I’ve lost count of how many times a client has sworn their RAM was dead, only for me to find a nvlddmkm.sys timeout or a buggy audio driver causing DRIVER_IRQL_NOT_LESS_OR_EQUAL. The symptoms look identical on the surface: sudden reboots, freezes, the occasional Blue Screen of Death (BSOD). But the fix? Worlds apart. One costs $0 and a driver update; the other costs a DIMM swap and an RMA.

Here is the exact workflow I use on the bench to separate silicon failure from code rot.

The Cheat Sheet: RAM vs. Driver Symptoms

Don’t guess. Compare what you’re seeing against this table. If you check three or more boxes in the “Likely RAM” column, stop updating drivers and start testing hardware.

Symptom / Behavior	Likely Bad RAM / Hardware	Likely Bad Driver / Software
BSOD Error Codes	`MEMORY_MANAGEMENT` (0x1A), `PAGE_FAULT_IN_NONPAGED_AREA` (0x50), random varying codes	`SYSTEM_THREAD_EXCEPTION_NOT_HANDLED` (0x3B), `DRIVER_IRQL_NOT_LESS_OR_EQUAL` (0xD1), `KMODE_EXCEPTION_NOT_HANDLED` (0x1E)
Crash Timing	Random. Idle, light browsing, heavy load—no pattern. Often worse when warm.	Deterministic. Happens only when printing, gaming, waking from sleep, or plugging in USB.
Application Behavior	Silent data corruption (files won’t open, zip CRC errors), Chrome “Aw Snap” pages constantly.	Specific app crashes (Explorer.exe restarts, game crashes to desktop, video driver resets).
Boot Behavior	POST codes beeping, long memory training times, fails to boot with XMP/EXPO on.	Boots fine into Safe Mode. Crashes only after specific driver loads (usually GPU/Audio/Network).
Event Viewer (System Log)	`WHEA-Logger` events (Corrected/Uncorrected Hardware Errors), `Memory` events.	`Event ID 1001` (Bugcheck) pointing to a specific `.sys` file (e.g., `rtwlane.sys`, `nvlddmkm.sys`).
MemTest86 Result	Errors appear immediately or within first 2 passes. Row hammer errors.	Zero errors after 4+ passes (standard).
Response to Reseat	Fixes it temporarily (oxidation), or changes error pattern.	No effect.

Phase 1: The “Zero Cost” Physical Checks

Before you run a single software tool, kill the power. Pull the PSU cord. Hold the power button for 15 seconds to drain flea power.

Reseat the DIMMs. Pull them out. Blow the slots out with compressed air (hold the can upright). Look at the gold contacts—see any oxidation or “burn” marks? Clean contacts with a pencil eraser gently, then 99% IPA. Reseat firmly until the retention clips click.
Test Single Sticks. If you have 2x16GB or 4x8GB, test one stick at a time in the primary slot (usually A2, second from CPU). This isolates a bad stick vs. a bad slot vs. the IMC (Integrated Memory Controller).
Disable XMP/EXPO. Reset BIOS to Optimized Defaults. Run at JEDEC speeds (4800/5600 MT/s). If stability returns, your memory controller (CPU) or the RAM itself can’t handle the rated overclock. That’s a hardware limitation, not a driver bug.

Pro Tip: On Ryzen 7000/9000 and Intel 12th/13th/14th gen, the memory controller is on the CPU. If single sticks pass MemTest but dual channel fails at XMP, your CPU IMC might be degrading. I’ve seen this on i9-13900K/14900K a lot lately.

Phase 2: Software Triage (Windows Built-ins)

1. Windows Memory Diagnostic (Quick Sanity Check)

Hit Win + R, type mdsched.exe, choose “Restart now and check for problems.”

Use case: Catches catastrophic failures (stuck bits, dead rows) in 5 minutes.
Limitation: It runs in a limited environment, doesn’t test all addressing modes, and misses subtle timing errors. A “Pass” here means nothing. A “Fail” means stop—hardware is dead.

2. Driver Verifier (The “Stress Test” for Code)

This is the nuclear option for driver hunting. It forces drivers to crash immediately when they do something illegal, rather than corrupting memory silently for hours.

:: Run as Administrator
verifier /standard /all

Reboot.
If you boot loop: Boot into Safe Mode (Shift + Restart -> Troubleshoot -> Advanced -> Startup Settings -> 4) and run verifier /reset.
If you get a BSOD: Check the bugcheck. It will name the .sys file. That is your culprit. Update or uninstall that specific driver.

Phase 3: The Gold Standard — MemTest86 (Free Edition)

If Windows Memory Diagnostic passes but you still crash, you must run MemTest86 from a USB stick. It runs outside the OS, testing every addressing mode, cache state, and hammer pattern.

Download the Free Edition USB image from PassMark.
Flash to USB (Rufus / BalenaEtcher).
Boot from USB. Select Config -> Test Selection -> Passes: 4.
Hit S to start. Walk away. Grab coffee. This takes 2–6 hours depending on capacity.

Interpreting Results:

Zero Errors (4 Passes): Your RAM is physically sound. Stop swapping sticks. Look at drivers, thermals, or PSU.
Errors on Pass 1, Test 0-4 (Address/Walking bits): Hard failure. RMA the stick.
Errors only on Test 13 (Row Hammer) or Test 9 (Bit Fade): Marginal stability. Try loosening timings (tRFC, tREFI) or dropping voltage/clock speed.
Errors move slots when you move the stick: Bad Stick.
Errors stay on same slot regardless of stick: Bad Motherboard Slot / CPU IMC.

Phase 4: The AI Diagnostic Assistant (Copy-Paste This)

Don’t stare at Event Viewer guessing. Dump the raw data into an LLM (ChatGPT, Claude, DeepSeek, local Llama) with this specific prompt. It parses minidump analysis and Event Viewer XML better than most humans because it has the entire Windows kernel symbol database in its weights.

Step 1: Grab the Data

Option A: Minidump (Best)

Win + R -> sysdm.cpl -> Advanced -> Startup and Recovery -> Settings.
Ensure “Write debugging information” is Small memory dump (256 KB) or Automatic memory dump.
Dumps live in (image upload pending) Zip the last 3–5 .dmp` files.

Option B: Event Viewer (If no dumps)

Win + X -> Event Viewer -> Windows Logs -> System.
Filter Current Log -> Event Sources: BugCheck, WHEA-Logger, Memory, Kernel-Power.
Select the last 20 events -> Right Click -> Save Selected Events As -> SystemCrashes.evtx.

Step 2: The Prompt

Copy this block, paste into your AI, attach the .zip (dumps) or .evtx (logs).

SYSTEM CRASH ANALYSIS PROMPT

Act as a Senior Windows Kernel Debugger. Analyze the attached crash dumps / Event Viewer export.

Context:

OS: Windows 10/11 (Version: winver output here)

Hardware: CPU / Motherboard / RAM Kit (Part #) / GPU / PSU (Wattage)

XMP/EXPO Status: Enabled / Disabled / Manual Timings

Recent Changes: (Driver updates, hardware swaps, Windows updates)

Task:

Identify the BugCheck Code and Parameter 1/2/3/4 meaning for the last 3 crashes.

Determine the Faulting Module (.sys file or ntoskrnl.exe offset).

Classify root cause probability: Hardware (RAM/CPU/Mobo/PSU) vs Software (Driver/OS Corruption/AV) vs Configuration (XMP/Overclock/Thermals).

If Hardware: Pinpoint specific component (e.g., “Bit flip at physical address 0xXXXXXXXX implies DIMM in Slot A2” or “WHEA 0x124 Cache Hierarchy Error implies CPU L2/L3”).

If Software: Provide exact driver name, vendor, and version to update/rollback. Link to official download if possible.

Provide a Prioritized Action Plan (Next 3 steps I should take physically or in software).

Output Format: Markdown table for Crash Summary, then Bullet points for Root Cause & Action Plan.

Why this works

The model sees BUGCHECK_STR: 0x1A and Arg1: 0x41790 and knows instantly that 0x41790 maps to PAGE_TABLE_CORRUPTION usually caused by a page table entry (PTE) bit flip—classic RAM/CPU cache—whereas Arg1: 0x41201 is PFN_LIST_CORRUPT, often a driver double-freeing memory. It saves you hours of !analyze -v in WinDbg.

Phase 5: The “RAM is Actually Dead” Replacement Checklist

You ran 4 passes of MemTest86. It threw errors. You swapped slots, errors followed the stick. It’s dead. Before you order replacement:

Check	Why It Matters
Verify QVL (Qualified Vendor List)	Go to motherboard manufacturer site. Search your board -> Memory QVL. Filter for your exact CPU generation. Buying “fast” RAM not on the QVL is the #1 cause of “bad RAM” RMA loops.
Match Die Revision (If adding to existing)	If you have 2x16GB and buy another 2x16GB, they must be same IC (Samsung B-die, Hynix A-die, Micron Rev.E). Mixing dies kills stability. Check Thaiphoon Burner on existing sticks.
Check CPU Cooler Mounting Pressure	Critical for LGA1700 / AM5. Over-tightening the cooler bends the PCB, lifting pads off the CPU substrate or cracking the IMC. Torque to spec (usually 0.5–0.6 Nm / ~5-6 in-lbs) in a cross pattern.
PSU Cable Check	Are you using two separate PCIe cables for the GPU (not daisy-chained)? Is the EPS 8-pin (CPU power) fully seated? Voltage droop on VDDQ/VDD rails looks exactly like bad RAM.
BIOS Version	Update to latest before installing new RAM. AGESA / Microcode updates fix memory training algorithms.
RMA Process	Most major brands (Corsair, G.Skill, Crucial, Kingston) offer advanced RMA (cross-ship). Pay the hold fee, get replacement in 2 days, return dead kit. Don’t wait 3 weeks for standard RMA.

Summary: The Decision Matrix

Crashes are random, MemTest86 fails -> Hardware. RMA RAM / Check CPU IMC / Check PSU.
Crashes are deterministic (specific game/app), MemTest86 passes 4 passes -> Driver / Software. Run Driver Verifier / Update specific .sys driver / DDU GPU drivers.
Crashes only with XMP on, MemTest86 passes at JEDEC -> Configuration. CPU IMC lottery loss. Loosen timings, drop frequency, or increase VDDQ/SA voltage slightly (Intel) / VDDIO/CLDO_VDDP (AMD).
Crashes leave WHEA_UNCORRECTABLE_ERROR (0x124) or CLOCK_WATCHDOG_TIMEOUT (0x101) -> CPU / Cache / Voltage. Not RAM. Check thermals, voltage, degradation.

Stop guessing. Run the USB stick. Read the dump. Fix the right thing.

PC Keeps Crashing? How to Tell if It’s a RAM Issue or a Bad Driver