the unit of failure.

The safe procedure, in full. Power down properly: shut down if possible, press and hold the power button if the machine is stuck, then unplug the power (laptop: unplug the charger and, where the design allows, remove the battery) and wait thirty seconds to a minute. The wait isn't ceremony — it lets the board's capacitors discharge and breaks any stuck power state, and it's the step that makes the next ones safe. Ground yourself: touch bare metal on the case before touching components, and keep handling to the edges — modules are gripped by their sides, never the chips or contacts. (The static risk is real but routinely overstated; the habits above are the whole of the protection that matters.) Open and look first: are the modules seated level, with the end latches fully down? A module that's in but not clicked is the commonest finding and the whole fix. Reseat one module: unclip the latches, lift the module straight out, and press it back in until both latches click. For the first test, use one module only. The slot question, honestly: which slot a single module goes in is board-specific — slot numbering and the recommended slots differ between boards, and a single module in the wrong slot can fail to boot on some designs. The board's manual (or the silkscreen marking on the board itself) is the source of truth; where it's unavailable, the slots are usually numbered and the first or second slot is the typical recommendation — one module, one slot, and a note of which is which, is the working setup. The test matrix: module A in the recommended slot → boot? If no: module B in the same slot → boot? If no: module A in a different slot. Each permutation isolates the variable: a module that boots in slot one but not slot two implicates the slot; a module that fails in every slot while its partner boots implicates the module; the result that matters most is the one that boots, because it names both the good part and the fault.

What the results mean, and the honest exit. One module boots and the other doesn't → the bad module: run the machine on the good one (a single module is a perfectly normal configuration) and replace the other; no emergency, and the diagnosis is complete. A slot that fails with known-good modules → board-side, and the motherboard section is the next read with professional assessment as the likely resolution. Neither module, in any slot, with the display path proven good and the board's report (if any) read → memory is exonerated as the primary suspect, and the fans-spin sequence's remaining steps (the graphics card, the CMOS clear, the minimized configuration) take over. One caution for the record: memory faults are the most common component cause of a silent no-boot, which is exactly why they're checked first — but “most common” is not “the only,” and the permutation results are what convert a suspicion into a finding. A board that beeps a memory pattern specifically is confirming the same thing from the other direction (the diagnostic section).

What a failing power supply can look like — and the ordering that keeps this an honest diagnosis. The patterns: total no-power after the outlet, cable, and switch are all proven good; a machine that starts and dies under load (the starts-then-shuts-off behavior); random shutdowns while working; a click or hum coming from the supply itself; or a burning smell, which is a stop, not a symptom to note. But the ordering matters: a “no power” desktop is statistically more often the outlet, the cable, the strip, or the rear switch than the supply — the power chain outside the case has more links in it than the supply has inside it — so the supply is a suspect of last resort, reached after the no-power chain is proven. A supply that's suspected early and swapped on a hunch is how healthy machines get healthy new parts attached to their real, unfixed problems.

The safety rule, stated once and plainly, because it's the rule this section is built around: a power supply's enclosure is never opened. The components inside sit on or near mains voltage, and the large capacitors can hold a dangerous charge long after the unit is unplugged — “it's been unplugged for a day” is not a safe state for an untrained opening. This isn't a “be careful” caveat; it's the reason the home toolbox for a power supply contains no screwdrivers, and the reason every legitimate check below is external. What you can safely do: check the rear switch (on, and reachable — a switch that's cracked or stuck is itself the failure); check the cable and its seat; observe the unit's behavior (fan, sounds, heat, smell — from the outside, at a distance that's comfortable); and verify the two connectors on the desktop side, which is the step that surprises people most. The 24-pin main connector and the CPU power connector (4- or 8-pin, near the processor socket) both have latches, and both can be “in but not fully seated” — a half-seated CPU power connector produces exactly the no-start or start-then-die pattern, and the fix is a firm reseat with the latch clicking. That connector check is the highest-value minute in this section. And the quiet-supply note: many modern supplies run their fan passively at low load (silent until the system warms up), so a PSU fan that doesn't spin at idle is normal on a real share of units — don't read failure into a quiet supply.

The definitive home test, and the laptop distinction. On a desktop, the swap test is the diagnosis: a known-good power supply of adequate rating, installed in place of the suspected one (the connectors are standard, the old one comes out, the new one goes in — with the machine unplugged throughout). If the machine works on the swap, the original supply was the cause, and the repair is a part, which is the good kind of repair: understood, replaceable, and sized by wattage. If the machine doesn't work on a known-good supply, the supply is exonerated and the board is the suspect (that section) — which is why the swap is worth the one borrowed or bought part it costs. For laptops, the “power supply” is the external adapter, and the logic translates cleanly: the known-good compatible charger is the swap test, the jack is the connector to check for looseness or damage, and a brick that's hot, smelly, or damaged is retired without a trial. The stop line for both: a supply that's hot, clicky, or smells — or any machine that smells — is unplugged and finished with for the day; the professional section is where that story ends, and repeated power-on attempts on a suspicious supply are how a part failure becomes a board one.