Card Works in Another PC but Not Yours? How to Run a Valid Cross-Machine Swap Test — Card vs System Diagnosis Guide

Key Takeaways
When a graphics card fails in your PC but works in another, the swap alone proves very little. This guide shows how to change one variable at a time, read each outcome correctly, and tell a card-layer fault from a host-system fault before deciding what to inspect next.
Why a Graphics Card Never Runs Alone
A graphics card is only one part of a working display chain. In any host machine it depends on the PCIe slot and its lane wiring, the power supply and the cables feeding it, the motherboard BIOS or UEFI settings, the stability of system memory, the state of the driver and the operating system, the monitor and its cable, and the cooling and airflow inside that particular case. Change the host and you change most of those conditions at once, even when the card itself has not changed at all.
That is why the sentence it works in another PC is a starting observation, not a conclusion. It tells you that the card can produce a picture under some conditions. It does not tell you that the card is fault-free, and it does not tell you that your own machine is fault-free either. A second machine may simply be more tolerant: a stronger power supply, a different slot, a different BIOS configuration, a cleaner driver state, or a cooler case can all mask a marginal fault that your own machine exposes. Conversely, a fault that appears in your machine may have nothing to do with the card at all, and may be waiting in the host system for any card placed in it.
The purpose of a cross-machine swap test is therefore narrow and disciplined. It is not to prove that a card is good or bad in general. It is to separate, as cleanly as possible, two questions: does the fault follow the card, or does it stay with the first machine? Everything in this guide serves that separation, and nothing here tries to diagnose a deeper layer once the separation is made. Deeper layers have their own guides, and they are pointed to rather than rewritten here.
The One-Variable Rule
The single most common error in swap testing is changing several things at once and then treating the result as an answer. If you move the card, change the cable, use a different monitor, and update the driver in the same step, a pass tells you nothing about which change mattered, and a failure tells you nothing about which change caused it. A valid test changes one variable at a time and records the rest.
Start by fixing the test itself. Use the same test that failed before: the same game, the same workload, the same symptom to watch for. A card that crashed in a demanding game has not been cleared by idling on a desktop in another machine. Use the same monitor and cable where possible. If that is not possible, test the monitor and cable separately first, so they are not silently swapped along with the card. Keep the power cabling arrangement comparable: the same number of separate cables, not a splitter where the original used separate runs, and not separate runs where the original used a splitter. Note the slot used in each machine, because a top slot and a lower slot can differ in lane source and in PCIe generation. Note BIOS settings that differ between the machines, including above-4G decoding, resizable BAR, CSM versus UEFI mode, and any forced PCIe generation setting. Do not change several of those settings at once in the hope of a quick pass.
Power off the machine and disconnect its power before moving any card. Do not force a card into a slot. Do not mix modular power cables between power-supply brands or models, because modular cables are not interchangeable even when their plugs look similar. If you borrow a card, do so only with the owner's agreement, and note its original slot, cables, and driver state so the machine can be restored exactly as it was. A borrowed card that returns to a changed machine has created a second problem while you were trying to solve the first.
A second machine is only a valid control machine if it is itself stable. Before using it as a reference, check its own stability with a light, known workload. A second machine that is itself unstable, overheating, or under-powered cannot tell you whether your card is good; it can only add its own faults to the picture.
Setting Up the Test in Order
A useful order keeps the comparison honest and the record readable afterwards. First, write down the original fault in plain terms: what symptom appears, what triggers it, how long it takes to appear, and whether a restart clears it. Second, write down the original configuration: slot, power cables, monitor, cable, driver state, and any BIOS settings you know differ from defaults. Third, decide the single test workload that will be repeated in both machines. Fourth, move only the card, keeping everything else as close to the original as the second machine allows, and record every difference you could not avoid. Fifth, run the same workload for long enough to give the original fault a fair chance to appear.
| Item to record | In the first machine | In the second machine | Why it matters |
|---|---|---|---|
| PCIe slot used | Top slot, lower slot, or other | Top slot, lower slot, or other | Slots can differ in lane source and generation |
| Power cables | Number of separate cables and their routing | Same arrangement or noted difference | A changed cable arrangement changes the power test |
| Monitor and cable | Exact monitor and cable | Same set, or separately tested set | A different display chain is a different test |
| Test workload | Game or task that triggered the fault | Identical game or task | A light task cannot clear a heavy-load fault |
| Duration | Time until the symptom appeared | Same or longer run time | A short pass is not yet a pass |
| BIOS differences | Above-4G, resizable BAR, CSM or UEFI, PCIe generation | Same settings or noted differences | BIOS defaults can hide or expose a fault |
That record is the test. Without it, two people can look at the same swap and reach opposite conclusions, and neither can show their working.
Reading the Four Outcomes
A swap test produces one of a small number of patterns, and each pattern means something different. Reading them carefully is more important than running the test quickly.
Outcome A: the suspect card fails in both machines, with the same symptom and the same trigger. This makes the card layer the prime suspect. It does not, by itself, say which card sub-layer is at fault. A power stage, memory, core, cooler or fan, or firmware layer can each produce a failure, and the swap alone cannot distinguish between them. The correct next step is to route by symptom to the guide for that layer, not to guess a sub-layer from the swap result.
Outcome B: the suspect card works in the second machine, and a known-good card also fails in the first machine. This makes the host-system layer the prime suspect. It does not identify which host part is responsible. The power supply and its cable arrangement, the slot and its seating or damage, lane sharing with other devices, BIOS settings, system memory stability, driver and operating-system state, and the monitor and cable are all still on the table, each with its own pointer. The correct next step is to separate those host suspects one at a time, not to replace parts at random.
Outcome C: the suspect card works in the second machine, but no known-good card was tried in the first machine. This test is incomplete. The second machine may simply be more tolerant, with a stronger power supply, a different slot, or a different driver state. Until a known-good card is tried in the first machine, you cannot tell whether the first machine would fail with any card or only with this one. An incomplete test should be labelled incomplete, not reported as a pass.
Outcome D: both cards fail in the first machine, but with different symptoms. Do not merge them into one fault. Treat each symptom separately, with its own trigger and its own record. Two different symptoms in the same machine may share a cause, but assuming that they do is exactly how a second, unrelated fault gets missed.
| Swap outcome | What it suggests | What it does NOT prove |
|---|---|---|
| Suspect card fails in both machines, same symptom | Card layer is the prime suspect | Which card sub-layer is at fault |
| Suspect card works elsewhere, known-good card fails in the first machine | Host-system layer is the prime suspect | Which host part is responsible |
| Suspect card works elsewhere, no known-good card tried in the first machine | Test is incomplete | That the card is good |
| Both cards fail in the first machine with different symptoms | Separate symptoms to investigate | A single shared cause |
If the Result Points to the Host System
A host-side result is the start of a second separation, not the end of the work. Each host suspect has its own pointer, and each should be checked on its own evidence rather than by swapping everything at once.
Power supply capacity and cable arrangement come first in many host-side cases, because a marginal supply can pass a light task and fail under load, and a changed cable arrangement can change the result without changing the card. Slot seating, slot damage, and lane sharing with other devices are next: a card that is not fully seated, a damaged slot, or lanes shared with another device can produce faults that look exactly like a card fault. Motherboard BIOS settings are a quieter suspect, because a reset to defaults or a forced PCIe generation can change behaviour without any visible change to the hardware. System memory instability is a separate guide in its own right, and the distinction between memory-related crashes and graphics-card crashes should be followed there rather than rewritten here. Driver and operating-system state is likewise a separate guide: a clean driver reinstall is a step in that guide, not a step to improvise inside a swap test. Monitor and cable faults are covered by the no-signal and port guides, and they should be pointed to rather than re-diagnosed here.
The discipline is the same at this stage as at the start. Change one host variable, repeat the same test, and record the result. A host fault that disappears after three simultaneous changes has not been found; it has been hidden again, and it will return when one of those changes is undone.
If the Result Points to the Card
A card-side result routes by symptom, and only by symptom. A failure that appears as a power-related shutdown points to a different follow-up from a failure that appears as visual corruption, and both differ from a failure that appears as overheating or fan behaviour. Power-stage fault, memory fault, core fault, cooler and fan fault, and firmware fault each have their own guides. This article does not re-diagnose those layers, because collapsing them into a swap guide would blur the very separation the swap was meant to achieve.
What the swap result does give you is a cleaner question to take into the next guide. Instead of asking whether the card or the machine is at fault, you can now ask which card layer matches this symptom, this trigger, and this timing. That narrower question is easier to answer honestly, and it protects you from replacing a card for a host fault or replacing host parts for a card fault.
Edge Cases That Break a Short Test
A fault that appears only under heavy load may not appear in a short swap test at all. If the original fault needed twenty minutes of a demanding game, a five-minute desktop session in the second machine proves nothing. The swap test must run long enough, and under the same load that triggered the fault. Until then, the correct label is not reproduced yet, not card proven good.
A fault that appears only at cold start, or only when hot, belongs in part to a temperature-timing guide, and the swap test should record temperature timing rather than ignore it. Note whether the fault appears from cold, after warming up, or only after a long session, and repeat that timing in the second machine.
A card that works in the second machine only because that machine forces a lower PCIe generation or disables a feature has not been proven good. It has been proven tolerant. A lower generation or a disabled feature changes the conditions of the test, and the card may fail again as soon as the original conditions are restored. Record any such difference explicitly, and do not report the result as a clean pass.
A pass on a shop test bench, with a different power supply and a different case, carries the same warning. It shows that the card can work under bench conditions. It does not prove that a home power supply is adequate, that a home slot is sound, or that a home driver state is clean. Bench results are useful, but they answer a bench question, not a home question.
Misconceptions to Leave Behind
It works elsewhere so it must be fine is false. Tolerance differs between machines, and a marginal card can pass in a tolerant machine while failing in a stricter one. The swap narrows the fault; it does not certify the card.
My machine is new so it cannot be the host is false. New machines have BIOS defaults, lane sharing, and driver states too, and a new power supply can still be marginal for a particular card under a particular load. Age is not evidence.
Swapping the card and the cable and the monitor all at once and getting a pass does not identify the culprit. It only tells you that something in that group mattered, and the group will need to be separated again before any part is blamed.
A single swap that changes several things at once proves nothing, however confident the result feels. The confidence should come from the record — same test, same trigger, same duration, noted differences — not from the speed of the swap.
Stop Points and What to Record Before Seeking Inspection
Stop testing if you notice a burning smell, scorch marks, a melted connector, a cracked board, a loose heatsink, or a detached fan. Stop if the test starts to feel like opening the card, rewiring power, or flashing firmware: none of those are part of a swap test, and none of them should be improvised to make a swap pass. Stop if repeated crashes risk data loss. In each of those cases, further swapping does not add information; it adds risk.
Before seeking inspection, your record should be able to answer five questions without guessing: what was the original symptom and trigger, what was the same test in both machines, what differed between the machines and by how much, what was the outcome in each machine, and whether a known-good card was tried in the first machine. A record that answers those five questions lets the next step start from evidence rather than from retelling.
For inspection in Hong Kong, the shop is Room 1209, Dali Plaza, Shek Mun. The telephone number is 2186 8010 and the WhatsApp number is 6558 6806. Bring the record described above, including the slot, cable arrangement, and test workload, so the inspection can start from the swap result rather than repeating it blindly.
Frequently Asked Questions
Does a card working in another PC prove the card is good?
No. It proves the card can work under the second machine's conditions. A more tolerant power supply, a different slot, a different BIOS configuration, or a cleaner driver state can all mask a marginal fault. Only a complete test, including a known-good card in the first machine, can separate a card fault from a host fault.
Does a card failing in another PC prove the card is faulty?
It makes the card layer the prime suspect when the symptom and trigger are the same in both machines, but it does not identify which card sub-layer is at fault. Power, memory, core, cooler, and firmware layers still need to be separated by symptom in their own guides.
Why must the same game or workload be used in both machines?
Because a fault that appears only under heavy load will not appear under a light task. Using a different, lighter task in the second machine changes the test itself, and a pass under that lighter task cannot clear a fault that was triggered by the original workload.
What if I do not have a known-good card to try in the first machine?
Then the test is incomplete, and it should be reported as incomplete. A card working elsewhere without that second check cannot distinguish a tolerant second machine from a faulty first machine. Seek a properly controlled check rather than treating the partial result as a conclusion.
Can a different monitor or cable invalidate a swap test?
Yes, if the monitor or cable is changed along with the card without being tested separately. The display chain is part of the test. Use the same monitor and cable where possible, or test the monitor and cable separately first, so a display fault is not mistaken for a card or host fault.
Does the PCIe slot matter in a swap test?
Yes. A top slot and a lower slot can differ in lane source and generation, and lane sharing with other devices can change behaviour. Always note which slot was used in each machine, and do not treat a pass in a different slot as identical to a pass in the original slot.
Should BIOS settings be changed during a swap test?
BIOS differences should be noted, not changed several at a time. Above-4G decoding, resizable BAR, CSM versus UEFI mode, and a forced PCIe generation can all affect the result. Changing several settings at once destroys the comparison, just as changing several hardware parts at once does.
How long should a swap test run?
At least as long as the original fault needed to appear, and under the same load. A short pass after a fault that normally takes twenty minutes is labelled not reproduced yet, not card proven good. Duration is part of the test, not an optional extra.
What if both cards fail in my machine but with different symptoms?
Treat each symptom separately. Do not merge different symptoms into a single fault. Each symptom has its own trigger and its own record, and merging them risks missing a second, unrelated fault while chasing the first.
When should I stop swapping and seek inspection?
Stop if there is a burning smell, scorch marks, a melted connector, a cracked board, a loose heatsink, or a detached fan. Stop if repeated crashes risk data loss, or if the next step would involve opening the card, rewiring power, or flashing firmware. Those are not swap-test steps, and further swapping at that point adds risk without adding information.