By Andrew Armstrong 
Bad RAM doesn’t announce itself. It produces crashes in unrelated programs, files that fail to open, Windows appearing corrupted. Symptoms that look like almost anything else. Which is why memory is worth testing early, before you spend money on theories.
Windows has a built-in tester. It’s better than nothing and considerably worse than the alternative.
Quick Facts
- Windows Memory Diagnostic is a quick screen, not a real test. It misses intermittent faults.
- MemTest86 is the standard. Bootable, so it tests memory without Windows in the way.
- Run at least four passes, ideally overnight. One pass proves very little.
- Any error at all means replace it. There’s no acceptable error count for RAM.
- Test sticks individually to identify which one is faulty.
Windows Memory Diagnostic. The quick check
# Launch it
mdsched.exeIt offers to restart now or on next boot, runs a test before Windows loads, and reports afterwards. If the result doesn’t appear, it’s in Event Viewer:
Get-WinEvent -FilterHashtable @{LogName='System'; ProviderName='Microsoft-Windows-MemoryDiagnostics-Results'} -MaxEvents 5 |
Select-Object TimeCreated, Message | Format-ListPress F1 during the test and choose the Extended test rather than Standard. The default is fast and shallow. Even then, treat a pass as “no obvious problem” rather than “memory is fine”.
MemTest86. The real test
Quick Steps
- Download MemTest86 and write it to a USB stick with the included tool.
- Boot from the USB. You may need to disable Secure Boot temporarily.
- Let it run. Four complete passes minimum. Several hours.
- Any error at all is a failure. Note which test number and which address.
- Re-enable Secure Boot afterwards if you disabled it.
The reason it’s better is simple: it runs without an operating system, so it can test the memory Windows would otherwise be occupying. A Windows-based tester can never check the regions it’s running in.
Finding which stick
An error tells you something is wrong, not what to replace. Isolate by elimination:
Quick Steps
- Note your current configuration. Which sticks in which slots.
- Remove all but one stick. Test it. Several passes.
- Repeat for each stick individually, in the same slot.
- If every stick passes alone but the set fails together, suspect the slot or the memory controller.
- To check slots: take a known-good stick and test it in each slot in turn.
That last check matters. A faulty slot presents exactly like a faulty stick, and people replace working RAM because of it.
Before you conclude the RAM is bad
- Turn off EXPO or XMP and retest. Memory running above its stable speed produces genuine errors from healthy sticks. This is the single most common false positive.
- Reseat everything. Remove and refit each stick firmly. Partially-seated RAM produces intermittent faults.
- Check the sticks are compatible with the board and each other. Mixed kits often can’t run at their rated speeds together.
- Check temperatures. Rare, but poorly-cooled memory in a hot case can throw errors under sustained load.
If disabling EXPO makes the errors disappear, the RAM isn’t faulty. It just can’t run at its advertised speed on your board with your CPU. That’s a real and common outcome, and running at a slightly lower speed is a perfectly reasonable resolution.
What bad RAM looks like in normal use
- Crashes in unrelated programs, with no consistent pattern.
- Blue screens with varying stop codes. Particularly MEMORY_MANAGEMENT and PAGE_FAULT_IN_NONPAGED_AREA.
- Files failing checksums, or archives reporting corruption on extraction.
- Windows appearing progressively damaged, with SFC finding errors repeatedly.
- Failures during large file copies or compilation.
That last group is a useful signal. Repeated SFC failures on an otherwise healthy machine often mean memory is corrupting data as it’s written. You’re fixing symptoms while the cause is still in a DIMM slot.
Glossary
| Term | What it means |
| MemTest86 | The standard bootable memory tester. Runs without an OS. |
| Pass | One complete run through all test patterns. Four minimum for confidence. |
| EXPO / XMP | Profiles running RAM at rated speed. A common cause of false errors. |
| DIMM | A memory module. A stick of RAM. |
| Memory controller | Part of the CPU that manages RAM. Faults here mimic bad sticks. |
| ECC | Error-correcting memory that detects and fixes single-bit errors. Server hardware. |
Frequently asked questions
Is Windows Memory Diagnostic good enough?
For a quick screen, yes. For ruling memory out, no. It’s less thorough than MemTest86 and can’t test the regions it occupies. Use it first, then MemTest86 before drawing conclusions.
How many MemTest86 passes do I need?
Four minimum, which usually means overnight. Intermittent faults can take hours to appear, and a single pass regularly misses them.
I got one error in eight hours. Is that acceptable?
No. There’s no acceptable error rate for memory. One error means one corrupted bit somewhere, which could be anything from a pixel to a saved file.
My RAM fails with XMP on but passes with it off. Is it faulty?
Not necessarily. It means it can’t run reliably at the advertised speed in your system. Run at a lower speed, or return it as not meeting spec if it’s new.
Can bad RAM damage anything else?
Not physically, but it corrupts data as it’s written. Files saved while faulty memory is in use may be silently damaged, which is why you should verify backups made during the affected period.
Gear We Recommend
Testing or replacing memory? Here’s what we’d buy.
Browse our General Tech Accessories picks on AmazonAs an Amazon Associate, TechyGeeksHome earns from qualifying purchases
Disclosure: this post may contain affiliate links. If you buy through one of them, we may earn a small commission at no extra cost to you. We only recommend products we’ve tested or genuinely rate.
Discover more from TechyGeeksHome
Subscribe to get the latest posts sent to your email.
