Sudden reboots, blue screens, programs closing for no reason, files that turn out corrupted after a copy: these symptoms can have software causes, but very often they hide a hardware problem. Faulty memory, an overheating processor, a power supply at its limit. The way out of guesswork is to put the system through targeted tests, one component at a time, watching what gives way and when.
When to test
- After building a PC or changing RAM, processor or graphics card.
- After enabling memory profiles such as XMP or EXPO or changing frequencies and voltages.
- When instability appears with no obvious cause.
- Before trusting a computer with important work, such as a home server or a machine that processes data for hours.
Basic rule: one component at a time
Running every test at once doesn't help: if the system freezes, you don't know who's to blame. Proceed step by step, starting with memory, the most elusive cause, then the processor, and finally the graphics card and the system as a whole. During all tests, keep a monitoring program open that logs temperatures, frequencies and, if available, voltages.
Testing memory
Faulty RAM is insidious because it doesn't always cause crashes: sometimes it silently corrupts data. The most widely used tools are MemTest86 and Memtest86+: they boot from a USB stick, outside the operating system, and write and read back data patterns across the whole memory.
Practical guidance:
- let it complete at least four full passes; for a thorough check, a whole night;
- even a single error is a failed result: healthy memory produces none;
- if errors appear, test one module at a time and, if possible, in different slots, to tell a faulty module from a slot or overly aggressive settings;
- if the errors disappear when you disable XMP or EXPO, the problem is the configuration, not necessarily the modules.
The memory diagnostic tool included in Windows is a simple alternative, but less thorough.
Testing the processor
Processor stress tests push it to maximum load for long periods. Among the best-known tools are Prime95, OCCT and, on Linux, stress-ng. Some workloads, such as those using advanced vector instructions, produce more heat than any real-world use: that's exactly why they're useful, but they need watching.
What to look at:
- Calculation errors: some tests compare results with expected values. An error indicates instability, even if the system doesn't freeze.
- Temperatures: every processor has a maximum temperature stated by the manufacturer. Approaching that limit triggers automatic frequency reduction, so-called thermal throttling.
- Frequencies: if the frequency collapses during the test, the processor is protecting itself from heat or from its power limit.
A run of half an hour to an hour reveals most cooling problems. To validate a modified configuration, longer tests are needed.
Testing the graphics card
Graphics benchmark programs and specific load tests put the GPU under strain. The signs to look for differ from those of the processor:
- artefacts: coloured dots, misplaced triangles, flickering. They may indicate faulty video memory or excessive temperatures;
- a driver crash with a black screen for a few seconds;
- a sudden shutdown of the whole PC under graphics load: the culprit is often the power supply, which can't cope with current spikes.
The combined test
Only after checking the individual components does a simultaneous load on processor and graphics card make sense. It's the toughest test for the power supply and for airflow in the case, because it produces maximum consumption and maximum heat together. A PC that passes the individual tests but shuts down in the combined one almost always has a power or internal temperature problem.
Reading the results
| Symptom | Likely cause |
|---|---|
| Errors in MemTest86 | faulty RAM, slot, overly aggressive memory profile |
| Calculation errors in the processor stress test | frequency or voltage instability, sometimes memory |
| Temperatures at the limit and falling frequencies | inadequate cooling, thermal paste, badly mounted cooler |
| Graphics artefacts | video memory, GPU temperatures |
| Total shutdown under combined load | insufficient or faulty power supply |
Caution during testing
Stress tests are meant to be tough, not destructive, but they need supervision. If temperatures keep climbing without levelling off, stop the test and look into the cooling rather than waiting for a shutdown. On laptops, very long loads make little sense: it's enough to check that the system copes with realistic heavy use.
In my work I always record temperature values and results before and after an intervention: a written comparison is worth more than any impression.
In short
- Test one component at a time: RAM first, then processor, graphics card and finally the combined load.
- Even a single memory error is a failed result.
- During stress tests, watch errors, temperatures and frequencies, not just crashes.
- Shutdowns under combined load often point to the power supply.
- Supervise the tests and record the results so you can compare them.
