How to Test a Graphics Card: Stability, Temperature and Artifact Checks

To test a graphics card, run a looping graphics load such as a stress test or a demanding benchmark scene while a monitoring tool logs temperature, clock speed, and power. A healthy card finishes the run without crashes, visual glitches, or large drops in clock speed.

Free utilities cover all of this: one tool creates the load, another records what the card does, and your eyes check for artifacts. The same method works for a new card, an overclock, a suspect card that crashes in games, and a used card you are about to buy. This guide explains what each kind of tool measures, how to read a run, and what to look for on a used card.

What Free Tools Measure

A GPU tester usually does one of two jobs: it generates load, or it records data. You normally need one of each.

An “FPS checker” is simply an overlay that displays frames per second while a game runs. It does not stress the card, but it shows how the card behaves during real use. Download any of these tools from the developer’s official site, and check the current documentation for supported GPUs.

Setting Up a Safe Run

Do a few checks before pushing a card hard.

1. Install the current driver and close unneeded programs.
2. Open a monitoring tool and make sure sensors report values, including temperature, clock speed, fan speed, and power.
3. Make sure the case has airflow and the card’s fans can spin freely.
4. Save your work. If a card is unstable, a stress run can freeze or restart the PC.
5. Start with a short run of a few minutes, then extend it if all looks fine.

Some stress tools place an unusually heavy load on the card, heavier than any game. Many reviewers suggest limiting such runs, and it is wise to stop immediately if temperatures climb toward the limit listed in the manufacturer’s documentation or if the fans sound wrong. A repeating benchmark scene is a gentler alternative.

How to Read a Stress Run

Temperature

Watch the GPU temperature as it rises and levels off. NVIDIA publishes a maximum operating temperature for each GPU, commonly in the high 80s to low 90s Celsius for recent cards, and the card slows itself down when it gets close. A temperature that rises steadily without leveling off, or sits at the limit with clocks dropping, points to a cooling problem such as dust, a failing fan, or poor case airflow.

Clock speed and throttling

A stable card holds a fairly steady clock. If the clock drops sharply during the run, the card may be limited by temperature, power, or both. Monitoring tools often show the reason for throttling. Modest changes as the card warms up are normal; large swings suggest a problem.

Power and fan behavior

Power draw should be consistent with the card’s rated board power. A very low reading while the load is high can mean the card is limited by a weak power supply or a poor cable connection. Fans should ramp up smoothly as temperature climbs, with no rattling or grinding.

Signs of Artifacts and Instability

Artifacts are visual errors caused by a GPU or its memory failing to compute correctly. Watch the screen closely during the run.

• Flickering or flashing colored dots, lines, or blocks across the image.
• Checkerboard or striped patterns, especially at high memory load.
• Distorted or stretched geometry, missing textures, or color corruption.
• A black screen or sudden driver reset, sometimes with a message that the display driver stopped responding.
• A system freeze or restart under load.

Possible causes include overheating, an unstable overclock, a weak power supply, a loose connector, or a failing card. To narrow it down, remove any overclock, reseat the card and power cables with the PC off, clean the cooler, and run again. If artifacts persist at default settings, the card may be defective, and a warranty claim or return is the safe path.

Checking a Used Card Before You Buy

If you buy a used card, a short inspection and test routine catches many problems.

1. Inspect the exterior. Look for bent fins, cracked shrouds, missing screws, burn marks on connectors, and corroded or scratched PCIe contacts.
2. Check the fans. They should spin freely and quietly without wobble.
3. Confirm the model. Open GPU-Z and compare the reported GPU name, memory size, and memory type with the listing. Counterfeit or relabeled cards exist.
4. Run a stress loop. Several minutes of a graphics load while logging sensors, then a longer run if possible. Watch for artifacts and throttling.
5. Run a memory test. A video memory test can expose faulty memory chips that a short benchmark misses.
6. Check the ports. Test each display output you plan to use.
7. Confirm the return policy. If you cannot test in person, buy only from a seller that accepts returns.

Be aware that cards used for mining or heavy workloads may have worn fans or dried thermal compound. That does not make them bad, but it affects how long they last and whether a cleaning or repaste is needed.

What a Passing Result Looks Like

A healthy card completes the run with no artifacts, no crashes, steady clocks, and a temperature that levels off below the manufacturer’s limit. Scores in a repeatable benchmark should be in line with published results from other owners of the same model; look for a result in a typical range rather than a specific number. If your card scores far below others, check that the power supply cables are connected, that the card sits in the right slot, and that the power mode is correct.

Testing After an Overclock or Undervolt

If you adjust clocks or voltage, stability checking becomes more important. Change one setting at a time, in small steps, and run a load after each change. A card can seem fine in a short run and then crash minutes into a game, so extend the run for a setting you plan to keep. If you see artifacts, a driver reset, or a freeze, return to the previous step or to default settings.

Real games and applications are a useful final check, since each loads the card differently. A setting that survives a synthetic loop may still fail in a particular title, especially one that stresses memory. Keep monitoring running in the background, and keep a record of what you changed so you can undo it. Most tools offer a reset to defaults, which is the safe path if something goes wrong.

Common Causes When a Run Fails

When the test fails, the cause is not always the card itself. Work through the likely causes in order, from easiest to hardest to fix.

1. Dust or blocked airflow, which causes overheating.
2. A loose or partially seated power connector, which causes crashes under load.
3. An outdated or corrupted driver, which a clean reinstall may fix.
4. An overclock or a memory profile that is too aggressive.
5. A weak or aging power supply that cannot deliver steady power.
6. A failing card, which shows artifacts or crashes even at default settings.

If the same fault appears on another system, the card is the likely culprit. If the fault disappears in another PC, look at the power supply, the motherboard slot, and the system’s cooling instead.

Related guides

• What Does Manage Gpu Performance Counters Mean On Nvidia Hardware
• Low Profile Nvidia Graphics Cards For Small Form Factor Pcs
• Nvidia App Error Your System Configuration Is Not Supported

Frequently Asked Questions

What is the best free tool to test a GPU?

There is no single best option. A common combination is a stress or benchmark tool, such as OCCT or Unigine Superposition, plus a sensor logger, such as GPU-Z or HWiNFO. Choose one from each group.

How long should I run a GPU stress test?

A few minutes reveals obvious faults, and a longer run of thirty minutes or more is a common choice for confirming stability. Stop immediately if temperatures approach the limit or the system becomes unstable.

What do GPU artifacts look like?

They appear as flashing dots, lines, blocks, checkerboard patterns, or distorted textures that are not part of the game. They often point to heat, an unstable overclock, a power problem, or failing memory.

Is FurMark safe for my graphics card?

Many cards handle it, but it creates an unusually heavy load that can push temperatures and power draw higher than games do. Watch the sensors, stop if temperatures climb too far, and consider a gentler benchmark if you have cooling concerns.