Choosing NAS hard drives: seven checks

A product-line name on a drive is like a pizza menu: "Margherita" tells you the genre, not whether the base is thin. This guide shows how to read the data sheet instead, so the box cannot surprise you.

The short answer

Pick NAS drives by these checks, in this order: recording technology (CMR, not SMR), workload rating against what your NAS really moves, bay rating and vibration sensors, your NAS maker's compatibility list, capacity against rebuild time and RAID level, and then warranty, noise and power. At every step, look up the exact model number, because one line can hold models with different specs [6][7]. We name no winner and quote no prices; both depend on the model number and change by the day. For failure statistics, read our guide on Backblaze Drive Stats. For what protects your data when a drive dies anyway, read the 3-2-1 backup guide. Both are linked under "Related pages".

How it works

1. A line name is not a spec. Manufacturers sell lines such as WD Red Plus, Seagate IronWolf Pro or Toshiba N300, and each line holds many model numbers. The specs belong to the model. Seagate's IronWolf Pro 20 TB data sheet shows two models side by side. ST20000NT001 is rated for 550 TB per year, 2.5 million hours MTBF and an unlimited number of bays. ST20000NE000 is rated for 300 TB per year, 1.2 million hours and up to 24 bays [7]. Seagate's IronWolf sheet lists two 8 TB models with different specs as well: one spins at 7,200 rpm and specifies at most 1 unrecoverable read error per 10^15 bits read, the other spins at 5,400 rpm and specifies 1 per 10^14 [6]. Same line, same capacity, different promise. So the first habit is to write down the model number, not the line.
2. CMR or SMR. In conventional magnetic recording (CMR) the tracks on the platter do not overlap. In shingled magnetic recording (SMR) they overlap like roof shingles, which packs more data onto a platter but makes overwriting data in the middle of a zone slower. Device-managed SMR (DM-SMR) hides this inside the drive, which does its housekeeping when it is idle [1][11]. Western Digital says its WD Red Plus and WD Red Pro are CMR and "should be used with ZFS systems", while WD Red (SMR) targets light small-office workloads. WD also says that a DM-SMR drive rebuilds a RAID "the same as a CMR drive, or slightly slower", and that ZFS resilvering leaves no idle time for the drive's background work [1]. In 2020 WD acknowledged that its 2 TB to 6 TB WD Red drives were DM-SMR, and it conceded that "in some cases" DM-SMR "can result in slower rebuild times" [10]. Its WD Red product brief today lists SMR for the models WD20EFAX to WD60EFAX, with a footnote that this is device-managed SMR [5]. The TrueNAS hardware guide, dated September 2026, is blunter: DM-SMR drives "function poorly" with TrueNAS, and you should expect degraded write performance and longer, less reliable resilver operations [11]. Both sides are vendors talking about their own platforms, and we found no independent measurement in our research. Our assessment: if you ever plan to rebuild, expand or resilver an array, take CMR and treat SMR as a cost you accept knowingly. Today's NAS lines are CMR according to their own data sheets: WD Red Plus [3], WD Red Pro [4], IronWolf and IronWolf Pro [2][6][7], Toshiba N300 [8]. SMR still exists nearby. Seagate's list shows BarraCuda 3.5-inch drives with 4 TB and 8 TB as SMR, and the very large Exos models (32, 36 and 44 TB) as SMR [2]. Drive Stats tables have no data on SMR drives either, as our Backblaze guide explains. Lines change, so check the model number against the list on the day you buy.
3. Workload rating. The rating is the amount of data the drive may transfer per year. WD defines it as user data transferred to or from the drive, annualized: TB transferred × (8760 ÷ power-on hours) [3]. Toshiba counts data "written, read or verified" and says each drive "is designed to perform up to" its rating, after which it "may be expected to decline", without changing the warranty [9]. The ratings in the data sheets we read: WD Red Plus 180 TB per year [3], WD Red Pro 550 [4], IronWolf 180 [6], IronWolf Pro 550 or 300 depending on the model [7], Toshiba N300 180 [8]. Two points get overlooked. First, reads count. By the vendors' definitions a scrub or a rebuild looks like any other read to the drive, and we found no vendor statement that excludes them. Synology recommends scrubbing periodically and gives "monthly" and "every six months" as schedule examples [17]. If a scrub reads a whole 16 TB drive, monthly scrubs alone add 192 TB per year and six-monthly scrubs 32 TB. That is our arithmetic, and whether your NAS reads the whole drive or only used space depends on its file system and RAID mode. Second, you can measure your own number. On Linux, smartctl -x /dev/sda prints the ATA device statistics log with "Power-on Hours", "Logical Sectors Written" and "Logical Sectors Read" [18]. Multiply the sectors by the logical sector size (512 bytes on most current drives, shown by smartctl -i), add both, divide by the power-on hours and multiply by 8,760. The device statistics log exists only on some recent devices [18], and NAS firmware may hide the raw drive.
4. Bays and vibration. Spinning platters and seeking heads shake the neighbours. Rotational vibration (RV) sensors let a drive anticipate that disturbance and counteract it [4]. Data sheets give a number of bays too. In the sheets we read: WD Red Plus "up to 8 bays" [3], WD Red Pro "unlimited" [4], IronWolf 1 to 8 bays [6], IronWolf Pro unlimited or up to 24 depending on the model [7], Toshiba N300 up to 8 in the European data sheet from March 2025 [8], but "up to 12" on Toshiba's US product page [9]. When two pages from one maker disagree, ask the maker, or pick a model whose sheet is clear. RV sensors: the Red Pro sheet lists them for every model [4], the IronWolf sheet lists them for the models it covers [6], the N300 has them [8], and the Red Plus sheet has no RV row at all [3]. Absence in a sheet is not proof of absence in the drive, but you cannot build a purchase on it. Our assessment: match the bay rating to the bays you fill, and treat the bay rating as the vendor's design limit.
5. Compatibility lists. Your NAS maker tests drives and publishes a list, and the list works by model number. QNAP says its customer service cannot give technical support for devices not on its list, and it advises updating the drive firmware before installation [13]. QNAP tests include cold and warm starts, I/O stress, thermal measurements, SMART and a hot-plug and RAID rebuild test, and it removes drives when a firmware or hardware change in production breaks compatibility [14]. Synology's rules depend on the model. For models from 2025 on, running DSM 7.3, HDDs on the compatibility list work everywhere. Unlisted HDDs can be used for new storage pools in DS Plus and desktop FS models, while several other series only accept them as migrated drives, and drives on the incompatibility list are refused [12]. The policy does not apply retroactively to older models [12]. WD sends its own customers to the NAS makers' lists for the final word [1]. Open the list for your exact NAS model, search the drive's model number, and look at the date and the firmware column where there is one.
6. Capacity against rebuild time. When a drive fails, the array runs without that protection until the replacement is rebuilt. Synology gives a formula: estimated time = capacity of one drive ÷ speed of one drive. For a repair it names 125 MB/s as the ideal HDD speed when the NAS runs nothing else, and 10 MB/s while it runs other services. Synology counts in binary units, so a drive that DSM shows as 10 TB is about 10,485,760 MB [15]. A 16 TB drive is about 15,258,789 MB in that notation. By Synology's formula, our arithmetic gives 33.9 hours at 125 MB/s and 17.7 days at 10 MB/s. The second figure is Synology's conservative floor for a busy NAS, not a prediction. A second risk sits in the data sheets. They state a maximum unrecoverable read error rate, 1 per 10^14 bits for some models and 1 per 10^15 for others [3][4][6]. Rebuilding a four-drive RAID 5 of 16 TB drives reads the three surviving drives, 48 TB or 3.84 × 10^14 bits. If errors occurred at exactly the data sheet maximum, the chance of at least one is about 32 percent at 1 per 10^15 and about 98 percent at 1 per 10^14 (Poisson estimate, our arithmetic). The sheet gives a ceiling, not a measured rate, and we found no primary source for how far real drives stay below it. Read the number as a reason to ask, not a forecast. Synology's guidance points the same way: RAID 5 tolerates one failed drive and is "ideal" up to seven drives, RAID 6 tolerates two and suits "large storage arrays where the chance of a second failure is higher" [16]. Two-parity levels cost capacity, and none of them replaces a backup.
7. Warranty. Data sheets list 3 years for WD Red Plus [3], 5 years for WD Red Pro [4] and IronWolf Pro [7], 3 years for IronWolf [6] and for the N300 [8]. Seagate adds three years of "Rescue Data Recovery Services" to IronWolf and IronWolf Pro, and notes they are not available in all countries [6][7]. WD points to a regional warranty page [3]. Read your own country's terms before you count on any of it. Our assessment: a warranty is a promise to replace a drive, not to bring your data back, and a longer warranty is a tiebreaker, not a substitute for a backup.
8. Noise and power. Data sheets list noise in dBA or dB and power in watts. Compare within one sheet only, because test methods differ and your enclosure, mounting and room change what you hear. Spreads inside one line are large. In the WD Red Pro data sheet, the 10 to 16 TB models list idle noise between 20 and 34 dBA, and the air-filled models list the higher value, with idle power between 2.8 and 6.1 W [4]. Toshiba lists 20 dB at idle for its 12 to 22 TB N300 and 34 dB for the 4 to 10 TB models [8]. To turn watts into energy, use kWh per year = watts × 8,760 ÷ 1,000. A drive that averages 3 W uses about 26 kWh a year, one that averages 8 W about 70 kWh. Multiply by the price per kWh on your own bill; we quote no tariff. Four drives are four times that. Data sheet idle values are lab values at room temperature, so measure your own box as described in our guides on homelab energy costs and power measurement.

A worked example

This is a made-up setup, not a report on our systems. We have no documented drive purchases to share. Assume a four-bay NAS that you want to fill with four 16 TB drives in RAID 5, and three candidate lines.

Step 1, find the model numbers. On a machine you control, lsblk -d -o NAME,MODEL,SERIAL,ROTA lists them, and sudo smartctl -i /dev/sda prints model, firmware and sector sizes [18]. For drives you have not bought yet, take the part number from the shop page and open the manufacturer's data sheet for exactly that number.

Step 2, run the filters. CMR: the sheet or Seagate's list must say so [2]. Bays: four bays is within the rating of every line we listed. Workload: assume 40 TB of file traffic a year across the array plus a scrub every three months over 10 TB of used space per drive. That is 40 TB of scrub reads per drive plus roughly a quarter of the file traffic, far below 180 TB, so in this setup the rating does not decide. A busy video-editing share would change that.

Step 3, compatibility. Search each model number on your NAS maker's list [12][13]. Drop what is missing.

Step 4, rebuild. At 125 MB/s a 16 TB drive takes about 34 hours, at 10 MB/s about 18 days (Synology's formula [15]). Check the data sheets' read error rate for your capacity. If both numbers make you uneasy, price a RAID 6 or SHR-2 layout against the capacity you lose [16], and make sure a tested backup exists first.

Step 5, tiebreakers. Compare idle watts and idle dBA between the exact models, convert watts to kWh with your own tariff, and note warranty length and country terms.

Checklist

• Write down the exact model number of every drive you consider, and open the manufacturer's data sheet for that number, not for the line.
• Confirm that the sheet or the manufacturer's list says CMR. If neither says it, skip the drive or ask the maker in writing.
• Compare the workload rating with your own yearly transfer, and include scrubs and rebuilds. Measure it with smartctl -x if your drive supports the device statistics log.
• Match the bay rating to the number of bays you fill, and find the RV-sensor row in the sheet.
• Search the model number on your NAS maker's compatibility list, and note the list's date and the firmware column.
• Calculate the rebuild time with Synology's formula for your capacity, and decide the RAID level before you buy.
• Read the unrecoverable read error rate for your capacity and treat it as a reason for a second parity drive and a backup, not as a forecast.
• Check warranty length, regional terms and whether data recovery is offered in your country.
• Compare idle power and idle noise between the exact models, and convert watts to kWh with the tariff on your bill.
• Compare the model number on the delivered label with your order, record smartctl values on arrival, and start the long self-test as described in our Drive Stats guide.