How to Test a Power Supply: 13 Steps, 60 Min [2026]

Your PC won’t POST, or it reboots mid-game right when the GPU load spikes. Before you spend $700 on a new graphics card or a fresh motherboard, check the part that actually feeds power to everything else: the power supply. A bad or marginal PSU causes more mystery crashes than any other single component, and most builders never test it because they assume the wall outlet and the “on” light mean it’s fine. It doesn’t, and that light doesn’t tell you much.

This tutorial walks through three testing methods across 13 detailed steps, from a five-minute paperclip check to full multimeter rail verification, using the same ATX voltage tolerances that PSU manufacturers test against, updated as of our August 2026 review pass. By the end you’ll know exactly how to test a power supply, read the results against real ATX 3.1 specifications, and decide whether to keep it, RMA it, or replace it. Total time: about 60 minutes if you do all three tests back to back.

·

Google · Preferred Sources

Don't miss new tech stories on Google

Add Tech Insider once in the Google app and our stories appear in your news suggestions.

Add Now

Why Your Power Supply Might Be Crashing Your PC

A power supply doesn’t have to die completely to cause problems. Most PSU-related failures show up as intermittent instability long before the unit refuses to turn on at all. That’s what makes diagnosis tricky: a failing PSU mimics symptoms of a bad GPU driver, an unstable RAM overclock, or a corrupted Windows install.

Watch for these signs before you start testing:

• Random reboots under load, especially during gaming when transient power demand spikes as the GPU boosts.
• No POST or no power-on at all: fans don’t spin, no lights, nothing on the display. This one is the classic dead-PSU symptom, though a bad motherboard, RAM, or front-panel switch can look identical.
• Burning smell or visible scorching around the PSU housing or a power connector. Stop using the unit immediately if you see or smell this.
• Coil whine under heavy load. It’s common in modern high-wattage gaming systems and isn’t automatically a defect, but paired with instability or excess heat it’s worth investigating.
• Shutdowns specifically tied to GPU load, which is the most common complaint on RTX 50-series and RX 9000-series builds because transient current spikes can push a marginal PSU past its limit for a fraction of a second.
• Failed boots after a fresh GPU or storage upgrade, which usually means you’ve crossed the PSU’s real wattage or connector ceiling, not that a part is defective.

If any of these sound familiar, don’t just swap parts at random. Testing the PSU first takes about 15 minutes for the quick checks and saves you from returning a perfectly good graphics card.

Prerequisites: Tools, Software, and Safety Gear

You don’t need every tool listed here for a basic check, but a full three-method test (paperclip, dedicated tester, multimeter) covers every failure mode a home builder can realistically diagnose.

• A Phillips-head screwdriver to remove the PSU from the case if you need full access.
• A standard metal paperclip or a jumper wire for the startup test.
• A dedicated PSU tester (optional but recommended). Budget LCD models run $12 to $20; Thermaltake’s updated Dr. Power III Pro, announced at CES in January 2026, adds four real-time rail readouts (+12V, +5V, +3.3V, +5VSB), automatic and manual test modes, and support for ATX v3.1 supplies up to PCIe Gen5 16-pin connectors, including 150W-600W sense-signal detection on 12V-2×6 cables.
• A digital multimeter with DC voltage measurement, accurate to at least two decimal places. Any $15 to $30 unit from Klein Tools, Fluke, or a generic auto-ranging model works; reviewers chasing lab-grade precision, like Vivid Repairs’ December 2025 teardown of Corsair’s ATX 3.1-certified HX1200i, lean on a Fluke 287 to log multi-rail voltage, though that’s well beyond what a home diagnosis needs.
• HWiNFO (latest release) or HWMonitor for software rail monitoring during a load test. Motherboard-bundled utilities (MSI Center, Armoury Crate, ASUS AI Suite) work too, but their sensor data comes from onboard monitoring chips and is less precise than a direct probe reading.
• A stress-testing tool such as OCCT, AIDA64, or a combined GPU/CPU stress test, to put real load on the rails while you watch for voltage droop. See our GPU and CPU stress test guide if you haven’t run one before.
• A non-conductive work surface: an anti-static mat, a wooden desk, or a cardboard box. Never test a PSU sitting on a metal case panel.
• 15 to 60 minutes, depending on how many of the three test methods you run.

Safety note before you touch anything: unplug the PSU from the wall before opening the case, removing cables, or performing the paperclip test. Modern ATX units hold a small residual charge in their internal capacitors even after the switch is off, so treat every PSU as live until you’ve confirmed otherwise.

Step 1-3: Confirm the Symptoms and Prep for Safe Testing

Step 1: Rule out the obvious first. Check that the wall outlet works (test it with a lamp or phone charger), confirm the PSU’s rear power switch is set to “I” (on), and make sure the 24-pin and CPU power cables are fully seated at both ends. A high number of “dead PSU” cases turn out to be a half-inserted 8-pin CPU connector.

Step 2: Note exactly when the problem happens. Does it crash only during gaming, only on cold boot, only after the system has been running for an hour? GPU-load crashes point to transient response and connector issues. Cold-boot failures point to capacitor aging. Random crashes with no clear pattern often point to a marginal +12V rail under fluctuating load.

Step 3: Power down completely and unplug the system. Shut down, flip the PSU’s rear switch to “O,” then pull the cable from the wall. Press and hold the case power button for five seconds to drain any residual charge from the motherboard capacitors. Now you can safely open the case.

Step 4-5: Visual Inspection and Removing the PSU

Step 4: Inspect the unit and every cable before you test anything electrically. Look for bulging or discolored capacitors on the PSU’s exhaust grille, melted or brown-tinted pins on any 12V-2×6 or 8-pin connector, frayed cable insulation, and dust buildup blocking the fan. A melted connector is a stop-use condition on its own. Replace the PSU and don’t attempt to reuse the damaged cable, regardless of what the rest of the tests show.

Step 5: Disconnect the PSU from every component. Unplug the 24-pin motherboard connector, CPU power (4+4 or 8-pin), GPU power, and all SATA/peripheral cables. If you’re using a dedicated tester or doing the paperclip test, you can leave the PSU mounted in the case for this step. You only need to fully remove it if you want to bench-test it away from the build, which is worth doing if you suspect a short elsewhere in the system.

Step 6-7: The Paperclip (Jumper) Test

The paperclip test is the fastest way to check whether a PSU can start up and hold a stable “on” state outside the case. It’s a pass/fail check, not a health certificate. It tells you nothing about voltage accuracy, ripple, or behavior under real load. Use it to answer one question: does this PSU turn on at all?

Step 6: Bridge the PS_ON pin to ground on the 24-pin connector. With the PSU unplugged, set it on a non-conductive surface. Find the 24-pin ATX connector (the largest one). Locate the green wire (PS_ON, typically pin 16) and any black wire (ground). Bend a paperclip into a U-shape and insert both ends so it bridges the green pin to an adjacent black pin. Attach a minimal load first, such as an old case fan plugged into a 4-pin Molex or a spare SATA drive, because some PSUs won’t stay stable with zero load attached.

24-PIN ATX CONNECTOR - PAPERCLIP TEST WIRING
=============================================
Pin 16: PS_ON  (green wire)   ---
                                  |-- bridge with paperclip
Pin 15 or 17: COM (black wire) ---

Before bridging:
1. PSU unplugged from wall
2. PSU on non-conductive surface
3. Minimal load attached (spare fan or old HDD)
4. No other components connected

Plug in AC power, flip switch to "I": PSU fan should
spin within 1-2 seconds if the unit is functional.

Step 7: Plug in and observe. With the load attached and the jumper bridged, plug the PSU into the wall and flip the rear switch on. A functional unit spins its fan immediately and keeps running steadily. If the fan doesn’t spin, spins and immediately stops, or spins erratically, the PSU has failed the most basic startup check. Unplug it right away. Don’t leave a jumpered PSU running unattended, and don’t touch any exposed pins while it’s energized.

Step 8-9: Testing With a Dedicated PSU Tester

Step 8: Connect the 24-pin (and optionally 4/8-pin CPU, PCIe, and SATA connectors) to the tester. A dedicated PSU tester automates the paperclip test and adds a quick readout of each rail’s voltage on an LCD screen. Plug the 24-pin cable into the tester’s matching port, then power the PSU on from the wall. Budget testers ($12 to $20) cover the essential rails; Thermaltake’s Dr. Power III Pro, unveiled at CES in January 2026, steps that up with four real-time rail readouts, automatic and manual test modes, and dedicated support for ATX v3.1 PSUs up through PCIe Gen5 16-pin connectors, including 150W-600W sense-signal detection on the 12V-2×6 line used by RTX 50-series and RX 9000-series cards, which matters if you specifically want to check the GPU power path.

Step 9: Read the display and compare against ATX tolerances. The tester shows +12V, +5V, +3.3V, and +5VSB (standby) readings, often alongside a pass/fail beep or LED. Compare each number against the ATX tolerance table below. A reading a few tenths of a volt off nominal is normal; anything outside the listed min/max range means the PSU is out of spec.

SAMPLE PSU TESTER LCD READOUT
==============================
+12V  : 12.08 V   [PASS - spec range 11.20-12.60V]
+5V   : 5.03 V    [PASS - spec range 4.75-5.25V]
+3.3V : 3.31 V    [PASS - spec range 3.14-3.47V]
+5VSB : 5.01 V    [PASS - standby rail, +/-5% tolerance]
PG    : OK        (Power Good signal within 100-500ms)

Result: PSU passes basic startup and rail-presence test.
Note: this does NOT confirm rail stability under load.

Step 10-11: Multimeter Testing for Precise Rail Readings

A multimeter gives you the most accurate reading of the three methods, because you’re measuring voltage directly at the connector pin rather than through a tester’s internal circuitry. It’s also the only method that lets you check individual pins rather than a single averaged rail.

Step 10: Set up the multimeter and probe the 24-pin connector while it’s running. With the PSU jumpered on (using the paperclip method above, with a load attached) and running, set your multimeter to DC voltage, dialing a manual-ranging meter to the 20V DC range, the setting a July 2026 PSU-testing tutorial from Ireland Focus recommends for getting a clean, accurate read on rails that sit well under that ceiling. Touch the black probe to any ground pin (black wire) and the red probe to the pin you want to measure. On the 24-pin connector: pin 9 (purple) is +5VSB, pins 4 and 6 (red) are +5V, pins 10 and 12 (yellow) are +12V, and pins 1, 2, and 11 (orange) are +3.3V. Take each reading with the system idling, then again while a stress test is running on another rig or via a load tester if you have one.

Step 11: Log every reading and compare it against the table in the next section. Write down each rail’s idle and, if possible, load voltage. A rail that reads fine at idle but sags outside tolerance under load is exactly the failure mode that causes GPU-load crashes, and it’s the one the paperclip test and most budget testers can’t catch, because neither applies meaningful load.

MULTIMETER LOG - SAMPLE SESSION
================================
Rail     Idle Reading   Load Reading   ATX Range        Result
+12V     12.05 V        11.72 V        11.20-12.60 V    PASS
+5V      5.04 V          4.98 V        4.75-5.25 V      PASS
+3.3V    3.29 V          3.22 V        3.14-3.47 V       PASS
+5VSB    5.02 V          n/a           4.75-5.25 V      PASS

Load test: OCCT Power Supply test, 10 minutes, GPU+CPU combined
Ambient temp: 24C
Verdict: rails hold within ATX tolerance under sustained load.

Step 12-13: Software Load Monitoring and the Final Call

Step 12: Reassemble the system and monitor rails in software while running a real workload. Reconnect everything, boot normally, and open HWiNFO or your motherboard’s monitoring utility. Run a combined GPU and CPU stress test for 10 to 15 minutes while watching the +12V and +5V rails in the sensor panel. Software readings come from onboard sensors rather than a direct probe, so treat them as a trend indicator rather than a certified measurement. Use them to catch voltage droop or instability over time, then confirm anything suspicious with the multimeter method above.

Step 13: Make the call. If every test passes (clean startup, correct tester readout, multimeter readings inside ATX tolerance under load, and stable software monitoring during a stress test) your PSU is healthy and the instability lies elsewhere. Check RAM timings, GPU drivers, or thermal throttling next. If any rail reads outside tolerance under load, or the unit fails to start at all, it’s time to replace it. Don’t attempt to repair a PSU yourself; the internal capacitors can hold a dangerous charge even when unplugged, and it’s not a board-level fix a home builder should take on.

Understanding ATX Voltage Tolerances

Every number in the tables above traces back to Intel’s ATX power supply design guide, the same specification PSU manufacturers test their units against before certification, a compliance pipeline that’s grown into real money: 24MarketReports counted 78,528 professional PSU-test units sold in 2025 at an average price of $2,113 per unit, according to its March 2026 update. Knowing the exact tolerance windows is what turns a vague “seems fine” into an actual pass/fail decision.

Rail	Tolerance	Minimum	Nominal	Maximum
+12V	+5% / -7%	11.20V	12.00V	12.60V
+5V	±5%	4.75V	5.00V	5.25V
+3.3V	±5%	3.14V	3.30V	3.47V
+5VSB (standby)	±5%	4.75V	5.00V	5.25V
-12V (legacy)	±10%	10.80V	12.00V	13.20V

Two things stand out here. First, the +12V rail gets the widest downward tolerance (-7% versus -5% on the other rails), because it’s the rail under the heaviest transient load from the CPU and GPU, and engineers built in extra headroom for momentary sag. Second, under peak current load the practical tolerance window on +12V effectively widens to about 10% under the ATX design guide’s peak-load allowance, which is exactly why a reading like 11.72V under load in the sample log above still counts as a pass, not a red flag. A July 2026 how-to from NogenTech frames the general rule the same way most testers use in practice: keep every rail within about ±5% of its nominal value during testing, and treat the wider +12V downward allowance as the one deliberate exception.

The current baseline spec most vendors reference is ATX Version 3.1, which replaced the original 12VHPWR GPU power connector with the redesigned 12V-2×6 connector and trimmed the required power-loss hold-up time from 17 milliseconds to 12 milliseconds, according to Wikipedia’s ATX specification history. Intel tightened that margin further in its own validation criteria: the company’s Desktop Platforms PSU Test Plan, updated in February 2026 and spanning both the ATX12V and ATX12VO specs, requires the +12V rail to hold up for a full 16 milliseconds at full load, giving system builders more of a buffer against a momentary power blip than the base ATX spec alone implies. This isn’t just a paper spec anymore: Cybenetics has been independently certifying ATX v3.1 units throughout 2025 and into 2026, including the ASRock Phantom Gaming PG-1000G (certified January 6, 2025), the ASRock Taichi TC-1650T (report ID 25PS2737A, February 2025), the Seasonic Focus GX-750 ATX3 (report ID 2749, certified February 20, 2025), the Super Flower Leadex III 750W (report ID 2781, certified April 14, 2025), and Corsair’s RM850x, whose January 2026 review from Vivid Repairs confirmed both ATX 3.1 compliance and PCIe 5.1 transient-response testing. Mercury-PC’s own ATX 3.1 roundup, current as of August 2026, had already put 8 models through the same PCIe 5.1-readiness gauntlet. That same March 2026 24MarketReports update pegged the overall PSU-testing equipment market at $161M for 2026, with growth projected to $263M by 2034, a sign the compliance testing this spec demands isn’t slowing down. If your PSU’s label or manual mentions ATX 3.0 or 3.1 and a 12V-2×6 (or still 12VHPWR) connector, these are the tolerance numbers it was built and certified against.

80 PLUS Efficiency Ratings Explained

Efficiency rating isn’t a health test, but it matters for testing context. A PSU running far below its rated efficiency at a given load, burning more watts as heat than the certification implies, is often an early sign of aging capacitors, even before rail voltages drift out of tolerance. Knowing your unit’s certified efficiency numbers gives you a baseline to compare against if you ever measure input-versus-output power with a wall meter.

Certification	20% Load (115V)	50% Load (115V)	100% Load (115V)
80 PLUS (White)	80%	80%	80%
80 PLUS Bronze	82%	85%	82%
80 PLUS Silver	85%	88%	85%
80 PLUS Gold	87%	90%	87%
80 PLUS Platinum	90%	92%	89%
80 PLUS Titanium	92%	94%	90%

These figures come from the official 80 PLUS certification program, and they’re measured at three load points because a PSU rarely runs at 100% capacity in a real system. Most gaming rigs sit closer to the 20-50% range even during heavy sessions, which is exactly where Bronze through Titanium units are tuned to hit their peak efficiency. See How-To Geek’s breakdown of the certification program and Wikipedia’s full 80 PLUS ratings tables for the complete 230V figures too, if you’re building or troubleshooting outside North America.

PSU Tester Comparison: Which Model to Buy

You don’t need an expensive tester for basic diagnosis, but if you’re regularly building or repairing PCs, the jump from a $13 unit to a premium model buys real functionality, mainly around newer GPU power connectors. Case in point: Thermaltake’s Dr. Power III Pro, announced at CES in January 2026, backs its higher price with four real-time rail readouts, automatic and manual test modes, PCIe 150W-600W sense-signal detection on 12V-2×6 connectors, and a 3-year warranty, well beyond what budget LCD testers offer; as of our August 2026 pricing check, Tech-Insider listed it at $44 to $48.

Tester	Price (USD)	Connectors Supported	Best For
Generic LCD ATX tester	$12-$16	20/24-pin, 4/8-pin, SATA, Molex	Quick pass/fail checks on standard builds
Kingwin Digital PSU Tester	$16.99	20/24-pin, 4/8-pin CPU, SATA, Molex, floppy	Home builders wanting a bundled screen readout
Newegg/branded LCD tester	$18-$19	20/24-pin, 4/8-pin, SATA, Molex	Budget backup unit for a home workshop
Thermaltake Dr. Power III	$44-$48	Adds PCIe 12+4-pin (ATX 3.1 / 12V-2×6)	Testing modern RTX 50-series / RX 9000-series GPU power paths

If you’re specifically troubleshooting a system built around a current-generation GPU, connector support matters more than the price gap suggests. Most sub-$20 testers were designed before the 12+4-pin PCIe connector became standard, so they can confirm your 24-pin and CPU rails are healthy but tell you nothing about the dedicated GPU power path, which is precisely where the melting and partial-seating issues around 12VHPWR and 12V-2×6 connectors have shown up in real-world reports. For that check, Thermaltake’s Dr. Power III Pro, launched at CES in January 2026 with 150W-600W PCIe sense-signal detection on 12V-2×6 cables and a 3-year warranty, and whose Newegg listing confirmed full ATX12V v3.1 support as of August 2026, or an equivalent ATX v3.1-rated tester, is worth the extra cost.

The 12V-2×6 Connector and Why Cable Seating Matters

If your PSU testing turns up a healthy unit but you’re still seeing shutdowns specifically under GPU load, the connector itself deserves a second look before you blame the PSU. The most consistently reported cause of localized heating on 12VHPWR and its successor, 12V-2×6, isn’t a defective PSU or GPU. It’s incomplete seating combined with a tight cable bend near the connector housing.

Here’s the mechanism: when the connector isn’t fully clicked in, or the cable is forced into a sharp bend right at the plug (often to route it behind a case’s side panel), the individual power pins lose contact area. Electrical resistance rises at that contact point, and because the current running to a high-end GPU is substantial, even a small resistance increase generates disproportionate heat, the same physics behind any loose electrical connection overheating. The fix costs nothing: reseat the connector fully until it clicks, and leave at least 35mm of straight cable before the first bend.

• Push until it clicks. A connector that looks “in” but isn’t fully seated is the single most common cause of GPU power connector heating.
• Avoid tight bends near the plug. Route the cable so it curves gradually, not immediately at the connector housing.
• Never daisy-chain adapters unnecessarily. Each adapter joint is another point where resistance can creep up.
• Check the connector after the first week of use. A slight discoloration or warmth (beyond general case airflow warmth) on the plug is worth investigating immediately. Unplug the system and inspect before continuing to use it.

This is also why the testing steps above specifically call out visual inspection of the connector before you touch a multimeter to anything. A perfect voltage reading on the 24-pin rail doesn’t rule out a connector-level heating problem on the GPU’s dedicated power cable.

Common Pitfalls When Testing a Power Supply

Most bad PSU diagnoses come from skipping a step, not from using bad tools. Here’s what trips up builders most often.

• Treating the paperclip test as a full diagnosis. It only confirms the PSU can start and idle. A unit that spins its fan on the paperclip test can still fail catastrophically the moment you put real GPU load on it. Always follow up with a multimeter or software load test if the paperclip test passes but symptoms continue.
• Testing with zero load attached. Some PSUs, especially older or budget non-modular units, won’t hold a stable output with nothing connected, and will report false-fail behavior (fan surging, shutting down) that has nothing to do with an actual defect. Always attach at least a spare fan or drive.
• Reading software sensor values as gospel. Motherboard monitoring software estimates rail voltages through onboard sensors that vary in accuracy between vendors. Use software to spot trends and instability, then confirm with a direct multimeter reading at the connector before making a replacement decision.
• Skipping the visual inspection. A perfectly “in spec” voltage reading doesn’t matter if there’s a scorched connector pin about to fail. Visual inspection catches physical problems electrical testing can miss entirely.
• Testing on a conductive surface. Setting a PSU on a metal case panel or the conductive kind of anti-static bag (not the pink foam kind) during a paperclip test risks shorting the unit or damaging it further. Use wood, cardboard, or a proper insulated mat.
• Ignoring standby (+5VSB) rail issues. A failing standby rail causes wake-from-sleep problems and intermittent boot failures that look like a motherboard issue. It’s the rail most builders forget to check because it’s not on the visible startup path.
• Assuming any voltage reading close to 12V, 5V, or 3.3V is “fine.” Compare against the actual tolerance windows in the table above, not a rough eyeball guess. A rail reading 10.9V on +12V looks close to nominal but is actually out of the ATX spec’s minimum.

Troubleshooting: What Your Readings Actually Mean

Once you’ve collected readings from one or more of the three test methods, use this list to translate them into an actual diagnosis.

QUICK DECISION CHECKLIST
=========================
[ ] Wall outlet confirmed working
[ ] All power cables fully seated (24-pin, CPU, GPU)
[ ] No burning smell / no melted or discolored pins
[ ] Paperclip test: fan spins and stays on
[ ] Tester/multimeter: +12V within 11.20-12.60V
[ ] Tester/multimeter: +5V within 4.75-5.25V
[ ] Tester/multimeter: +3.3V within 3.14-3.47V
[ ] Rails hold steady under a 10-min stress test
[ ] No new coil whine paired with instability

All boxes checked -> PSU is healthy, look elsewhere.
Any box unchecked -> retest that item, then replace
the unit if the failure repeats.

• Fan doesn’t spin at all during the paperclip test. The PSU has failed the most basic function check. Confirm the wall outlet works and the rear switch is on before concluding the unit is dead, then replace it.
• Fan spins briefly, then stops. This usually indicates the PSU’s protection circuitry is tripping, often due to insufficient load, a short somewhere in the tested cables, or an internal fault. Try again with a proper load attached; if it still cuts out, the unit is failing.
• +12V reads correctly at idle but drops below 11.20V under load. This is the classic “fine until you game” failure. The PSU can’t hold its rail under transient demand. Replace it, especially if you’re running a current-generation GPU.
• +3.3V or +5V reads outside tolerance while +12V is fine. Less common on modern GPU-heavy builds since most of the load sits on +12V, but still a fail. Older SATA drives and some motherboard VRMs depend on these rails directly.
• +5VSB reads outside tolerance. Expect wake-from-sleep failures, USB devices not charging while the system is off, and occasional failure to power on from a cold start. Replace the unit even if the main rails test fine.
• Tester shows “PG” (Power Good) failure or delayed signal. The Power Good signal should arrive within roughly 100-500 milliseconds of power-on. A missing or badly delayed signal means the motherboard may refuse to complete POST even if the raw voltages look acceptable.
• Readings pass every test, but crashes continue. The PSU is very likely not the cause. Move to checking RAM stability, especially if you’re running an XMP/EXPO overclock (see our DDR5 overclocking guide), GPU driver version, and thermal throttling next.
• Coil whine present but all voltages test within tolerance. Usually cosmetic and audible only, not a failure. If it’s new, loud, or paired with any of the other symptoms above, treat it as a warning sign worth monitoring rather than an immediate replacement trigger.
• System boots fine on integrated graphics but fails with the GPU installed. This points to insufficient PCIe or GPU-connector power rather than a general PSU fault. Recheck your PSU’s rated wattage against your GPU’s recommended power supply, and inspect the 12V-2×6/PCIe connector seating as described above.

Advanced Tips: Testing Under Real Gaming Load

The three core methods catch most failures, but a few advanced techniques help when you’re chasing an intermittent problem that only shows up occasionally.

Log sensor data over an entire gaming session, not just a synthetic benchmark. Synthetic stress tests like OCCT’s power supply test apply sustained, predictable load, which is useful for finding a baseline problem but doesn’t always reproduce the sharp, brief power spikes that real games generate when a GPU boosts clock speeds mid-frame. Set HWiNFO to log continuously in the background during a two-hour play session, then review the log for any voltage dips that coincide with the timestamps of a crash or stutter.

Size your replacement PSU around transient spikes, not average draw. A system built around a top-tier RTX 50-series card can momentarily demand well above its rated TDP for milliseconds at a time, which is why current build guides recommend generous headroom rather than sizing a PSU to exactly match a GPU’s advertised wattage. An April 2026 RTX 5090 PSU roundup from Siemens-Mobile makes the same point from the flagship end: it stresses buying a unit with a native 12V-2×6 connector and full ATX 3.1 testing behind it, rather than relying on an adapter, precisely because that card’s transient spikes are the hardest to absorb. A mid-tier build, something in the RTX 5070/5070 Ti or RX 9070 class, needs meaningfully less headroom than a flagship-tier system, but the same margin-of-safety principle applies at every tier.

SIMPLE PSU SIZING WORKSHEET
============================
GPU rated TDP:              320 W
CPU rated TDP:              125 W
Motherboard/RAM/storage:     60 W
Fans/AIO pump/RGB:            30 W
------------------------------------
Sub-total (steady state):   535 W

+ Transient spike headroom (30%):  ~695 W
+ Future-upgrade margin:          ~750-800 W target PSU rating

Rule of thumb: size for steady-state load plus 30%
transient headroom, then round up to the nearest
common wattage tier (650W / 750W / 850W / 1000W).

Cross-check GPU-load shutdowns against undervolting. If your PSU tests clean but you still see shutdowns specifically at peak GPU boost clocks, a slight GPU undervolt can reduce transient spikes without materially affecting performance, buying you time before a PSU replacement. Our GPU undervolting tutorial covers the process step by step.

Test with the case side panel on, not off. A PSU tested outside the case, with an open-air fan, will report better thermal behavior than the same unit crammed into a poorly ventilated case. If you’re chasing a heat-related intermittent fault, always run your final verification test with the case fully closed and running under normal airflow conditions.

Keep a written log across multiple test sessions. A single reading tells you the PSU’s state at one moment. A log spanning a week of testing (idle, gaming, cold boot) reveals whether a rail is degrading gradually, which is common in aging capacitors, versus failing suddenly, which usually points to a specific component fault.

Building the Complete Project: A Home PSU Test Bench

If you build or repair PCs regularly, it’s worth assembling a small permanent test bench rather than repeating the manual paperclip setup every time. Here’s a complete, working setup you can build in under an hour.

• An old ATX case or an open-frame test bench mount to hold a “known good” motherboard and PSU pair as a reference point.
• A permanently wired jumper switch, a simple momentary switch soldered to a spare 24-pin connector’s PS_ON and ground pins, instead of a paperclip, so you can start any PSU with the flip of a switch.
• A dedicated PSU tester left connected as a permanent first-pass readout every time you test a new unit.
• A load bank: a couple of retired mechanical hard drives and an old case fan, wired to always be attached during a test, so you never accidentally test a PSU at zero load.
• A multimeter with alligator-clip probes left semi-permanently attached to a spare 24-pin connector’s ground and +12V pins, so you can get a fast reading without fiddling with test probes each time.
• A logging spreadsheet or notes file where you record every PSU you test: model, wattage, certification, idle/load readings, and verdict. Over time this becomes a genuinely useful reference for which brands and models in your experience hold up and which don’t.

This setup turns a 45-minute diagnostic process into a five-minute check, which matters if you’re testing PSUs for friends, a small repair business, or just your own growing pile of “maybe this one still works” units in a closet.

When to Replace, RMA, or Repair

Once testing confirms a fault, the next question is what to do about it. Don’t attempt an internal repair yourself. PSU capacitors can retain a dangerous charge long after the unit is unplugged, and it’s not a safe project for a home workbench.

Check your warranty first. Premium PSU brands back their units for years, not months. Corsair and Seasonic commonly offer 10-year (and in some Seasonic lines, up to 12-year) coverage on higher-tier models, be quiet! typically covers 5 to 10 years depending on the series, and Cooler Master’s coverage varies similarly by product line. EVGA still sells and supports PSUs under its own warranty terms even after exiting the graphics card business, so don’t assume a brand exit from GPUs means their PSU support disappeared too. If your unit is inside its warranty window and fails a documented test, RMA it rather than buying a replacement outright.

Know the realistic lifespan. A quality gaming PSU typically lasts 5 to 10 years of regular use, with higher-end 80 PLUS Gold and above units often running longer if kept clean and away from sustained high-load conditions. Heat, dust accumulation, and long stretches at high load are the biggest factors that shorten a PSU’s working life, which is also why a case with decent airflow protects more than just your GPU and CPU temperatures.

Buy up, not sideways, when replacing. If your old PSU failed under a GPU upgrade’s load, don’t just buy an equivalent-wattage replacement. Recalculate your full system’s power draw with headroom for transient spikes, and prioritize a unit with native ATX 3.1 / 12V-2×6 support if you’re running a current-generation GPU. See our graphics card installation guide for the full connector and clearance checklist when you’re ready to install the new pairing.

Frequently Asked Questions

Is the paperclip test safe to do myself?
Yes, as long as you follow the sequence: PSU unplugged from the wall while you set up the jumper, placed on a non-conductive surface, with a load attached, and you never touch exposed pins once it’s powered on. It’s the same basic startup check technicians use, just without dedicated test equipment.

Can a power supply test fine and still be the cause of my crashes?
Yes. The paperclip test and many budget testers only confirm the unit starts up and holds nominal voltage with light or no load. A rail can pass every basic check and still sag out of tolerance the moment a GPU pulls a real transient spike during gaming, which is why the multimeter and software load-monitoring steps matter for genuinely intermittent problems.

What voltage should the +12V rail read on a healthy PSU?
Between 11.20V and 12.60V under the ATX specification’s +5%/-7% tolerance, with 12.00V as the nominal target. A reading anywhere in that window, including under load, counts as within spec.

Do I need a multimeter, or is a dedicated PSU tester enough?
A dedicated tester covers most home-diagnosis needs and is faster to use. A multimeter is worth adding if you want pin-level precision, need to test under actual load rather than the tester’s light-load state, or want to verify a specific rail the tester doesn’t break out individually.

Why does my PSU only cause problems when the GPU is under load?
This is the most common PSU-related complaint on current-generation builds. It usually means either the +12V rail is sagging under the GPU’s transient current spikes, or the dedicated GPU power connector isn’t fully seated and is generating localized heat and resistance. Test both: rail voltage under a stress test, and a visual/physical recheck of the 12V-2×6 or PCIe connector.

Is it normal for a PSU to make coil whine noise?
Some coil whine under heavy load is common on modern high-wattage units and doesn’t by itself indicate a fault. Treat it as a warning sign only if it’s new, unusually loud, or shows up alongside voltage readings outside ATX tolerance or actual instability.

How often should I test my PSU if I’m not having problems?
There’s no need for routine testing on a healthy system. Test after any GPU or major component upgrade, if you notice new coil whine or instability, or roughly once a year on a PSU that’s already past its fifth year of service, since that’s when capacitor aging starts becoming a realistic factor.

Will testing my PSU void the warranty?
External testing (the paperclip test, a dedicated tester, or multimeter probing at the connector) doesn’t open the unit or void the warranty. Never open the PSU casing itself; that does void coverage on virtually every brand and exposes you to a real electrical hazard from charged internal capacitors.