Your SSD is at 97% "Healthy" and it's Dying: The SMART Data Lie (5 Hard Lessons)

1. Introduction

Imagine the ultimate workstation: an Intel i9-13900KS, 128GB of high-speed RAM, and a $2,000 RTX 4090. This is a machine built for absolute dominance, yet it is currently losing a battle against a single, microscopic piece of silicon. The system stutters, the OS hangs, and every attempt to clone the boot drive results in a cryptic failure and endless OS crashes.

The most maddening part of this nightmare, the part that nearly broke me, is that the diagnostic software reports the Samsung 980 Pro NVMe is at 97% health.

To a rational mind, 97% is an "A" grade. But in the world of high-performance storage, that number can be a mask for a walking corpse. As I spent hours troubleshooting the "Error 9" and "Error 0" loop during cloning the hard drive, I realised I wasn't just fighting a software bug. I was witnessing the final gasp of a dying drive that refused to admit it was over.

"The 97% that is showing above is a big trick, don't believe it... Media and Data Integrity Errors: My number was 75561. In NVMe, this number should be zero. Any number above zero means the physical cells (NAND) have started to decompose and are actually dying."

2. Takeaway 1: The "97% Health" Trap (Why SMART Data Lies)

In the storage industry, we rely on SMART data (Self-Monitoring, Analysis, and Reporting Technology) to tell us when to panic. However, the "Health Percentage" headline is often a distraction. In my case, the drive looked healthy because the Available Spare (ID 03) was reported as 47 in hex, which translates to roughly 71% of its reserve blocks still being available. Because the drive hadn't yet exhausted its physical fallback cells, the firmware calculated a "passing" health score.

The truth was buried in the raw hexadecimal values. The Media and Data Integrity Errors (ID 0E) showed a staggering 75,561 errors. Unlike the Health Percentage, which is a weighted average of spare blocks, ID 0E is a raw counter. There is no "A" grade here; anything above zero is a failing grade. When an NVMe reports even a single integrity error, the "Health Percentage" becomes a lie; the drive is no longer reliable for data storage.

"75,561 errors... This essentially means the drive is critically failing and doesn't have any healthy spare space left to move data to. Since we can't repair it this way, our best option might be to try cloning the drive again, but this time configuring Macrium to just skip any bad sectors."

3. Takeaway 2: The Samsung 980 Pro’s "Self-Destruct" Legacy

The Samsung 980 Pro is a powerhouse, but early production runs were haunted by a notorious firmware bug that caused the drive to prematurely exhaust its NAND cells. In this recovery scenario, the irony was thick: the drive reported only 48TB of Total Host Writes. On a drive rated for 1,200TBW (Total Bytes Written), this is a 4% usage rate. It should have been in its prime, yet it was clinically finished.

My drive was running version 5B2QGXA7, which was the firmware released to "fix" the self-destruction bug. Unfortunately, for many users, like me, this patch was a day late and a dollar short. If the NAND cells have already begun to physically decompose due to the bug's prior behaviour, the software patch cannot resurrect the dead silicon. It is a defensive necessity to keep Samsung Magician updated, but once those spare blocks are hit with 75,000 errors, the "fix" is merely a witness to the funeral.

"The Samsung 980 Pro... had a very famous 'Firmware Bug' that makes the drive destroy itself. I have the version that is supposed to 'fix' the problem, but it's clear the update came too late or after the drive was clinically finished."

4. Takeaway 3: Windows is "Polite," Linux is "Aggressive"

When silicon fails, Windows tools are often too "safe." Software like Macrium Reflect or the standard Windows Installer uses the Volume Shadow Copy Service (VSS) to create a stable snapshot of the drive. On a failing NVMe, VSS often "chokes" on bad sectors before the copy even begins. This results in Error 9, which is a physical Cyclic Redundancy Check (CRC) / Read I/O failure—the silicon equivalent of a brick wall.

To get the data off, I had to move from "Intelligent" methods to "Forensic" methods.

"Forensic Sector Copy... ignores the file system logic entirely and does a bit-by-bit transfer. This will take a long time because it is copying the empty space along with the data, but for a failing drive, it is the only way to bypass the logic that is currently triggering Error 9."

5. Takeaway 4: The WinPE Settings Gap (The Rescue Media Blindspot)

One of the most frustrating moments in data recovery is realising your software is ignoring you. I had checked "Ignore Bad Sectors" in my Macrium desktop app, yet the cloning process kept failing in the WinPE (Rescue Media) environment. This is because WinPE is a "sandboxed" environment; it does not inherit the configuration or registry keys of the host Windows installation.

I had to use a "dirty" hack to win this fight: the XML Override. Since the WinPE UI often hides advanced settings, you must save your backup as an XML Definition File, open Notepad from within the WinPE terminal, and manually find the <IgnoreBadSectors> tag. Changing that value from N to Y (or false to true) manually injects the "ignore" flag into the engine's execution logic, forcing it to bulldoze through the dead NAND cells.

"The Macrium instance running on your PC—most likely the WinPE Rescue Environment—has a stripped-down UI that completely locks out those advanced global defaults. The full Windows installation on your laptop exposes the complete feature set."

However, forcing the clone is only half the battle. If you find that the "Error 9" disappears but you still can't boot or install the OS, the problem might not even be the drive anymore.

6. Takeaway 5: When Your High-End Hardware Sabotages Itself

Sometimes, the I/O errors aren't the drive's fault at all. In high-end builds using an i9-13900KS and 128GB of RAM, the CPU's memory controller is under massive electrical strain. During a Windows installation, the WinPE environment extracts the massive payload entirely into a RAM disk (X:).

If you are running aggressive XMP or EXPO profiles, even microscopic memory instability can scramble the data during this decompression phase. This manifests as the dreaded Error 0x80042444. The system reports a read/write failure, but the "corruption" actually happened in the RAM before it ever touched the SSD. The counter-intuitive solution? Downclock that $600 CPU and high-end RAM to JEDEC base specs (4000MHz or 4800MHz) just to get the OS installed stably.

"Even slight instability during XMP/EXPO profiles will cause silent data corruption during the decompression phase, manifesting as this exact read/write error... Drop the memory frequency down to the JEDEC base spec."

7. Conclusion: Lessons from the Silicon Graveyard

The ultimate resolution for me wasn't a clever software fix; it was an RMA (Return Merchandise Authorisation). When a drive hits 75,000 integrity errors with only 48TB written, the silicon is physically exhausted. No amount of chkdsk or "Remapping" will make that drive safe for your data again. Sometimes, the most professional technical move is to admit the hardware has failed and walk away from the clone.

Every SSD, no matter how fast or expensive, is ultimately a collection of cells with a finite lifespan. If your drive reported 99% health today, but the raw integrity errors were climbing, would you trust it? Always look past the "Health" headline and check the raw hex values—the truth is usually hidden in the errors, not the percentages.

Your "Is My Drive Dying?" Checklist

• [ ] Check ID 0E: In CrystalDiskInfo, is "Media and Data Integrity Errors" above 0? If yes, back up immediately.
• [ ] Update Firmware: Check Samsung Magician or your manufacturer's tool for critical firmware patches.
• [ ] Verify WinPE Settings: If using rescue media, manually check XML/config files for IgnoreBadSectors flags.
• [ ] Stable Specs: If installing Windows on high-end hardware, drop RAM/CPU to JEDEC/Base speeds temporarily.
• [ ] RMA or Bin: Don't trust "repaired" silicon with 70,000+ integrity errors.

"No software (Macrium, Victoria, etc.) could 'fix' the dead silicon. The successful outcome... was the definitive hardware diagnosis that prevents further time wasted on a doomed migration."

Some of the Windows OS crashes I was getting