Quick Reference -- Start Here

This guide covers Unraid boot device selection across USB and internal NVMe configurations.

If you read nothing else, read this section.

---

Your situation in one question: Are you considering migrating to internal boot, or selecting a USB boot drive?

---

If you are considering internal boot migration

The reliability case for internal boot is real -- but it depends entirely on the hardware selected and how it is deployed.

The most common deployment pattern -- incorporating the boot partition into an existing cache NVMe -- expands the failure scope significantly compared to USB boot.

A failed USB boot drive loses OS configuration.

Recovery is disruptive but bounded -- replace the drive, restore configuration from backup, done.

A failed NVMe in combined boot plus cache duty loses OS configuration and all cached data simultaneously.

AppData, Docker configurations, VM images, and everything resident in the cache pool at failure are gone together.

If you are deploying internal boot on a dedicated drive separate from cache, this risk does not apply.

If you are incorporating the boot partition into your existing cache drive or pool, understand this failure scope before migrating.

TPM licensing introduces a second risk layer that USB licensing does not have.

Most modern motherboards use fTPM -- firmware TPM running inside the CPU rather than a dedicated physical chip.

fTPM state is stored in the motherboard's SPI flash alongside the BIOS firmware.

BIOS updates, Intel microcode updates, AMD platform firmware updates, and CMOS clears can all reset fTPM state -- changing the TPM identifier and triggering a license mismatch requiring a transfer.

A USB drive carries its GUID through any system event.

An fTPM identifier can change without warning during routine maintenance.

If your motherboard has a physical TPM header, a discrete dTPM module at $10-20 eliminates this vulnerability. If you are relying on fTPM, verify your specific platform's BIOS update history for fTPM resets before migrating.

The hardware agnosticism tradeoff.

A USB boot drive moves to any replacement hardware in a disaster recovery scenario -- no BIOS configuration, no M.2 slot requirement, no EFI boot entry management.

Internal boot with TPM licensing ties you to M.2 slot availability and TPM compatibility on replacement hardware simultaneously.

For the full internal boot hardware selection guide including NVMe NAND type recommendations and mirrored pool failure risks, see the post below.

---

If you are selecting a USB boot drive

Most USB drives implement a hardware read-only protection mode in their controllers -- when the drive detects imminent failure it locks itself read-only rather than allowing further writes that could corrupt existing data.

The result is a drive that can no longer boot the system but preserves the configuration in a fully readable state for immediate recovery.

Insert a replacement drive, copy the config folder, boot.

The failure that felt catastrophic resolves in minutes with zero data loss.

This self-preservation behavior is not as standardized in NVMe drives. When used in combined boot and cache configurations a failing NVMe is less likely to protect the data it shares with the OS partition.

Three rules cover most USB purchase decision situations:

Avoid: Planar TLC, any QLC, USB 3.x drives in USB 3.x ports, unverified budget drives, SanDisk (proprietary controllers).

Target: Planar MLC from 2007-2010 if you can find it.

Verified 3D TLC from Samsung, Kingston, or Transcend if you cannot.

Industrial MLC or SLC if budget allows.

Verify before committing: Use ChipGenius or Flash Drive Information Extractor to confirm NAND type (5 second check)

Do not assume based on brand or purchase date alone.

🚨🚨The single best current purchase available anywhere: 🚨🚨

64GB Innodisk 3ME industrial USB drive -- eBay item 326046070546 -- currently $3.99

P/N: DEUA1-64GI61BW1SC

Industrial MLC, 3,000 P/E cycles, 60-bit ECC, 30μ gold contacts, power-fail firmware, metal housing, S.M.A.R.T. capable*

Confirmed Toshiba MLC NAND via Flash ID decode on purchased units.

This is surplus liquidation stock from a commercial fleet application.

MLC NAND production is being phased out across the industry.

When this listing stock is gone it will not be replaced at the same price.

If you have an unused USB 2.0 drive from a quality brand sitting in a drawer from 2006-2012, verify it with ChipGenius before spending anything. You may already own the right answer.

For the full USB drive selection guide including NAND transition timeline, identification tools, and NOS MLC sourcing, see the post below.

---

Redundant boot without consuming internal slots

Unraid 7.3 also supports a mirrored USB boot pool -- two USB drives forming a ZFS mirror providing redundancy against single drive failure, consuming zero internal slots.

This configuration exists. It was not prominently documented in the 7.3 release materials.

Two industrial USB drives at $4 each on internal headers via inexpensive adapters provides redundant boot with complete hardware portability and zero slot consumption.

For full configuration details see the post below.

Note

Layout was AI-assisted.

Technical content was verified through primary sources, manufacturer documentation, and ChipGenius obtained data.

Corrections, comments and criticism welcome.

*Innodisk USB Drive 3ME -- iTracker Health Monitoring: Confirmed

The Innodisk 3ME (DEUA1-64GI61BW1SC) drive health monitoring available via Innodisk’s proprietary iTracker utility --

vendor-specific S.M.A.R.T. like metrics such as health percentage and erase count; not standard ATA S.M.A.R.T.

Confirmed attributes surfaced by iTracker on purchased units from the eBay surplus listing:

• Health percentage -- 99.93% on all tested units confirming factory-new condition

• Average erase count -- 2 across all units, consistent with production bench testing baseline

• Firmware version -- O0917v1

• Controller -- SM3261 confirmed

• Capacity -- 60.46 GB usable on all 64GB units

Individual anti-static packaging and identical erase counts across the batch confirm these are brand new drives that were never deployed.

iTracker is available directly from Innodisk support on request.

Reference the drive's part number DEUA1-64GI61BW1SC and the SM3261 controller when requesting the utility.

Identifying NAND Flash Type

To determine the NAND flash type (along with other details like controller and capacity), use these free utilities:

• ChipGenius: A simple tool for extracting drive information.

• Flash Drive Information Extractor by ANTSpec: Provides more detailed specs in most cases.

Notes on Identification:

• These tools typically identify basic NAND types (e.g., MLC, TLC, QLC).

• Distinguishing 3D TLC from planar TLC requires decoding the Flash ID and researching the NAND chip's specifications online (e.g., via manufacturer datasheets).

• If unwilling to research further: do not assume "TLC" means "3D TLC" based on purchase date alone — planar TLC stock from 2018–2020 production remains in retail circulation today. The only safe shortcut is buying a drive with manufacturer-confirmed 3D NAND (such as the Samsung BAR Plus, which explicitly documents V-NAND) and verifying with ChipGenius on arrival.

• SanDisk Exception: These utilities do not fully read SanDisk drives due to their proprietary controllers. Given SanDisk's documented GUID uncertainty this is largely academic -- SanDisk drives are not recommended for Unraid boot duty regardless of what verification reveals.

• Linux users: The equivalent verification path uses lsusb -v -d VID:PID to retrieve controller and device identifiers. This provides less complete information than ChipGenius -- NAND type is not directly reported and requires cross-referencing the controller and device strings against community databases. If Windows access is available even temporarily -- a friend's machine, a dual boot, a Windows VM -- ChipGenius verification is significantly more reliable and produces definitive results in seconds. The Linux path works but requires additional research steps to reach the same conclusions.

If shopping for New-Old-Stock (NOS) drives, target production years 2006-2009 for higher reliability

These years used MLC NAND with larger node sizes, offering better endurance.

NAND Transition Timeline: 4 -16 GB Consumer USB Flash Drives

2007 -- 100% MLC. No TLC from any manufacturer in any consumer product.

• Toshiba/SanDisk: 70nm → 56nm transition completed H1 2007. 56nm MLC dominant by year end. Largest cells of any major supplier -- best raw charge retention and endurance of the consumer era.

• Samsung: 51nm MLC entered mass production April 2007, transitioning from 60nm. Samsung USB drives of 2007 shipped with 51–60nm MLC.

• IMFT (Intel/Micron): 50nm MLC in shipping USB drives (Kingston DataTraveler, Lexar JumpDrive). Solid endurance at this node.

• Hynix: 48–51nm MLC. Comparable to Samsung and IMFT 50nm parts in practical endurance.

Excellent NOS target year across all brands.

---

2008 -- 100% MLC. Major node transitions mid-year across all suppliers, but no TLC in any consumer USB product.

• Toshiba/SanDisk: 56nm → 43nm mid-year. 43nm MLC dominant by H2 2008. Still excellent endurance territory.

• Samsung: 51nm → 40nm transition. 40nm Samsung MLC in shipping USB drives by H2 2008.

• IMFT: 34nm IMFT NAND began production in 2008, with drives using this node (Kingston, Lexar) appearing in the market from late 2008 onward.

• Hynix: 41-48nm MLC, transitioning toward 32nm by late 2008.

The node spread in 2008 is significant: a Toshiba-based drive is 43–56nm; a Samsung-based drive is 40–51nm; an IMFT-based drive from late 2008 could already be 34nm.

All are MLC, all suitable for Unraid -- but cell size and endurance vary meaningfully.

---

2009 -- TLC debuts exclusively at 16GB+. First year of cell-type divergence between manufacturers.

• Toshiba/SanDisk: 43nm → 32nm MLC in 4–8GB. Toshiba and SanDisk introduced TLC NAND chips in 2009 — appearing only in select Cruzer 16GB+ budget lines. First consumer TLC product anywhere.

• Samsung: 40nm → 32nm MLC in shipping 4–16GB drives.

• IMFT: 34nm MLC throughout 2009. Zero TLC. All Kingston and Lexar USB drives were MLC regardless of capacity.

• Hynix: 32–41nm MLC. No consumer TLC production.

4–8GB: ~100% MLC across all manufacturers.

16GB: MLC dominant; TLC appearing only in Toshiba/SanDisk budget lines.

---

2010 -- Samsung enters TLC production; 16GB becomes the primary intermix battleground.

• Toshiba/SanDisk: 32nm → 24nm MLC in 4 – 8GB. TLC expanding across more 16GB budget SKUs. 24nm MLC entering production Q3 2010 — appearing in shipping drives by late 2010/early 2011.

• Samsung: Began mass-producing TLC in 2010.

• Own-brand 16GB budget lines began using TLC. 32nm MLC still in 4–8GB Samsung drives.

• IMFT: 34nm → 25nm MLC. 25nm NAND entered mass production in 2010 and began appearing in Kingston and Lexar USB drives. MLC-only - produced no consumer TLC at the 25nm node.

• Hynix: 26–32nm MLC. No consumer TLC.

4GB: ~100% MLC.

8GB: MLC dominant; TLC only in Samsung and Toshiba/SanDisk own-brand budget lines.

16GB: Rapidly intermixing -- TLC dominant in vertically integrated brands; MLC persisting in third-party brands sourcing from IMFT or Hynix.

---

2011–2012 -- The inflection point. Toshiba/SanDisk's 19nm node hits volume production January 2012, producing MLC and TLC simultaneously.

From this node onward across all manufacturers, cell type cannot be assumed — it must be verified.

• Toshiba/SanDisk: 24nm → 19nm. Both MLC and TLC variants from the same process. 16GB predominantly TLC across all brands including third-party holdouts by end of 2012. 8GB rapidly intermixing.

• Samsung: 32nm → 27nm → 21nm. TLC spreading from 16GB into 8GB own-brand budget lines through 2012. Samsung's CTF architecture meant MLC and TLC diverged at the firmware/programming level rather than requiring different fab processes.

• IMFT: 25nm → 20nm MLC. Kingston and Lexar drives remained MLC at 4–8GB through 2011–2012. IMFT did not ship consumer TLC for USB drives at scale during this period.

• Hynix: 20–26 MLC. No significant consumer TLC from Hynix in this window.

4GB: Predominantly MLC -- 19nm TLC die sizes still uneconomical for 4GB.

8GB: Rapidly intermixing -- TLC dominant in own-brand and white-label; MLC persisting in IMFT-sourced and Hynix-sourced drives.

16GB: ~80–90% TLC across all mainstream brands by end of 2012.

---

2013–2014 -- TLC completes its takeover of 16GB and 8GB; 4GB turns.

• Toshiba/SanDisk: 19nm TLC dominant in 8-16GB. 15nm MLC production April 2014; 15nm TLC June 2014. 15nm TLC die sizes finally make 4GB economical for TLC. Older MLC fabs serving 4GB begin retiring.

• Samsung: 21nm → 16nm CTF. TLC universal in Samsung own-brand 8GB+ lines by 2013.

• IMFT/Micron: 20nm → 16nm. Consumer USB partners (Kingston, Lexar) begin transitioning entry-level 8GB+ to TLC at 16nm. 4GB IMFT-sourced drives hold MLC longest of any supplier combination.

• Hynix: 20nm → 16nm MLC, TLC following at same node. Transitioning entry-level 8GB lines to TLC through 2014.

2013 is the watershed for 8GB and 16GB.

2014 is the equivalent for 4GB.

After 2014, MLC in any consumer USB drive requires explicit verification.

---

2015–2017 -- Full TLC standardization at the worst planar nodes across all manufacturers.

Edited June 28Jun 28 by Lolight

NAND Flash Types: Technical Breakdown and Relevance to Unraid Boot Drives

This section explains the main NAND flash architectures used in USB drives and SSDs, focusing on their cell structure, historical timelines, key characteristics, and why they matter for Unraid's USB boot device.

Unraid loads its OS into RAM and keeps most operations (including standard logging) in memory, resulting in infrequent writes to the USB boot drive -- mainly for config changes, updates, and shutdowns -- making cell interference resistance (to avoid bit-flips/corruption) and low heat signature (to minimize long-term wear in 24/7 operation) key factors for drive longevity and stability.

1. Planar MLC (Multi-Level Cell) – 2 bits per cell (4 voltage levels)

Era: Approximately 2006-2015 in production. Best NOS targets are 2007-2010 -- the window where node sizes remained large enough to deliver maximum reliability while capacity was sufficient for current Unraid installations. No longer in production at consumer scale.

Architecture: Cells laid flat on a silicon wafer at progressively shrinking node sizes. Unlike planar TLC, MLC's two-bit design meant node shrinking was less catastrophic -- wider voltage margins provided meaningful tolerance even as geometries tightened.

Node size timeline (consumer USB drives):

• 2005–2006...............70 nm – 90 nm

• 2007.........................51 nm – 56 nm

• 2008–2009...............34 nm – 43 nm

• 2010.........................25 nm – 32 nm

• 2011–2012...............19 nm – 24 nm

• 2013–2015...............15 nm – 19 nm (final planar generations)

Drives at 34nm and above represent the most reliable consumer MLC generations. Node size shrinks with each year -- earlier is better within the MLC era.

Key characteristics:

• Interference: Four voltage levels with wide margins between states make bit-flips from neighboring cells rare -- even at smaller nodes. This is the fundamental reliability advantage of MLC over TLC and QLC. The controller managing MLC is handling a significantly less demanding error correction workload than any TLC or QLC equivalent.

• Heat: Low. Wide voltage margins mean the controller's error correction workload is minimal. Runs cool under sustained use and cooler still in Unraid's near-idle boot application. USB 2.0 power draw compounds this advantage -- MLC's inherent thermal efficiency combined with USB 2.0's lower power envelope makes this the coolest running category available outside of industrial SLC.

• Endurance: 1,500–5,000 P/E cycles in consumer products. Higher end at 34nm and above -- lower end at 19-25nm late-node variants. In Unraid's near-idle boot application even the lower end of this range represents effectively unlimited operational life. P/E cycle exhaustion is not a realistic failure mode for MLC in this application.

• Unraid fit: Excellent -- the best consumer option outside of industrial SLC and the most practical recommendation given current market conditions. NOS drives from 2007–2010 with 34–56nm nodes are the primary target. Sealed examples surface on eBay periodically at $10–25 -- verify with ChipGenius on arrival. Check the USB Flash section of this forum for confirmed MLC models with documented component output before purchasing.

2. Planar TLC (Triple-Level Cell) – 3 bits per cell (8 voltage levels)

Era: ~ 2011-2019 in consumer USB drives (declining from 2017 onward as 3D TLC took over)

Architecture: Same flat layout as planar MLC but storing 3 bits per cell, combined with continued aggressive node shrinking to 15nm and below. This compounded both density and reliability problems simultaneously -- more bits per cell demanding tighter voltage margins, at geometries where those margins were already compromised by physical cell size.

Key characteristics:

• Interference: Cells shrunk to 15-19nm leave extremely thin barriers between neighbors. Charge leakage between cells -- primarily a quantum tunneling and long-term retention loss phenomenon rather than a thermal effect -- is the dominant data corruption mechanism. This is a passive process that continues regardless of whether the drive is being written to. A drive that reads correctly today may accumulate sufficient charge leakage to fail after months of always-on uptime.

• Heat: Higher than MLC under equivalent workloads due to more complex controller error management at marginal voltage states. Not as severe as QLC but meaningfully worse than MLC or large-node planar TLC from earlier in the era.

• Endurance: 300 - 1,000 P/E cycles in practice. Lower end at 15nm and below -- upper end only at early large-node 19nm+ variants. Substantially lower than the MLC range it replaced and the 3D TLC that followed it.

• Unraid fit: Avoid. Drives from this era are prone to silent, gradual data corruption over months of uptime. A drive that reads correctly today may fail to boot after a year of Unraid use.

3. 3D TLC (Triple-Level Cell) -- 3 bits per cell (8 voltage levels)

Era: Displacing planar TLC in consumer USB drives gradually through 2019–2022. Planar TLC assembled stock remains in retail circulation -- purchase date alone does not confirm 3D TLC. Verify via ChipGenius or buy only from manufacturers that document 3D NAND explicitly.

Architecture: Cells stacked vertically rather than shrunk horizontally -- density achieved by building upward rather than inward. Layer counts started at 24-32 and now exceed 100 in current production. This was the industry's direct response to the reliability ceiling late-node planar TLC had reached.

Key characteristics:

• Interference: Vertical stacking puts physical distance between adjacent cell strings, significantly reducing coupling compared to planar designs.

• Heat: Moderate - USB 2.0, High - USB 3.x The larger 3D die footprint dissipates heat better than late-node planar TLC -- but form factor matters as much as NAND type. A compact plastic USB 3.x drive runs significantly hotter than a full-size metal-bodied drive with identical NAND.

• Endurance: Overlaps with low-range consumer MLC at the high end but doesn't match large-node MLC at its best. In Unraid's near-idle boot application endurance is largely irrelevant -- idle heat and controller quality are the dominant reliability variables.

• Unraid fit: Good -- with conditions. The practical choice when NOS MLC or industrial options are unavailable. Use full-size metal body, USB 2.0 port where possible, authorized retail only. The Bar Plus is the ceiling of the consumer 3D TLC category for this application -- and real-world always-on failure reports confirm even that ceiling has documented limitations

4. 3D QLC (Quad-Level Cell) -- 4 bits per cell (16 voltage levels)

Era: Widespread in consumer USB drives from approximately 2020 to present. The dominant NAND type in the cheapest currently available drives.

Architecture: Vertically stacked cells storing 4 bits per cell. Exclusively 3D -- there is no planar QLC. Optimized entirely for density at the expense of every reliability margin.

Key characteristics:

• Interference: 16 voltage states packed into a single cell leave microscopic margins between levels. Minor charge coupling from a nearby write can shift a cell by one state, silently corrupting data without triggering an immediate error. Silent bit-rot is a characteristic failure mode rather than an edge case.

• Heat: Highest of any consumer NAND type. LDPC error correction runs continuously regardless of user I/O -- compensating for QLC's inherently marginal voltage states keeps the controller significantly warmer than MLC or TLC equivalents even at complete idle. This is not activity-dependent heat. It does not stop.

• Endurance: 100–300 P/E cycles in practice -- the lowest of any consumer NAND type. In a high-write application this matters enormously. In Unraid's near-idle boot application P/E exhaustion is less likely than controller or thermal failure first.

• Capacity note: QLC dies are not economically produced below 128GB. Any QLC-based drive at 64GB or below is almost certainly a cheap generic, a counterfeit, or built from down-binned rejected NAND from higher capacity production runs -- combining QLC's inherent weaknesses with pre-existing manufacturing defects.

• Unraid fit: Avoid entirely. The continuous idle heat generation and controller stress make QLC actively unsuitable for always-on boot duty regardless of capacity, brand, or price paid.

Practical Consumer USB drive ranking for Unraid boot use:

• Best: Planar MLC NOS from 2007-2010 (34-56nm nodes). Large P/E budget (~5,000 cycles), wide voltage margins that minimize controller workload, and no TRIM dependency. Will outlast the server hardware in near-idle boot duty.

• Acceptable for new purchases: 3D TLC from a reputable manufacturer, full-size metal body, USB 2.0 port. Sufficient P/E budget (1,000-3,000 cycles) and modern controllers with better wear management than their planar-era predecessors. Authorized retail only. Backups non-negotiable.

• Avoid: Planar TLC (any era) -- thin P/E budget consumed by background controller activity independent of Unraid's writes. Late-node 15nm is particularly poor. Early-node 19-24nm is mediocre but survivable with a quality controller.

• Avoid: Any QLC.

• Avoid: Counterfeits and cheap generics regardless of stated NAND type.

• Avoid: USB 3.x drives in USB 3.x ports -- lower power draw of USB 2.0 operation meaningfully reduces thermal stress in always-on duty.

USB Flash Drive Mirrored Boot Pool -- Confirmed Supported Configuration (Unraid 7.3)

Unraid 7.3's boot pool architecture supports USB flash drives and USB DOMs as boot pool members.

This means a mirrored boot pool providing redundancy against single drive failure is achievable using USB hardware without consuming any internal drive slots, SATA ports, or M.2 slots.

License slot consumption note: Each USB drive in a mirrored boot pool counts toward your license's drive allocation regardless of whether the dedicated boot pool option is used.

Two drives in a mirrored USB boot pool consume two license slots.

Users on Basic or Starter licenses who already have an existing cache NVMe or cache pool should be aware that incorporating the boot partition into that existing drive or pool -- rather than adding dedicated USB boot pool members -- avoids any additional license slot consumption entirely.

The mirrored USB boot pool configuration is most practical for Unleashed and legacy Unlimited license holders where the drive count ceiling doesn't apply.

How It Works

The configuration has two variants depending on licensing method.

With TPM licensing:

Drive 1 -- USB flash drive or DOM, 8GB minimum, boot pool member

Drive 2 -- USB flash drive or DOM, 8GB minimum, boot pool member

License -- stored on motherboard TPM chip, USB drive removed entirely

With USB licensing (no TPM available):

Drive 1 -- USB flash drive or DOM, 8GB minimum, boot pool member

Drive 2 -- USB flash drive or DOM, 8GB minimum, boot pool member

Drive 3 -- existing USB license drive, with no size restrictions, holds the license key as before

In both variants the two boot pool drives form a ZFS mirror.

If either fails the system continues booting from the surviving drive.

In the USB licensing variant the license drive remains completely separate from the boot pool -- it is never partitioned or reformatted and requires no changes from the current setup.

The Minimum Size Requirement

Boot pool member devices must be 8GB or larger when used as dedicated boot pool members.

The license drive is not subject to this restriction and can remain whatever size it currently is.

The NAND Quality Consideration

The 8GB minimum requirement meaningfully improves the hardware selection picture for this configuration compared to beta.1's 16GB minimum.

The optimal USB drives documented in this guide -- NOS MLC from 2007-2010 at 34-56nm nodes -- are typically 4GB or 8GB.

The 4GB drives remain ineligible.

The 8GB drives from the same era are now viable boot pool candidates -- and represent the best available NAND for this application at any price point.

Verified 8GB NOS MLC drives from the optimal era should be the first target for anyone building a mirrored USB boot pool.

For users unable to source verified 8GB NOS MLC drives the same NAND verification discipline that governs all USB drive selection applies.

Target verified 3D TLC or better from a reputable manufacturer.

Use ChipGenius or Flash Drive Information Extractor to confirm NAND type before committing any drive to boot pool duty. Avoid unverified budget drives regardless of stated specifications.

Industrial USB DOMs at 8GB or above -- from ATP, Innodisk, or Swissbit — are the optimal new-purchase choice for boot pool members where budget allows.

Their MLC or SLC NAND, documented specifications, and internal header mounting make them the most reliable available new-purchase option for this role.

Note that verified 8GB NOS MLC consumer drives from the optimal era deliver superior node geometry to most currently available industrial DOMs at lower cost -- the industrial option's advantage is supply chain confidence and new-from-distributor provenance rather than NAND quality superiority.

The License Drive Remains Unchanged

The existing license USB drive continues operating exactly as before -- no reformatting, no migration, no minimum size requirement beyond Unraid's standard 4GB minimum.

For users currently running a 4GB or 8GB MLC drive as their boot and license device that drive stays in place as the license holder while the two new boot pool drives handle the actual boot process.

The 4GB NOS MLC drives that are ineligible for boot pool membership due to the minimum size requirement are specifically well suited for the permanent license holder role -- their optimal NAND quality makes them ideal for the always-on always-present function they now occupy.

Hardware Agnosticism Fully Preserved

Unlike internal NVMe boot this configuration maintains complete hardware portability.

All three USB drives physically move to any replacement hardware.

No M.2 slot availability required.

No TPM compatibility required.

No EFI boot entry management.

No GRUB module installation.

The disaster recovery scenario — move drives to any available hardware and boot -- works identically to traditional USB boot.

For Users With Internal USB Headers

Both USB DOMs and standard USB flash drives can connect to internal motherboard USB headers -- DOMs natively, standard flash drives via an inexpensive 9-pin to USB-A adapter or adapter cable available for $5-10.

Either approach physically protects the boot pool drives from accidental contact while consuming no external ports.

Two drives on internal headers as the boot pool mirror plus the existing license drive on an external rear port is the cleanest available configuration -- redundant boot, physically protected drives, no internal slots consumed, full hardware agnosticism preserved.

Industrial USB DOMs remain the optimal choice for internal header mounting due to their verified NAND specifications and industrial-grade construction.

Standard flash drives on 9-pin adapters are a practical, accessible or sometime even superior alternative where DOM sourcing or budget is a constraint -- NAND type verification via ChipGenius applies regardless of which approach is used.

Combined with TPM licensing if available the internal header configuration eliminates the external USB drive entirely while maintaining full boot redundancy.

Practical USB Mirrored Boot Pool Recommendations

• Best: Two ATP, Innodisk, or Swissbit industrial USB DOMs at 8GB or above on internal headers.

  Verified SLC, pSLC or MLC NAND, documented specifications, physical protection.

  Existing license drive remains or TPM licensing used instead.

• Good: Two verified 3D TLC USB flash drives at 8GB or above from Samsung, Kingston, or Transcend -- ChipGenius confirmed NAND type before purchase.

  Full-size metal body preferred for thermal management.

  Can be mounted internally via inexpensive 9-pin to USB-A adapter for physical protection. Existing license drive remains as is.

• Avoid: Unverified budget drives at any capacity.

  The boot pool's always-on duty makes NAND quality as important here as in any other Unraid boot device role.

Note on existing small MLC drives: The 4GB MLC drive is too small for boot pool membership but remains the optimal license holder.

Keep it in place unchanged -- its role shifts from boot device to license anchor while the boot pool handles actual boot duty.

The Bottom Line

For most existing Unraid users the migration path requires no changes to current hardware beyond adding two new USB drives.

The existing license drive stays.

The boot pool adds redundancy that USB boot has never previously had.

No internal slots consumed.

No TPM required.

No NAND quality complexity beyond the existing USB drive selection guidance this guide already provides.

This configuration was not mentioned in the 7.3 promotional materials or tutorial videos.

It was confirmed through community questioning in the beta forum thread.

Consider it the most accessible redundant boot option available in 7.3 -- achievable with two USB drives costing a few dollars each and hardware most Unraid users already own.

Internal Boot Device Selection -- The Same Physics, Higher Stakes (Unraid 7.3.x)

Unraid 7.3 introduced internal boot -- the ability to boot from an NVMe, SSD, or eMMC instead of a USB flash drive.

The 7.3 release notes acknowledge directly that "manufacturers have quietly shifted to cheaper NAND, endurance ratings have dropped, and flash failures have become more common" as the motivation for this change.

This is the correct diagnosis of the USB failure cause.

What the release notes do not address is that the same NAND quality variable applies directly to internal boot device selection.

The hardware choice made at boot device selection determines whether internal boot delivers its reliability premise -- or reproduces the same failure in a form factor where the consequences are larger.

Understanding Internal Boot Failure Consequences

A failed USB boot drive loses OS configuration.

Recovery involves replacing the drive and restoring configuration from backup -- disruptive but bounded in scope.

An NVMe in combined boot plus cache pool duty -- the most common internal boot deployment pattern -- loses OS configuration and all cached data simultaneously in a single failure event.

AppData, Docker configurations, VM images, and any data resident in the cache pool at the time of failure are all lost together.

Understanding this failure scope helps calibrate the hardware selection decision that follows.

The right NVMe choice can make internal boot genuinely more reliable than USB.

The wrong choice expands the failure scope without improving the failure probability.

NAND Type Hierarchy For NVMe Internal Boot

The same NAND type hierarchy that governs USB boot device selection applies to NVMe selection -- with different capacity thresholds reflecting NVMe market economics.

Optimal: Intel Optane M.2

Optane is not NAND flash.

It uses 3D XPoint technology -- a fundamentally different storage mechanism with no floating gates and no charge leakage physics.

The controller has essentially no idle maintenance workload -- it does not need to continuously scan cells, measure charge states, correct errors, or refresh drifting cells.

In always-on near-idle Unraid boot duty this means near-zero idle controller heat regardless of how long the drive has been running.

Intel discontinued Optane in 2022.

The M.2 form factor devices -- 16GB to 32GB capacities -- are available NOS and used on eBay at $15-30.

Either capacity is more than sufficient for Unraid boot duty since Unraid runs from RAM.

Verify the specific model before purchasing -- not all Optane M.2 devices use the same interface.

The Optane H10 is a hybrid device combining Optane and QLC NAND -- avoid it.

The Optane Memory M10 and P1600X are pure Optane and the correct targets.

Optane combined with TPM licensing eliminates USB dependency entirely while providing the best possible always-on boot device characteristics.

This is the architecturally optimal internal boot configuration.

Current implementation note -- license drive allocation

Internal boot devices count toward your license's drive allocation.

This applies regardless of whether you use the split boot pool or the dedicated boot pool option.

Dedicated boot pool resolves the architectural split -- a drive used purely for boot no longer requires a forced data partition -- but the drive still consumes one license slot.

Dedicated boot devices still count for licensing with no current plans to change this.

Who this affects in practice:

Most users migrating to internal boot will incorporate the boot partition into their existing cache NVMe or cache pool.

That configuration consumes no additional license slots -- the cache drive was already counted before migration.

Nothing changes in their slot allocation.

The users specifically affected are those following the guide's most technically sound recommendations -- a dedicated Optane drive for boot, or a mirrored USB boot pool for redundancy without consuming internal slots.

These configurations add one or two new devices to the license count respectively.

The practical consequence by license tier:

Unleashed, Lifetime and legacy Pro -- unaffected, no drive count ceiling applies.

Starter, Legacy Basic -- 6 attached devices maximum.

Legacy Plus -- 12 devices maximum.

A dedicated Optane boot drive consumes one slot, mirrored Optane boot pool -- two slots.

A mirrored USB boot pool consumes two slots.

Users on Starter or Basic licenses with 5 or 6 drives already assigned have no migration path to these configurations without either removing a data drive or upgrading their license.

The entry licenses are the most penalized by dedicated boot device slot consumption.

Classic USB boot consumes zero drive slots from your license. Every internal boot configuration consumes one or two.

On a Starter, legacy Basic or legacy Plus license that difference is real money.

Good: 3D TLC from verified manufacturers

3D TLC NVMe from Samsung, WD, Crucial or similar is the best practical consumer option when Optane is unavailable.

Samsung 980, WD SN770, and Crucial P3 at capacities below 500GB are verified 3D TLC.

Above 500GB the probability of QLC increases significantly depending on manufacturer and SKU -- verify before purchasing via latest reviews if available.

The same controller quality caveat applies as in the USB section -- a verified 3D TLC drive from an established manufacturer with a known controller is meaningfully better than an unverified drive claiming equivalent specifications.

Avoid: QLC NVMe at any capacity

Consumer NVMe drives at 500GB and above from budget manufacturers are overwhelmingly QLC.

At 1TB and above QLC is essentially universal outside of explicitly pro-grade drives.

The same continuous idle controller heat that makes QLC USB drives unsuitable for always-on boot duty applies identically to QLC NVMe -- and in a combined boot plus cache role the failure scope expands to include cached data alongside OS configuration.

The capacity threshold where QLC becomes the likely NAND type is lower for NVMe than for USB -- budget NVMe at 256GB is already at risk depending on manufacturer.

Verify with manufacturer documentation (not reliable) or latest online hardware reports before committing any NVMe to internal boot duty.

Mirrored NVMe Boot Pool -- Sequential And Correlated Failure Risks

A mirrored NVMe boot pool addresses random independent failure -- if one drive fails unexpectedly the other continues operating.

This is the redundancy the mirror is designed to provide and it works as intended for that specific failure mode.

Two additional failure risks apply that mirroring alone does not address.

The sequential failure window -- when one drive fails after years of always-on thermal stress the remaining drive is at a similar degradation level from identical conditions.

During the rebuild window -- before a replacement arrives and the mirror completes -- the degraded single drive carries both boot and cache data.

The probability of the second drive failing during that window is meaningfully higher for thermally stressed hardware approaching end of life.

The NAND type determines how wide that window is and how risky it becomes -- which is why hardware selection remains relevant even with a mirrored configuration.

The correlated failure risk -- drives from the same production batch share manufacturing tolerances, firmware versions, and wear leveling algorithms.

A firmware bug that causes one drive to fail will cause the other to fail on the same timeline -- sometimes simultaneously.

A documented real-world example: a firmware defect caused certain enterprise SSDs to brick themselves at exactly 32,768 hours of operation -- mirrored pairs failed at the same moment, rendering the redundancy completely useless.

Wear leveling synchronization compounds this -- two identical drives receiving identical write patterns will approach their endurance limits simultaneously.

The mitigation for correlated failure is straightforward -- mix brands or models in any mirrored boot pool.

A Samsung 980 mirrored with a WD SN770 uses different controllers, different firmware, and different NAND from different manufacturers.

The probability of a shared defect approaches zero.

If identical drives are unavoidable purchase them from different vendors at different times to ensure different production batches.

This diversification principle applies equally to USB flash drive mirrored boot pools -- two drives of identical make and model from the same purchase share the same correlated failure risks as identical NVMe drives.

The Hardware Agnosticism Trade-off

Internal boot introduces a hardware dependency that USB boot does not have.

A USB boot drive physically moves between any machine with a USB port -- no BIOS configuration, no M.2 slot requirement, no EFI boot entry management.

In a disaster recovery scenario where replacement hardware is whatever is immediately available this flexibility has genuine practical value.

TPM licensing compounds the dependency -- the license is tied to a specific motherboard's TPM chip rather than a portable physical device.

Moving to emergency replacement hardware requires both M.2 slot availability and TPM compatibility simultaneously.

Users who prioritize hardware agnosticism and disaster recovery flexibility should weigh this dependency against the physical connector robustness that internal boot provides.

A quality MLC USB drive in traditional USB boot configuration already achieves the reliability standard internal boot is designed to reach -- for users running legacy drives that have proven reliable, migration may offer less incremental benefit than the hardware dependency cost warrants.

Internal boot's reliability advantage over USB boot materializes when the NVMe hardware selected is genuinely superior to the USB drive being replaced.

The NAND quality awareness this guide provides is what makes that determination possible.

Practical Internal Boot Recommendations:

• Best: Intel Optane M.2 16-32GB with TPM licensing. Optimal technology for always-on near-idle duty, complete USB elimination, effectively unlimited endurance. Available NOS/used at $15-30.

• Good: 3D TLC NVMe from Samsung, WD, Crucial etc. -- verified manufacturer documentation, backed by latest online reviews confirming 3D TLC. Combined with USB licensing if TPM unavailable, TPM licensing if available.

• Avoid: Any QLC NVMe regardless of capacity, brand, or price. The continuous idle controller heat, high write amplification and expanded failure scope in combined boot plus cache duty make QLC unsuitable for this role.

• Avoid: Unverified budget NVMe at any capacity. Apply the same counterfeit and undisclosed NAND type caution that governs USB drive purchasing.

Evaluate for your situation: Internal boot with USB licensing -- the USB drive remains required, the boot architecture gains complexity, and the reliability improvement depends entirely on the NVMe hardware selected.

Users with functional high quality USB drives running reliably may find their existing configuration already represents the reliability standard this feature is designed to achieve.

License allocation note: The internal boot device(s) currently counts toward your license's drive allocation.

Verify available slots before migrating, particularly on Starter, Legacy Basic and Legacy Plus licenses.

TPM Implementation -- Stability Varies By Type (Unraid 7.3)

TPM licensing anchors your Unraid license to a hardware identifier the same way USB licensing anchors it to the drive's GUID.

What the promotional materials don't address is that not all TPM implementations provide the same identifier stability -- and the type of TPM your motherboard uses determines whether the licensing anchor is as reliable as the feature implies.

Two distinct TPM implementations exist with meaningfully different reliability characteristics for license anchoring.

dTPM -- Discrete TPM

A physical dedicated chip -- either soldered to the motherboard or installed via an add-on header module.

The TPM functionality runs in its own dedicated silicon entirely independent of the CPU and chipset.

Its identifier is burned at manufacture and stored in the chip's own non-volatile memory.

For Unraid licensing this means the identifier persists regardless of BIOS updates, CMOS clears, SPI flash reflashing, or other system maintenance events.

A dTPM module can also be physically moved between compatible motherboards in some configurations -- carrying its identifier with it similarly to how a USB drive carries its GUID.

dTPM is the more reliable implementation for license anchoring.

It is also the less common one -- most modern consumer motherboards ship without a discrete module installed even when a header is present.

fTPM -- Firmware TPM

Not a physical chip.

A TPM implementation running as firmware within the CPU's secure execution environment -- AMD's Platform Security Processor or Intel's Platform Trust Technology.

The TPM state is stored in a dedicated region of the motherboard's SPI flash -- the same flash that stores the BIOS firmware.

This is the default TPM implementation on the vast majority of modern consumer motherboards.

Most users migrating to Unraid TPM licensing will be using fTPM without necessarily knowing it.

The reliability concern for license anchoring is specific and documented.

fTPM state stored in SPI flash is vulnerable to events that dTPM is immune to.

BIOS updates -- the most common trigger.

Some BIOS updates clear or reset fTPM state as part of the firmware update process.

AMD Ryzen platforms have documented fTPM disruption on certain BIOS updates.

Intel 13th and 14th generation users receiving microcode stability updates face the same risk.

A BIOS update that resets fTPM state changes the TPM identifier -- triggering a license mismatch that requires a transfer.

CMOS clears -- resetting BIOS to defaults or clearing CMOS can reset fTPM state on some motherboard implementations.

A user who clears CMOS while troubleshooting a boot problem may inadvertently trigger a license mismatch.

Motherboard failure -- fTPM state exists in the failed board's SPI flash and is not recoverable.

A dTPM module removed from the failed board carries its state to replacement hardware.

An fTPM's state is lost with the board.

The License Transfer Budget Implication

The documented USB license transfer allowance -- one self-service transfer per year via the automated system, additional transfers requiring support contact -- was designed around deliberate hardware changes.

Users considering fTPM licensing on platforms with frequent BIOS updates should verify the transfer terms with Unraid support before migrating -- particularly if their hardware or update practices make involuntary fTPM resets a realistic possibility.

Identifying Your TPM Type

In Windows --> Device Manager --> Security Devices --> Trusted Platform Module shows the manufacturer.

AMD, Intel or Standard as manufacturer indicates fTPM.

A dedicated chip manufacturer indicates dTPM.

In BIOS -- Security or Trusted Computing settings typically show AMD fTPM, AMD PSP, Intel PTT, or Intel TXT for firmware implementations versus a specific TPM version number for discrete modules.

In Unraid --> Tools --> System Devices shows TPM information after enabling it in BIOS.

Practical TPM Recommendations

• Most stable for licensing: A dTPM discrete module installed in the motherboard's TPM header.

  Identifier persists through BIOS updates and maintenance events.

  Recommended for users who update BIOS frequently or who have experienced fTPM instability on their specific platform.

• Acceptable for most users: fTPM on platforms with infrequent BIOS update histories.

  Understand that BIOS updates carry a risk of fTPM state reset.

  Keep a record of your current license state before performing BIOS updates. Avoid clearing CMOS unnecessarily.

• Verify before migrating: Whether your board uses fTPM or dTPM, and whether your specific BIOS update history has produced fTPM resets on your platform.

  AMD Ryzen users in particular should check community reports for their specific motherboard model before committing to fTPM licensing.

• Legacy hardware without TPM: USB licensing continues working exactly as before.

  Internal boot remains available with USB licensing -- the USB drive stays required for license validation.

  No TPM header or module purchase is necessary unless TPM licensing is specifically desired.

Note on header availability: Many modern motherboards have a TPM header physically present but ship without a module installed.

Installing an inexpensive dTPM module in that header -- typically $10-20 -- provides the more stable implementation where fTPM instability is a concern.

Verify your specific motherboard's header pinout before purchasing -- 14-pin and 20-pin headers are not interchangeable.

PCIe Lane Sharing -- Verify Before Migrating

Consumer motherboards frequently share PCIe lanes between NVMe slots and SATA ports through the chipset.

Adding a dedicated NVMe boot drive to a secondary or tertiary M.2 slot can silently disable SATA ports currently in use or downgrade their bandwidth -- without any obvious warning during the migration process.

This is not a theoretical risk.

Consumer motherboard architecture commonly routes NVMe slots 2 and 3 through the chipset rather than directly to the CPU.

Populating these slots can disable SATA ports 5 and 6 on the same chipset lanes, or force remaining SATA connections into reduced bandwidth modes.

Before committing any NVMe slot to internal boot duty:

• Consult your motherboard manual's M.2 and SATA compatibility matrix -- typically found in a table showing which ports become unavailable when specific slots are populated

• Verify that no currently active SATA data drives share lanes with the intended boot slot

• Confirm the slot's direct CPU versus chipset connection -- only slot 1 connects directly to the CPU on most consumer boards

This verification takes five minutes and prevents a migration that silently degrades or disables existing storage connections.

Edited June 23Jun 23 by Lolight

I bought this one for < 20 Eur. Brand new one.

16GB Transcend JetFlash 180I (industrial-grade USB flash drive with SLC mode, ECC https://us.transcend-info.com/embedded/product/embedded-flash-solutions/jetflash-180i  ).

Edited February 3Feb 3 by bagican

10 minutes ago, bagican said:

I bought this one for < 20 Eur.

That's a very good one, a steal for <20 Eur.

Edited March 20Mar 20 by Lolight

Thanks for this.

I added a companion topic at https://forums.unraid.net/topic/196975-tested-usb-flash-drives-good-and-bad/