Your PC crashes at random. A blue screen appears out of nowhere, you reboot, and everything works fine again for a week. You blame Windows, then the graphics driver, then the SSD. You reinstall the system twice. You update the BIOS. And the crashes keep coming, just unpredictable enough that you start wondering if you're imagining them.
I've been down that road. Twice. The second time, it was a single faulty RAM stick that took me eleven days to identify, mostly because I kept testing the wrong component. RAM is one of the easiest parts to diagnose and one of the hardest to suspect, because when it fails, it rarely fails loudly. It just corrupts a bit here, a bit there, and lets the rest of your system take the blame.
Key Takeaways
- Random blue screens, corrupted files, and failed installs are the three most common symptoms of faulty RAM — but each has other possible causes, so testing matters more than guessing.
- Windows Memory Diagnostic (mdsched.exe) is fast and built in, but it misses intermittent faults that MemTest86 catches on the third or fourth pass.
- Test one stick at a time, in different slots. "The RAM is bad" is almost always wrong — usually it's one stick, one slot, or one XMP profile.
- XMP and EXPO memory profiles can cause errors that look like hardware failure but disappear at default speeds.
- A faulty module can pass every test at idle and fail under load. If your tests come back clean but the crashes continue, run MemTest86 overnight, not for ten minutes.
Signs your RAM is failing: what actually shows up on your screen
Most people expect RAM failure to look like a dead stick — the machine won't POST, black screen, beep codes. That happens. But it's the least interesting case, because it's obvious. The frustrating failures are the ones that leave you running a working computer that misbehaves.
Here's what I've seen over the years, across my own builds and the machines friends bring me when they've given up:
What are the symptoms of faulty RAM?
Blue screens with memory-related stop codes (MEMORY_MANAGEMENT, PAGE_FAULT_IN_NONPAGED_AREA, IRQL_NOT_LESS_OR_EQUAL) are the classic sign. Random freezes that force a hard reset. Applications that close with no error message. Files that open as garbage or refuse to open at all. And — my personal favorite — installations that fail at 74% with a meaningless error code, every single time, no matter which version of the installer you download.
That last one cost me a full afternoon once. I assumed the ISO was corrupted. It was a bad stick in slot 2.
One symptom people rarely mention: successful installs that produce a system which then installs fine, boots fine, and corrupts a working file three weeks later. That's the insidious version. A single bit flips in the wrong place, and you lose a document you spent two days on.
What does corrupted RAM look like?
Not like sparks and smoke. It looks like a Windows profile that loads with missing desktop icons. A ZIP archive that fails its CRC check on extraction. A photo that displays perfectly in the thumbnail and won't open at full size. A game that crashes to desktop during a load screen, always at the same spot, and then works fine the next day.
If you're seeing a pattern of data-level problems — files that don't match what you saved, checksums that fail, browser tabs that crash on load — the memory subsystem is a serious suspect. Not the only one, but a serious one.
How to run a RAM diagnostic test properly (not the version that wastes your evening)
There are two tools worth your time on Windows. One is already installed. The other is what you actually need.
Windows Memory Diagnostic: fast, free, and not always enough
Press Win + R, type mdsched.exe, hit Enter. You'll get a choice between restarting now and scheduling a test at next boot. Pick the restart. The test runs in a plain blue screen before Windows loads, and takes between 15 and 40 minutes depending on how much RAM you have and which pass setting you choose.
What most people don't realize: the default is two passes of the basic test. That's a smoke test. It catches a stick that's completely dead. It will absolutely miss a stick that fails once every few billion operations — which is exactly the kind of fault that causes your random crashes.
When the test ends, Windows reboots and then tells you nothing useful unless you go hunting in Event Viewer. Look for MemoryDiagnostics-Results under Windows Logs → System. No errors reported doesn't mean the RAM is fine. It means the RAM didn't fail in those two passes.
MemTest86: the standard, and why the number of passes matters
MemTest86 runs from a USB stick outside the operating system, which is the point — it can hammer the memory without Windows in the way. You'll find it at memtest86.com. The free version is enough. Write it to a USB key, boot from it, and let it run.
Here's the part the quick guides skip. One pass proves almost nothing. A stick with intermittent faults can pass once and fail on pass five. My rule, after getting burned: minimum four full passes, which for 32 GB of DDR5 takes several hours. If you have the patience and the crashes are frequent, run it overnight and let it do eight or ten.
Read the results carefully. MemTest86 reports error counts, addresses, and the specific test that failed. Write them down. If the same address range errors repeatedly across passes, you've found your culprit. If errors appear in different addresses each pass, the fault may be in the memory controller or the slot rather than the stick itself.
How can I test my RAM?
The answer depends on whether you're testing to diagnose a fault or to confirm a suspicion.
- If you have symptoms, skip the built-in tool and go straight to MemTest86. Four passes minimum.
- If MemTest86 shows errors, stop. Don't run more tests. Move to isolation, which I'll cover below.
- If MemTest86 shows no errors after four passes but your crashes continue, test the memory at default speeds with XMP or EXPO disabled, then test again mid-load using a stress tool like Prime95 while monitoring for WHEA errors in Event Viewer.
- If everything passes and the machine still misbehaves, RAM is probably not your problem. Look at the power supply, the storage, and the motherboard next — in that order.
Isolating the faulty stick (this is where most people get it wrong)
Saying "my RAM is dead" is like saying "my car is broken." It doesn't help. You need to know which stick, in which slot, under which profile. That is what actually fixes the problem.
Here's the process, and I'll be direct: it's tedious, but it's the only reliable way.
| Test step | What to do | What it tells you |
|---|---|---|
| 1. Baseline | Run MemTest86 with all sticks installed, XMP on | Confirms errors exist, but not where |
| 2. Single stick, slot A | Remove all but one stick, retest in the primary slot | Isolates whether a specific stick is faulty |
| 3. Same stick, slot B | Move it to a different slot, retest | Distinguishes bad stick from bad slot |
| 4. Repeat per stick | Do this for each module, one at a time | Identifies all faulty modules, not just the first |
| 5. All sticks, XMP off | Reinstall everything, disable XMP/EXPO, retest | Tells you if instability is a profile problem, not hardware |
Step 5 is the one that surprised me most. I had a kit that failed four MemTest86 passes at its rated XMP speed and ran flawlessly for a full night at JEDEC defaults. That's not a dead module — that's a memory kit the motherboard doesn't want to run the way it was sold. The fix isn't a return; it's a BIOS update or a slightly relaxed timing.
A word on laptops: the process is identical if your laptop has removable SODIMMs, but many thin models solder memory to the board. If that's your machine, you can't isolate anything by swapping. Your options are running MemTest86 from USB (still works) and, if errors appear, using your warranty. Don't open a sealed laptop to "check the RAM" unless you're comfortable losing the warranty over it.
Common questions that come up after the tests
How do I know it's RAM and not the SSD or CPU?
Look at the shape of the failures. RAM faults are random and address-dependent — different applications crash, different files corrupt, and rebooting sometimes clears the problem for hours. A failing SSD tends to produce errors tied to specific files or read/write operations. A CPU issue usually shows up under sustained load, not at idle.
If you want to be sure, run MemTest86 and a disk health check back to back. One of them will tell you something conclusive.
How do I check RAM without another computer?
You don't need one — MemTest86 boots from USB on the machine you're testing. What you can't do is test from inside Windows and expect reliable results for an intermittent fault. The operating system is using the memory while it tests it, which masks exactly the kind of errors you're looking for. That's the whole reason MemTest86 exists as a boot environment.
What to do once you know
If a stick fails, and it's under warranty — and most kits sold in the last few years carry a lifetime warranty — the process is straightforward. Photograph the MemTest86 error screen before you pull anything. RMA departments ask for it, and being able to say "errors at addresses 0x1A2F0000 and 0x1A3B0000 on pass 3" moves the case along faster than "it crashes sometimes."
If nothing failed and the crashes continue, you've eliminated a major suspect and can move on with a clear conscience. That's worth the evening, even when the answer is "not here."
The thing I keep coming back to, after all this: RAM is boring until it isn't. It sits there quietly for years, does exactly what it was sold to do, and then one day flips a bit in a file you care about. The tests above are dull. Running them is the least interesting way to spend four hours. But the alternative — guessing, replacing parts, reinstalling Windows a third time — costs you far more than that.
And if your tests pass and the machine still crashes? Save the next crash dump. That file will tell you more in five minutes than a week of swapping components. You just have to know where to look.
