How to stress test a PC, and what the results mean

When a stress test is worth running

• After building or upgrading a PC, or fitting a new cooler.
• After switching on a memory profile (XMP or EXPO), overclocking or undervolting.
• When games crash, or the PC freezes or restarts by itself.
• When performance falls away after a few minutes of heavy work.
• Before selling a PC, or just after buying a used one.

Before you start

• Save your work and close other programs. The PC will be slow and its fans loud while the test runs.
• Plug a laptop in; on battery it runs slower and the results mean less.
• Stay with the PC the first time, and know how to stop the test. Every test has a stop button, and if the PC stops responding, holding the power button for a few seconds turns it off.

A stress test does not harm a healthy PC. Processors and graphics cards protect themselves from heat by slowing down, and switch off before any damage if that is not enough. What a stress test can do is bring out a fault that was already there — which is the point.

What each kind of test finds

The processor

Heavy calculation on every core. Tools that check their own answers, such as Prime95 and OCCT, find instability as well as heat: a wrong answer from a processor is never normal. Cinebench loads the processor just as hard but does not check its answers, so it finds heat problems only.

Memory

Patterns written to memory and read back. MemTest86 runs from a USB stick before Windows starts and can test nearly all of the memory; Windows Memory Diagnostic (search for it in the Start menu) is a simpler built-in version; OCCT and TestMem5 test from inside Windows. After switching on XMP or EXPO, memory errors are the most common cause of instability.

The graphics card

3D rendering at full load. 3DMark’s stress tests loop a scene twenty times and check that the frame rate holds steady; OCCT’s 3D test looks for errors; FurMark loads a card harder than any game, which makes it a harsh test of cooling. The games you play are a fair test too.

The power supply

Nothing tests a power supply directly, but loading the processor and graphics card together comes closest. A PC that restarts only under that combined load, and passes each part on its own, often has a power supply or cable problem.

How long to run it

• 15 to 30 minutes shows the heat picture. Temperatures climb for the first several minutes as the heatsink and the air in the case warm up, then level off.
• An hour or more after changing memory or overclock settings, because rare errors take time to appear. Memory testers need several full passes, which can take hours.
• Stop at the first error. One is enough to prove a problem, and running on proves nothing more.

What temperatures are normal?

Modern processors are designed to run hot. Recent AMD Ryzen chips hold themselves at up to 95 °C under full load, and recent Intel Core chips at up to 100 °C, boosting as high as their cooling allows. Reaching that temperature is not a fault in itself, and laptops often sit at 90 °C or more under full load by design.

What matters is whether the processor slows down, and why. Watch its speed in Task Manager (Performance → CPU → Speed) during the test:

• A drop in the first minute or two, which then holds steady, is usually the processor settling from its short-term power limit to its long-term one. That is by design.
• A speed that keeps falling for many minutes as the temperature climbs is heat. So is a speed that ends well below the processor’s base clock.

Graphics cards usually run between about 60 and 85 °C under a long load, and each has a maximum temperature its maker publishes. Some tools also show a hotspot or memory temperature; those always read higher than the main figure, and that is normal.

Signs of instability

• The test reports an error, or stops a worker — Prime95 stops the worker and says a hardware failure was detected.
• The PC freezes, shows a blue screen, or restarts.
• Windows logged a hardware error. Open Event Viewer → Windows Logs → System and look for: WHEA-Logger events, which record hardware errors, including corrected ones that point to a PC on the edge of stability; event 4101, the graphics driver stopping and recovering; and Kernel-Power 41, the PC restarting without shutting down properly.

If it fails

Errors or crashes

• After switching on XMP or EXPO: update the motherboard firmware first, because memory compatibility improves with updates, then try the next speed down. Memory speed explains the settings.
• With an overclock or undervolt: step it back until the test passes.
• At default settings: a part is faulty. Test the memory one stick at a time with MemTest86 first, since it is the most common culprit and the easiest to prove.
• Restarts under combined load: check the power supply is rated for the system, and give the graphics card a separate cable from the power supply for each of its power sockets rather than one split cable.

Heat

• Clean the dust from filters, fans and heatsinks, and check every fan actually spins.
• Check the fan settings in the firmware; a silent profile can let the processor reach its limit early.
• If the cooler is several years old, refit it with fresh thermal paste.
• Improve the case’s airflow: air in at the front, out at the back and top.
• For a laptop, clean its vents and use it on a hard, flat surface.

Bottleneck’s heat and stability check

Bottleneck has a heat and stability check built in. It runs the processor and memory at full load — and the graphics card, if you include it — for 5 minutes to an hour. Every processor thread runs calculations whose answers are known, while memory is written and read back, so a single wrong answer proves the PC unstable, and the check stops there. Nothing is overclocked or changed.

All the while it watches whether heat slows anything down: the processor’s speed and work rate over the run, Windows’ own cooling limits, and the graphics card’s temperature. It tells a power limit that settles in the first minutes from heat that builds over the run. Afterwards it reads the hardware errors Windows logged, graphics driver resets and unexpected restarts. It stops if the graphics card gets too hot to continue, and if the PC restarts partway through, it reports what happened the next time it opens.

The verdict is free: passed, runs hot, or unstable. Pro adds the full detail, and Technician adds a certificate to hand to a customer.