When a Blue Screen Means Replace the Part The Evidence Test
Two mistakes fight each other in blue screen repair: replacing parts on hunches (expensive, often wrong) and refusing to replace a part that has clearly failed (weeks of crashes). This page draws the line with evidence thresholds you can check yourself.
Watch the full method in the video - identifying a blue screen cause, step-by-step
Replacement is justified when a specific test fails, not when you run out of ideas. RAM that errors in MemTest86, a drive with growing SMART failures, a GPU that fails VRAM tests at stock clocks - those are verdicts. A PC that 'still crashes after I reinstalled everything' is not a verdict; it is an unfinished diagnosis. Below is the honest threshold for each major part, plus the order to test in so the cheapest evidence comes first. One dump never justifies a purchase. Patterns prove; failed tests convict.
The evidence ladder - buy only at the top step
Step 1 - Suspicion: a stop code or a gut feeling (never buy here)
Step 2 - Pattern: dumps repeatedly point one way
Step 3 - Isolation: the fault follows the part, slot or stick
Only step 3 plus step 4 together justify spending money.
The ladder every replacement decision should climb. Skipping steps is how good parts end up in the bin.
1
RAM - Replace When a Test or a Stick Swap Convicts It
Evidence threshold: any MemTest86 error at stock settings, or crashes that follow one specific stick across slots while the other stick runs clean for days.
Errors only at XMP but clean at base speed mean unstable profile, not dead RAM - run slower or exchange for a QVL kit; that is a settings outcome, not a failure
Errors at base speed in every slot, from one stick - replace that stick (kits under warranty go back as a kit)
No errors after 4+ passes and crashes unchanged with each stick alone - stop blaming RAM and move on
2
SSD or HDD - Replace on SMART Evidence, Not on Slowness
Evidence threshold: SMART status Caution/Bad, reallocated or pending sector counts that grow between checks, or the drive vanishing from BIOS intermittently.
Back up first - a drive at this stage can die during the argument
CRITICAL_PROCESS_DIED patterns plus bad SMART together are a conviction
A healthy-SMART drive with install crashes usually points elsewhere - cable, controller mode, or RAM corrupting what gets written
SSDs also die suddenly with perfect SMART history - if the PC cannot see the drive at all in two machines, that absence is itself the evidence
3
GPU and PSU - the Load Pair
GPU threshold: VRAM test errors at stock clocks, artifacts on screen, or stability that returns when the card is underclocked - an ageing card confessing. PSU threshold: crashes or hard restarts under peak load with no dump written, temporarily cured by a borrowed known-good supply.
Always clear the graphics driver first (clean install) - driver faults mimic dying cards daily
Kernel-Power Event 41 in Event Viewer after each incident supports the power story, though it records the restart, not the reason
Never buy a bigger PSU on suspicion alone - borrow, substitute, prove
4
When NOT to Replace Anything Yet
Keep troubleshooting instead when any of these are true.
All dumps name the same third-party driver - that is software until proven otherwise
Crashes began right after a specific update or install and you have not rolled it back yet
The PC has never been tested at stock - overclocks and XMP void every other conclusion
Only one or two crashes total, weeks apart - collect more dumps; the pattern has not spoken yet
A part 'failed' a test you ran once, quickly, while overclocked - retest properly at stock before spending
Keep your evidence when you do replace: screenshot the MemTest86 errors, the SMART report, the dump list. Shops and warranty desks approve claims with proof in minutes - and argue with stories for weeks.
Watch the Complete Video Guide
The full blue screen identification method - dump files, stop codes and the fix - shown live, step-by-step.
My PC crashes daily but every test passes. Replace something anyway?
No - passing RAM, healthy drive, clean drivers, stock settings, and continuing crashes means the net widens: motherboard, CPU (rare), power quality at the wall, or a device you have not disconnected yet. Substitution testing with borrowed parts beats buying; a good repair shop's bench time is cheaper than a wrong motherboard.
Is it worth replacing parts in a very old PC?
Sometimes - RAM and SSD upgrades are cheap and move to a future build only sometimes (DDR generations and form factors change). Be honest about the maths: if the motherboard or CPU is the convicted part on an old platform, the money usually belongs towards a newer platform instead.
The repair shop says replace the motherboard. How do I verify?
Ask what test convicted it. A fair shop will name the evidence: known-good CPU and RAM failing in this board, or this CPU and RAM passing in another board. 'We tried everything' without a named test is a guess wearing a bill - get the stock-settings and dump evidence first.
This guide is based on the Technical Anshu blue screen video - stuck on any step, ask in the video comments and you will get an answer.