What Causes Samsung 870 EVO Failures?
The Samsung 870 EVO uses Samsung's MKX controller paired with Samsung V-NAND (3D TLC). Certain production runs exhibit premature NAND cell degradation, where the TLC cells lose charge retention faster than the controller's ECC algorithms can compensate. The drive accumulates uncorrectable read errors until the controller can no longer serve data reliably.
TLC NAND stores 3 bits per cell by distinguishing between 8 voltage levels. As cells degrade through program/erase cycles or manufacturing variability, the voltage margins between those 8 levels narrow. The controller uses Low-Density Parity-Check (LDPC) error correction to compensate, but LDPC has a correction ceiling. When a cell's voltage drift exceeds the LDPC correction capacity, the read fails as an uncorrectable error.
On affected 870 EVO units, this degradation happens well before the drive reaches its rated endurance (TBW). Users have reported failures at a fraction of the rated write endurance, indicating a manufacturing or firmware interaction anomaly rather than normal wear. Higher-capacity models (2TB and 4TB), which use denser NAND stacking, appear disproportionately represented in failure reports.
Symptoms of Samsung 870 EVO NAND Degradation
The failure progresses through stages. Early symptoms are subtle and easy to dismiss. By the time the drive shows 0 bytes, significant NAND degradation has accumulated.
- Stage 1
Intermittent read slowdowns
File copies stall or take longer than expected. Applications freeze briefly when loading files from the drive. The controller is spending extra cycles on ECC correction for marginal cells. S.M.A.R.T. health may still report 90%+ at this stage.
- Stage 2
Uncorrectable read errors in system logs
Windows Event Viewer shows "The IO operation at logical block address was retried" or "Reset to device was issued" errors. macOS Console logs show
IOAHCIBlockStoragetimeout errors. Specific files become unreadable or return corrupted data. The controller is hitting cells beyond LDPC correction capacity. - Stage 3
System freezes and boot failures
If the 870 EVO is the boot drive, Windows hangs during startup or enters a repair loop. The controller is failing reads on critical system files (NTFS metadata, registry hives, boot configuration). The system appears "frozen" because the OS is waiting on reads that the controller cannot complete.
- Stage 4
0 bytes capacity or drive not detected
The controller enters a protective state after exhausting its bad block remapping capacity. The drive may report 0 bytes in Disk Management, show an incorrect model string, or disappear from BIOS entirely. The firmware cannot maintain a coherent FTL mapping with the number of failed blocks.
Why S.M.A.R.T. Does Not Predict This Failure
Samsung's S.M.A.R.T. implementation normalizes raw error counts into a percentage that can mask thousands of reallocated sectors behind a "99% healthy" rating. CrystalDiskInfo and Samsung Magician both read these normalized values, not the raw counts.
S.M.A.R.T. Attribute 5 (Reallocated Sector Count) and Attribute 177 (Wear Leveling Count) are the relevant fields. Samsung's normalized health score is calculated from these attributes using a proprietary algorithm. The normalization compresses a wide range of raw values into a narrow percentage band. A drive with 500 reallocated sectors and one with 5,000 can both report "99% health" depending on how Samsung's firmware weights the count against the drive's total capacity and spare block pool.
This means the standard consumer monitoring tools provide a false sense of security. By the time Samsung Magician drops below 90%, the drive may already be experiencing read failures. For the affected 2021 batches, the S.M.A.R.T. health cliff is steeper than normal: drives go from "healthy" to catastrophic failure with minimal warning.
What Happens to an 870 EVO During a Sudden Power Loss?
A sudden power loss during a write operation can corrupt the 870 EVO's internal mapping tables, making data inaccessible even though the NAND cells still hold it. The MKX controller maintains these tables in DRAM & flushes them to NAND periodically. Power loss mid-flush corrupts the mapping metadata, and the controller can no longer locate your files.
The 870 EVO's TLC NAND stores 3 bits per cell using 8 voltage levels. During a write, cells transition through intermediate states before settling at their target level. Lose power mid-write, and some cells land between valid voltage states. The controller flags these as uncorrectable errors on the next read attempt, even though the surrounding cells are healthy.
Power loss also damages the Flash Translation Layer (FTL). The FTL is the index that maps logical block addresses (what your OS sees) to physical NAND locations (where the data actually sits). The MKX controller keeps the working FTL copy in volatile DRAM for speed. If power drops while the controller is writing FTL updates back to NAND, the on-disk FTL copy becomes inconsistent. The controller boots, finds a corrupted FTL, and reports the drive as 0 bytes or wrong capacity.
Why 870 EVO FTL Corruption Is Harder Than Older Samsung SATA
On older Samsung SATA generations (MGX/MHX on 840/850/860), PC-3000 SSD's Samsung utility could enter technological mode and scan NAND page metadata to rebuild the logical-to-physical map directly. Samsung locked that access path starting with this generation, so on an 870 EVO (MKX / Metis, S4LR059) there is no supported firmware-level FTL reconstruction; the controller is effectively a closed box at the firmware layer.
What remains is normal-mode imaging. If the controller can be stabilized (board-level repair of the power rails or PMIC, replacing any shorted passive component, restoring clean supply to the controller so it boots into its own normal mode) and if enough of the drive's own internal FTL is still coherent in the NAND service area, the drive may re-enumerate long enough to be imaged over SATA at the logical layer. That image is then processed on a separate workstation. FTL damage that the controller itself cannot reconcile on boot is currently not recoverable on the 870 EVO generation through any published PC-3000 path. Free evaluation confirms whether a given drive falls into the recoverable or non-recoverable bucket ($600–$900 firmware-tier on successful stabilization).
How We Recover Data from Degraded Samsung V-NAND
Samsung locked firmware-level technological-mode access on the 870 EVO generation. The published PC-3000 SSD recovery path for the MKX is therefore normal-mode imaging after hardware stabilization, not loader-based firmware rewrites. The goal is to get the original controller booting cleanly so its own LDPC engine can deliver as many blocks as possible over SATA.
- 01
Hardware stabilization on the test bench
The drive is powered on the PC-3000 SSD test bench with SATA link power management disabled and host timeouts extended. Power rails are checked against datasheet expectations; any shorted or sagging rail is repaired at the board level (PMIC, DC-DC inductors, decoupling caps) before any imaging attempt. The PC-3000 bench never asserts HIPM, DIPM, or DevSleep, which removes the host-side LPM failure surface described later.
- 02
Normal-mode imaging with error isolation
Once the controller enumerates, imaging proceeds in head-first passes that skip over sectors that trigger timeouts or uncorrectable ECC returns, building a map of healthy, slow, and non-responsive LBA ranges. Skipped ranges are revisited in later passes, since marginal sectors on degraded V-NAND often read on a subsequent attempt when surrounding read-disturb pressure is lower.
- 03
Extended retry imaging for marginal LBAs
For LBAs that fail on standard reads, the bench issues repeated reads with extended per-command timeouts, giving the MKX's on-chip LDPC decoder more time to complete soft-decision decoding. The per-sector pass rate on the second and third attempts is typically non-trivial on drives with early retention loss; sectors that remain unreadable after the retry budget are logged as bad and the image is delivered with those gaps marked for filesystem-level handling.
- 04
Image assembly and file verification
The recovered sectors are assembled into a full drive image. File system analysis extracts the directory structure and verifies file integrity. Files are delivered on your choice of return media via our mail-in service.
Samsung 870 EVO Hardware Encryption and Chip-Off Limitations
The MKX controller applies AES-256 hardware encryption to all data written to NAND, even without BitLocker or Samsung's software encryption enabled. The encryption key is bound to the original controller. If the controller dies, the NAND holds only ciphertext.
Chip-off (desoldering the NAND packages and reading them with a raw NAND reader) is not a recovery option for the 870 EVO. The binary dumps yield only AES-256 ciphertext without the controller's key. Board-level repair to revive the original MKX controller is the only path to data. For broader Samsung SSD recovery procedures, including other controller families, see our Samsung hub page.
Samsung MKX Controller Firmware Failure Modes
The MKX controller has four distinct firmware-visible failure modes. Samsung locked technological-mode access on this generation, so the repair path is different from older Samsung SATA (840/850/860). Identifying the mode determines whether the case falls into the $600–$900 firmware tier on successful stabilization, the $450–$600 board repair tier, or is currently not recoverable.
- FTL Corruption
The Flash Translation Layer maps logical block addresses to physical NAND locations. When the FTL becomes inconsistent, the drive reports wrong capacity or 0 bytes. Unlike older Samsung SATA generations, the 870 EVO does not expose a supported firmware-level FTL rebuild path in PC-3000 SSD; recovery depends on whether the controller, once hardware-stabilized, can reconcile its own FTL from redundant copies in the NAND service area on boot. When it can, normal-mode imaging proceeds; when it cannot, the case is generally not recoverable today. This is the most common firmware corruption mode on 870 EVO drives, particularly after power loss events.
- ROM Corruption
The ROM region stores the controller's boot code and drive identity. When this region is corrupted, the MKX may report a generic model string (e.g., "Samsung SSD" instead of "Samsung SSD 870 EVO 2TB") or fail to present valid identity data. The drive can appear in BIOS but be inaccessible to the OS. Because Samsung locked loader-based firmware rewrites on this generation, the 870 EVO does not have a user-servicable ROM reflash path through PC-3000; recovery depends on stabilizing the original controller at the board level. Cases where the ROM itself is damaged are generally not recoverable with tools publicly available today.
- Module Corruption
Individual firmware modules (S.M.A.R.T. tables, bad block management maps, wear leveling configuration) can corrupt independently of the main firmware. A corrupted bad block table causes the controller to read garbage from remapped sectors; a corrupted wear leveling map concentrates writes on a subset of NAND blocks. On the MKX, whether the drive survives this depends on the controller's own boot-time integrity checks and redundant module copies in the service area. When the controller recovers on boot, normal-mode imaging yields data; when it cannot, this class of failure is generally not recoverable without vendor-level tooling that is not commercially available.
- Controller Lockout
After accumulating too many uncorrectable errors or a firmware exception loop, the MKX may stop responding to ATA commands, time out during POST, or fail to appear in BIOS. On the test bench the drive is powered with SATA link power management disabled and extended timeouts so host-side aggression is removed from the equation. If the controller has a shorted power management IC (PMIC) or a failed passive, we locate the failure using a FLIR thermal camera and replace the component with a Hakko FM-2032 on an FM-203 base station. Once supply is clean and the controller boots on its own, imaging proceeds in normal mode.
Board Repair as Data Recovery
When the MKX controller itself is dead rather than firmware-corrupted, the case shifts from firmware recovery to board-level repair ($450–$600). The AES-256 encryption key lives on the controller silicon. A dead controller means the NAND is ciphertext. Replacing the controller with a new one doesn't work; the new chip has a different key.
The only path: revive the original controller. We use FLIR thermal imaging to pinpoint the failed component on the PCB, then replace it with Hakko FM-2032 microsoldering. Common failures include shorted PMICs, blown voltage regulators, & cracked solder joints under the controller BGA package. Once the original controller boots, the encryption keys are intact & the data is accessible. Board repair isn't a separate service from data recovery for encrypted SSDs; it is data recovery.
For a full breakdown of SSD recovery pricing across all tiers, see our cost page.
What Is Inside the Samsung MKX Controller?
The MKX (internal designation Samsung S4LR059) is an ARM Cortex-R4 processor fabricated on Samsung's 14nm FinFET process. It manages 8 flash channels with 8 chip enables per channel, striping data across every NAND package on the PCB. That 8-channel interleave is why chip-off on an 870 EVO yields fragments from multiple packages rather than sequential files.
The MKX allocates LPDDR4 DRAM as a write cache & FTL mapping buffer. Capacity scales with the drive: 512MB of DRAM on the 250GB & 500GB models, 1GB on the 1TB, 2GB on the 2TB, and 4GB on the 4TB. The working FTL lives in this DRAM. Power loss before the controller flushes DRAM contents to NAND is what causes the FTL corruption described in the section above.
MKX vs. MJX: What Changed from the 860 EVO
The MJX controller (860 EVO) uses the same ARM Cortex-R4 architecture. Both run AES-256 Class 0 hardware encryption with TCG Opal 2.0 & IEEE 1667 support. The difference is the ECC engine. MJX was calibrated for Samsung's 64-layer (V4) & 96-layer (V5) V-NAND. MKX ships with a refined LDPC decoder tuned for the 128-layer (marketed) & 136-layer (actual) Gen-6 V-NAND stacking.
This matters for recovery triage. Older Samsung SATA generations (840/850/860) had published technological-mode recovery paths that distinguished between MGX, MHX, and MJX firmware images. Starting with the 870 EVO, Samsung locked that access path, so the practical triage question on an MKX is not "which image to reflash" but "can the original controller be hardware-stabilized long enough for a normal-mode image."
How Does Samsung V-NAND Physically Degrade?
Samsung's V-NAND uses Charge-Trap Flash (CTF) instead of the floating-gate design found in older planar NAND. CTF stores electrons in a non-conductive silicon nitride insulator rather than a conductive polysilicon floating gate. A single oxide defect in CTF doesn't drain the entire stored charge the way it does in floating-gate cells. That's the advantage. The failure modes are different too.
- Early Retention Loss (Lateral Electron Migration)
In 3D CTF, the silicon nitride trap layer is continuously connected across vertically stacked layers. Electrons stored in the nitride migrate laterally through the shared insulator, causing the threshold voltage (Vth) of programmed cells to drift over time. This isn't a P/E cycle problem; it happens even on cells with low write counts. The 870 EVO's 136-layer stack height increases the surface area of shared nitride, increasing the lateral migration path length & the resulting Vth drift rate.
- Tunnel Oxide Trapping
Every erase cycle uses Fowler-Nordheim tunneling to clear the charge trap. During tunneling, electrons pass through the thin tunnel oxide layer. Some get stuck in trap states within the oxide itself instead of reaching or leaving the silicon nitride. Over accumulated P/E cycles, these oxide-trapped electrons shift the cell's baseline Vth upward, narrowing the voltage margins between TLC's 8 programmed states.
- Read Disturb
Reading a single word line requires applying a pass voltage (Vpass) to all unselected word lines in the same block. Vpass is high enough to keep those cells conducting without reading them, but it also nudges electrons in the trap layer. Over thousands of reads, the cumulative Vth shift from Vpass stress pushes cells into adjacent voltage states. On an 870 EVO used as a boot drive with constant OS reads, read disturb accelerates the degradation timeline.
Where This Meets Recovery
All three mechanisms cause the same end result: Vth distributions of the 8 TLC voltage states start overlapping. When overlap exceeds the MKX controller's LDPC correction ceiling, the read fails as an uncorrectable error. Because Samsung locked technological-mode access on the 870 EVO, the practical recovery path is not vendor read-retry issued from the host; it is extended-timeout normal-mode imaging that lets the controller's own on-chip LDPC decoder spend more time on each sector before giving up. Sectors recovered on a second or third attempt at extended timeout are typical on drives in early retention loss; on a 2TB 870 EVO with advanced degradation, this multi-pass imaging is time-intensive and some LBAs may remain unreadable.
Samsung 870 EVO Firmware Version History & the 2021 Manufacturing Anomaly
Samsung released three firmware versions for the 870 EVO. The firmware revision on your drive determines which production batch it came from, what the controller's ECC behavior looks like, & whether a live firmware update is safe or dangerous on a degrading drive.
- SVT01B6Q (Launch Firmware)
Shipped with drives manufactured early to mid 2021. This is the firmware revision most associated with premature failures. Drives on SVT01B6Q have shown NAND degradation well before reaching their rated TBW endurance. The LDPC parameters in this firmware version appear inadequately tuned for the Gen-6 V-NAND cell geometry, contributing to earlier ECC ceiling breaches.
- SVT02B6Q (Late 2021 Update via Samsung Magician)
Released late 2021 through Samsung Magician. Added a new vendor-specific S.M.A.R.T. attribute: ID 0xFC, labeled "Newly Added Bad Flash Block." This attribute tracks NAND blocks degrading in real time, separate from the existing reallocated sector count. Updating to SVT02B6Q does not reverse existing damage. On drives with accumulated errors, the firmware flash process itself can brick the controller if the ROM write encounters a bad NAND block mid-flash.
- SVT03B6Q (Factory-Only, Mid-to-Late 2023)
Factory firmware for newly manufactured drives. Cannot be applied to older SVT01 or SVT02 drives through Samsung Magician or any user-accessible tool. The inability to cross-flash indicates a fundamental hardware-level change in the NAND. Samsung Magician documentation references a "revised V6 process starting November 2022," confirming a silicon-level manufacturing change between production runs.
S.M.A.R.T. Attributes That Matter for Early Detection
Three S.M.A.R.T. attributes provide the earliest warning of the 870 EVO degradation pattern. Check the raw values, not Samsung's normalized health percentage.
| Attribute | ID | What It Tracks | Warning Threshold |
|---|---|---|---|
| Reallocated Sector Count | 5 | NAND blocks remapped to spare pool | Any non-zero raw value on a new drive |
| Uncorrectable Error Count | 187 | Reads that exceeded LDPC correction | Rising count over days (back up immediately) |
| Hardware ECC Recovered | 195 | Reads that required ECC intervention | Spike relative to drive's own baseline |
S.M.A.R.T. attribute 0xFC ("Newly Added Bad Flash Block") is only visible on drives running SVT02B6Q or later firmware. CrystalDiskInfo displays it; Samsung Magician may not surface the raw value.
Samsung 870 EVO Symptom-to-Root-Cause Diagnostic Reference
Different symptoms point to different hardware root causes, and each root cause requires a different recovery approach & falls into a different pricing tier. Use this table to identify the probable hardware failure mode based on diagnostic symptoms.
| Symptom | S.M.A.R.T. Indicator | Root Cause | Recovery Approach | Typical Tier |
|---|---|---|---|---|
| Slow transfers, intermittent freezes | Rising Attr 187 & 195 | V-NAND early retention loss; Vth drift exceeding soft-decode margin | Extended-timeout normal-mode imaging | $600–$900 |
| BIOS freezes on boot; drive causes POST hang | S.M.A.R.T. inaccessible (drive in BSY state) | MKX controller FTL panic; firmware exception loop | Hardware stabilization & extended-timeout normal-mode imaging | $600–$900 |
| 0 bytes or RAW in Disk Management | S.M.A.R.T. returns all zeros or generic model string | FTL mapping collapse; controller entered protective mode | Generally not recoverable without vendor-level tooling | $600–$900 |
| Drive not detected at all (no BIOS, no Disk Management) | N/A (drive doesn't respond to any command) | PMIC failure or shorted component on PCB | FLIR fault localization & Hakko FM-2032 component replacement | $450–$600 |
| Bricked after Samsung Magician firmware update | N/A (drive unresponsive post-update) | ROM or module corruption during interrupted flash | Board-level stabilization; generally not recoverable if ROM itself is damaged | $600–$900 |
Free evaluation determines the exact failure mode & tier. No diagnostic fees. Call (512) 212-9111 or submit your case online.
How Do SATA Link Power Management Bugs Affect the 870 EVO?
Separate from NAND degradation, a subset of Samsung 860/870 EVO failures trace to documented link-layer firmware exceptions. The Linux kernel carries an explicit quirk (ATA_HORKAGE_NO_NCQ_TRIM) for this family because queued asynchronous TRIM can trigger a firmware panic that the host sees as a port reset cascade. Aggressive host-initiated or device-initiated SATA Link Power Management (HIPM/DIPM) and DevSleep can compound the symptom. To the operating system the drive appears to drop off the bus mid-operation, producing I/O errors or a read-only filesystem remount.
Why Link-Layer Aggression Exposes the 860/870 EVO
The well-documented Samsung 860/870 EVO failure surface at the link layer is the Native Command Queuing (NCQ) TRIM path, not LPM specifically. When the host issues queued asynchronous TRIM alongside normal I/O, the drive's firmware can throw an exception, the ATA port goes into error-handling, and the kernel issues hard resets. The Linux kernel ships an explicit quirk for this family, ATA_HORKAGE_NO_NCQ_TRIM, which forces TRIM to non-queued so the failure never occurs. Aggressive host-initiated or device-initiated power management (HIPM/DIPM) on top of an already-fragile queued path makes the same symptom more likely to appear under light idle workloads.
This is why the same 870 EVO that tested fine in an external USB-to-SATA dock can drop offline when connected internally to a desktop motherboard or laptop AHCI port. USB-to-SATA bridges typically translate host I/O into simpler non-queued SCSI commands and do not implement HIPM/DIPM or pass queued TRIM through, so the firmware path that triggers the kernel resets is never exercised.
Symptoms That Point to LPM, Not NAND
- I/O errors that stop when you disable DIPM or NCQ TRIM
If setting the Windows power plan's SATA link state to "Active" (which disables HIPM/DIPM), or passing
libata.force=noncqtrimon a Linux kernel command line, makes the dropouts disappear, the root cause is link-layer or queued-TRIM handling, not the NAND. The data on the drive is intact; the link is the failure surface.- Kernel log entries showing link resets and hard resets on one SATA port
Linux dmesg output containing "ataN.00: exception Emask" followed by "ataN: hard resetting link" and "failed command: WRITE FPDMA QUEUED" or "SEND FPDMA QUEUED" on a Samsung 860/870 EVO is the signature of the NCQ TRIM firmware exception documented in Linux kernel bug trackers. The kernel mitigates this with the
ATA_HORKAGE_NO_NCQ_TRIMlibata quirk, which forces TRIM to non-queued. After the reset the drive re-enumerates, but the filesystem may have already remounted read-only to protect against further corruption.- Drive enters read-only mode after a short period of idle time
If the drive responds normally for the first few minutes of a session and only starts dropping after the system has been idle long enough for Slumber or DevSleep to engage, the failure is LPM-triggered. A drive that failed during sustained heavy writes is a different failure class (NAND wear or FTL overflow) covered earlier on this page.
Recovery Workflow for an LPM- or NCQ-TRIM-Locked 870 EVO
When an 870 EVO arrives in read-only or repeated-dropout state, the drive is moved off the original host entirely. On the PC-3000 SSD test bench (or an imaging bench with a controller that supports AHCI with LPM disabled), the link-layer aggression that triggered the firmware exception is removed from the equation:
- Power the drive with SATA link power management disabled and with NCQ TRIM suppressed. On Linux this is how the
ATA_HORKAGE_NO_NCQ_TRIMquirk is applied in the wild; on a PC-3000 bench, the equivalent configuration is set at the bench controller so queued TRIM is never issued and HIPM/DIPM are not negotiated. - Verify board-level supply rails before any imaging attempt. A drive that drops offline intermittently sometimes does so because a PMIC or capacitor is marginal rather than because of a pure link-layer issue. Rails out of spec are repaired (Hakko FM-2032 on FM-203, Zhuo Mao for BGA work) before the next imaging pass.
- Image the user area in normal mode over SATA with extended per-command timeouts and head-first pass-skipping, so marginal sectors are deferred rather than re-read until the drive faults. Timeouts give the MKX's on-chip LDPC decoder time to complete soft-decision decoding on sectors in early retention loss. Skipped sectors are retried on later passes.
- After imaging, the raw image is mounted on a separate workstation for filesystem-level processing. No further reads are issued against the source drive.
A link-layer lockout (LPM-triggered or NCQ-TRIM-triggered) on an 870 EVO with an otherwise healthy NAND array falls into the firmware tier on successful stabilization ($600–$900), because the fix is removing host-side aggression and imaging in normal mode rather than rewriting firmware. If the underlying cause turns out to be NAND degradation that was masked by the link-layer symptom, the case moves to the intensive imaging tier ($1,200–$1,500). If the controller itself is dead or the internal FTL cannot reconcile on boot, the case is generally not recoverable today. Free evaluation identifies which bucket applies before any paid work starts. Rush service is available: +$100 rush fee to move to the front of the queue.
How Much Does Samsung 870 EVO Recovery Cost?
Samsung 870 EVO firmware-tier recovery: $600–$900. If extensive multi-pass imaging is required for heavily degraded NAND, the case may reach the $1,200–$1,500 tier (intensive imaging). Free evaluation and firm quote before any paid work. No data recovered means no charge.
The Samsung MKX controller uses hardware AES-256 encryption. Recovery requires the original controller to be functional, which means chip-off is not an option for these drives. Firmware-level recovery through the controller is the only viable path.
Rush service: +$100 rush fee to move to the front of the queue. Call (512) 212-9111 for a free evaluation.
Do Not Run Recovery Software on a Failing 870 EVO
Consumer recovery software (Disk Drill, EaseUS, Recuva, R-Studio) issues thousands of read commands across the entire drive surface. On a drive with degrading NAND, every read stresses the marginal cells. Each failed read triggers the controller's internal retry logic, adding heat and electrical stress to cells that are already failing.
Software scans also trigger background controller operations. The controller may attempt to relocate data from failing blocks to spare blocks. If the spare pool is exhausted, the controller may enter a protective lockout. Each background operation risks further corruption of the Flash Translation Layer. A drive that was recoverable before the software scan can become unrecoverable after it.
If your Samsung 870 EVO is showing any of the Stage 1 or Stage 2 symptoms described above, power down the drive. Do not run chkdsk. Do not run Samsung Magician diagnostics. Do not attempt to clone the drive with dd or Clonezilla. Send the drive for professional evaluation while the maximum number of NAND cells are still readable.
Frequently Asked Questions
Why does my Samsung 870 EVO show 0 bytes?
The 0 bytes symptom on Samsung 870 EVO drives typically indicates severe NAND degradation or firmware failure triggered by accumulated uncorrectable ECC errors. When the controller encounters more bad blocks than it can remap, it may enter a protective state that reports 0 capacity. The data is still on the NAND cells, but the controller has lost the ability to read it through normal host commands.
Is the Samsung 870 EVO failure a known defect?
Yes. Samsung 870 EVO drives from early 2021 production runs have user-reported premature NAND degradation. Users on ServeTheHome, Tom's Hardware, and multiple Reddit communities reported identical symptoms: uncorrectable read errors, system freezes, and eventual 0-byte detection. Reports are concentrated among higher-capacity models (2TB and 4TB). Samsung released firmware updates but did not publicly acknowledge a manufacturing defect.
How much does Samsung 870 EVO data recovery cost?
Samsung 870 EVO firmware-tier recovery costs $600–$900. If the case requires intensive multi-pass imaging to extract data from heavily degraded NAND, it may fall into the $1,200–$1,500 tier. Free evaluation and firm quote before any paid work. No data recovered means no charge.
Can Samsung Magician detect this failure before it happens?
Not reliably. Multiple users have reported Samsung Magician and CrystalDiskInfo showing 99% drive health immediately before catastrophic failure. S.M.A.R.T. monitoring on Samsung drives uses a proprietary algorithm that masks the actual bad block count behind a normalized health percentage. A drive can accumulate thousands of reallocated sectors while still reporting near-perfect health.
Should I run data recovery software on my failing 870 EVO?
No. Running recovery software on a drive with degrading NAND forces the controller to repeatedly attempt reads on failing cells. Each failed read stresses the surrounding cells and can trigger garbage collection or reallocation that destroys additional data. Power down the drive and send it for professional evaluation. The fewer read cycles on degrading NAND, the more data survives.
How long does Samsung 870 EVO recovery take?
Firmware-tier cases take 3-6 weeks. Intensive multi-pass imaging of degraded NAND with extended per-command timeouts can extend the timeline depending on how many LBAs require repeated attempts before the controller returns them. Rush service is available: +$100 rush fee to move to the front of the queue.
Can data be recovered after a failed 870 EVO firmware update?
A failed firmware update usually corrupts the ROM or module area, not the user data NAND. The MKX controller's firmware resides in a separate NAND region from user data, so the cells that hold your files are typically unaffected. Samsung locked firmware-level technological-mode access on the 870 EVO generation, so the fix is not a routine PC-3000 reflash. Recovery depends on board-level stabilization of the original controller (voltage rail and PMIC work) so the drive re-enumerates long enough for a normal-mode image. If the controller will not revive, the case escalates to board repair or is deemed not economically recoverable; free evaluation determines which applies.
Does the Samsung 870 EVO encrypt data by default?
Yes. The MKX controller applies AES-256 hardware encryption to all data written to NAND, even without BitLocker or Samsung's software encryption enabled. The encryption key is stored in the controller's secure area. Recovery requires the original controller to be functional. Chip-off (desoldering the NAND chips) yields only encrypted, unreadable data without the controller's key.
What is the difference between firmware recovery and NAND swap on an 870 EVO?
Firmware-tier work ($600–$900) covers cases where the original MKX controller can be stabilized through voltage-rail and PMIC repair so the drive re-enumerates long enough for a normal-mode image; Samsung does not permit PC-3000 technological-mode access on the 870 EVO generation, so this is not a VSC-based firmware rewrite. Board-repair-tier work ($450–$600) covers PCB component-level repair when passive components or rails have failed. When the controller silicon itself is dead, the AES-256 key is lost and the case is generally not recoverable, since chip-off yields ciphertext. Free evaluation determines which path applies. A donor drive is a matching SSD used for its circuit board. Typical donor cost: $40–$100 for common models, $150–$300 for discontinued or rare controllers.
Related Samsung SSD Recovery Pages
Full SSD recovery service overview
Samsung controller-specific procedures
Controller families across all SSD brands
Diagnosis of 0-byte SSD symptoms
NAND degradation and wear-related failures
Firmware-level controller failures
Full SSD pricing breakdown by failure type
Samsung 870 EVO showing errors or 0 bytes?
Free evaluation. Firmware recovery: $600–$900. Intensive NAND extraction: $1,200–$1,500. No data, no fee.