How to run a GPU stress test safely

A stress test deliberately holds your GPU at or near full load. That is useful: it reveals weak cooling, unstable overclocks and failing hardware before a game or project does. It also raises temperature and power draw, so a little preparation matters.

Before you start

• Close games, video editors and extra browser tabs so only the test is using the GPU.
• On a laptop, plug in the charger and choose a high-performance power mode. Make sure vents are not blocked by a soft surface.
• Open a monitoring tool such as GPU-Z or HWiNFO so you can see temperature and clocks.
• Remove or reset any aggressive overclock if you are diagnosing crashes.

Choosing duration

Thirty seconds is enough to check that everything works. Sixty to one hundred and twenty seconds gives stable averages. For thermal behaviour you need at least three to five minutes, because heat builds gradually; the Thermal Throttle Test on this site runs five minutes for that reason.

Choosing intensity

Start at Medium. If your frame rate sits exactly at your monitor’s refresh rate the result is capped and tells you little about the GPU’s ceiling, so step up to High, then Extreme. On phones and thin laptops, Low and Medium are usually the right range.

When to stop

Press Stop, or Escape, if the screen glitches, fans reach a speed that alarms you, the system freezes or the temperature climbs steadily past what the manufacturer rates for your card. Most modern GPUs throttle to protect themselves, but a test should never be left running unattended.

After the run

Read the SMG Verdict. A weak Endurance score points to heat; a weak Spike control score points to background load, drivers or power settings. The advice section lists specific next steps.