Best NAS Hard Drives: CMR Picks Compared Per Terabyte

The enclosure is the cheap half of a NAS purchase. Drives are where the money goes. Eight drives at $200 to $500 each is $2,000 to $4,000 in a single order, against roughly $1,300 for the eight-bay box you put them in. Then they fail, and get upgraded, and you buy them again. Two rules decide most of the outcome: buy conventional recording, never shingled, and compare drives on price per terabyte rather than sticker price.

Below are three picks at genuinely different capacities, then the full field of ten verified drives ranked by what each terabyte actually costs. Every drive named on this page uses conventional magnetic recording, because a shingled drive does not belong in a parity group at any price.

Best NAS hard drives by build size

Beginner Smallest sensible NAS drive

Seagate

Seagate IronWolf 4TB NAS HDD (ST4000VN008)

$219.00

CMR recording, rotational vibration sensing and a NAS-rated workload figure. Below 4 TB the price per terabyte stops making sense.

Best for: A two-bay mirror where the library is documents and photos rather than video.

Check price

Intermediate Best capacity per dollar

Western Digital

WD Red Plus 8TB NAS HDD (WD80EFPX)

$353.99

Red Plus is the CMR line, which matters because SMR drives can take days to rebuild an array. 5,640 rpm also runs cooler and quieter than the Pro tier.

Best for: The default four-bay build. Buy all four at once so they age together.

Check price

Expert High capacity

Western Digital

WD Red Pro 20TB NAS HDD (WD202KFGX)

$836.99

Rated for 24-bay chassis and a 550 TB per year workload. The catch is rebuild time: a 20 TB member takes well over a day to resilver.

Best for: Large arrays running double parity. Do not run 20 TB drives on single parity.

Check price

Prices change often, confirm on Amazon. As an Amazon Associate we earn from qualifying purchases.

Why do the drives matter more than the box you put them in?

A four-bay enclosure is a one-time few hundred dollars. Four WD Red Plus 8TB drives at $353.99 each is $1,415.96, and that number repeats: drives are consumable, and a NAS bought today will be on its second set of spindles inside about six or seven years. An eight-bay build is worse, or better, depending on how you look at it: 8 x $250 is $2,000 at the cheap end and 8 x $500 is $4,000 at the top, against a box that lists around $1,300.

The practical consequence is that saving $150 on the enclosure and then buying worse drives is backwards. Spend the enclosure budget on the cheapest box with the right number of bays and the right processor, and put every remaining dollar into capacity and recording technology. Run your intended drive set through the RAID capacity calculator before you order, because the layout you choose changes how many of those drives you actually get to use. If you are assembling a whole system rather than replacing drives, the intermediate home server setup lists a complete build with the drive spend already accounted for.

CMR against SMR: the one specification that can cost you the array

Conventional magnetic recording (CMR) lays tracks side by side, so any sector can be overwritten in place. Shingled magnetic recording (SMR) overlaps tracks like roof tiles to squeeze more data onto the same platters. Overlapping means a write into an already-written region cannot happen in place: the drive must read the surrounding band, modify it and write the whole band back.

On a desktop that shows up as an occasional stutter, because the drive has a conventional cache region it can absorb bursts into and flush later while idle. Inside a RAID array that escape route closes. A rebuild writes to the replacement member continuously for hours with no idle window to flush in, so the cache fills and every subsequent write pays the full band-rewrite cost. Rebuilds that should take a day have been reported to run for several, and some controllers time out and drop the drive mid-resilver, which on single parity means you have lost the pool.

Never put an SMR drive in a RAID array. WD Red Plus and WD Red Pro are CMR. The plain WD Red line included SMR models. Seagate IronWolf and IronWolf Pro are CMR. Toshiba N300 is CMR. If the listing does not state the recording technology, look up the exact model number on the manufacturer specification sheet before you buy.

The trap is that SMR drives are not labelled loudly, and they are cheapest per terabyte, so a price-sorted search puts them at the top. That is the specific reason this page ranks by price per terabyte within a CMR-only field. A cheaper number that disqualifies itself on recording technology is not a better deal, it is a different product.

Which NAS drive is cheapest per terabyte?

This is the table the marketing pages will not show you, because the ranking it produces is not the ranking the brands want. Price per terabyte is the sticker price divided by the labelled decimal capacity, computed from the same verified prices used on the cards above. Every model name links to the current listing.

Prices are the verified listing prices and move constantly. Capacity is decimal, as printed on the label. Your NAS will report every one of these figures about 9.05 percent smaller because it counts in binary tebibytes. See the TB against TiB chart for the conversion.

Two things fall out of that ranking. The 8 TB and 16 TB models sit at the bottom of the per-terabyte column, around $36 per terabyte, and 4 TB is the worst value in the field at $55 to $62 per terabyte. Buying the smallest drive is almost never the cheap decision, it is the decision that fills your bays with capacity you will want to replace.

Work a concrete example. A four-bay array of 4 TB IronWolf drives costs $876.00 and gives 12 TB usable under single parity. The same four bays filled with 8 TB IronWolf drives costs $1,159.96 and gives 24 TB usable. You pay 32 percent more and get exactly double the storage, in the same chassis, drawing roughly the same power. The 4 TB build also leaves you with no upgrade path except replacing all four drives, whereas the 8 TB build has years of headroom.

How long does each drive size take to rebuild?

Resilvering is the process of reconstructing a failed member from parity onto a replacement drive. It is the most stressful operation an array ever performs, because it reads every sector of every surviving drive while writing continuously to the new one. Rebuild time scales directly with capacity, and it is the reason large drives change your layout decision.

Rates are sustained throughput for the rebuild itself. A NAS that is also serving files during the rebuild will be slower, sometimes considerably, and consumer units often throttle resilver speed deliberately to keep the array responsive.

The pattern is stark. An 8 TB member finishes in roughly 15 to 28 hours, which is a long night. A 20 TB Red Pro takes 37 to 70 hours, which is a long weekend, and a 24 TB member can approach 83 hours at the slow end. Under single parity the array is completely unprotected for that entire window. Past roughly 8 to 10 TB per drive, use double parity: RAID 6, RAIDZ2 or SHR-2 all keep protecting the array while the first rebuild runs. Our RAID 5 against RAID 6 comparison works through what that second parity drive costs you in capacity.

What does a workload rating in terabytes per year actually buy?

The workload rate is the annual volume of reads and writes the manufacturer designed the drive around and will honour under warranty. NAS-tier drives publish 180 TB per year. Pro and enterprise tiers publish 300 or 550 TB per year.

180 TB per year is about 500 GB every day, every day. A household NAS serving Plex, running photo sync and taking nightly backups does not come close: streaming a 4K film reads perhaps 60 GB, and an incremental backup writes a few gigabytes. The rating starts to matter when the array runs continuous surveillance recording, when you scrub the full pool weekly, or when the NAS is also a seedbox.

Run the arithmetic on a weekly scrub, because it is the one home workload that genuinely accumulates. A 48 TB array scrubbed every week reads 48 TB per pass, which is 2,496 TB per year spread across eight drives, so 312 TB per drive annually. That exceeds a 180 TB rating. Scrub monthly instead and it falls to 72 TB per drive, comfortably inside spec. The scrub schedule, not the drive tier, is what needed changing.

Error recovery time, the feature nobody advertises

Time-limited error recovery, sold as TLER by Western Digital and ERC by Seagate, caps how long a drive will keep retrying a sector it cannot read. NAS drives stop at around seven seconds and report the error. Desktop drives keep trying, sometimes for up to two minutes, because on a single-drive machine that stubbornness is the only chance of getting your data back.

In an array it is exactly backwards. The controller has parity and can reconstruct that sector instantly, so it does not need the drive to succeed. What it cannot tolerate is a member that stops responding for two minutes, so it marks the drive failed and ejects it. The drive is perfectly healthy. You now have a degraded array, and if a second desktop drive does the same thing during the rebuild, a single-parity array is gone.

This is the concrete answer to "can I just use cheaper desktop drives". You can, and the failure mode you are buying is an array that drops healthy members under stress. It is also why shucked external drives need checking rather than assuming: the drive inside is sometimes an excellent white-label enterprise unit and sometimes not, and the warranty position differs from a bare retail drive.

Rotational vibration sensing, and when you actually need it

Every spinning drive vibrates, and in a multi-bay chassis those vibrations travel through the backplane into every neighbour. Rotational vibration sensing puts accelerometers in the drive so the head servo can correct for movement that did not come from its own platters. Without it, heads spend longer settling, throughput drops and error rates climb.

In a two-bay box it is close to irrelevant. From roughly four bays up it becomes worth paying for, and in an eight-bay chassis with seven neighbours all seeking at once it is a genuine requirement. That is the honest reason the Pro tiers exist: the IronWolf Pro 16TB and the Red Pro line are rated for chassis of up to 24 bays, and the smaller NAS lines are not. If you are filling an eight-bay chassis, treat vibration sensing as mandatory rather than optional.

Two related specifications are worth a glance while you are there. Drives at 8 TB and above are commonly helium sealed, which cuts aerodynamic drag so they run cooler and draw less power than air-filled drives spinning at the same speed. And rotational speed itself is a trade rather than a ranking: the 5,640 rpm Red Plus runs quieter and cooler than a 7,200 rpm drive, while the faster drive rebuilds sooner. In a cupboard, pick quiet. In a basement rack, pick fast.

Warranty length is a proxy for how the drive was binned

Three years against five years is not just a support term. It reflects where in the qualification process the drive landed, and it changes the arithmetic on a large array. Drive annualised failure rates in large fleets run about 1.0 to 1.6 percent per year for modern 3.5 inch drives, rising past roughly five years, and the probability that at least one drive in a set fails follows 1 - (1 - afr)^(n x y).

Calculated at a 1.3 percent annualised failure rate, mid-range for modern 3.5 inch drives. Put your own drive count and horizon into the drive failure probability calculator for the exact figures.

Read the eight-drive row carefully. Over five years there is roughly a two in five chance that at least one member dies, and that is the ordinary case rather than bad luck. Under a three-year warranty, a failure in year four costs you the full replacement price. Under five years it costs you shipping. On a large array, that difference is most of the premium the Pro tiers charge, and it is why the Seagate Exos 20TB is interesting: it carries the same five-year term and 550 TB per year workload rating as a Red Pro at a lower price per terabyte, trading noise and idle power for the saving.

Why buy in two batches, from two retailers?

Buy the same model, at the same time, but not from the same order. Drives that roll off one production run share whatever that run got wrong, and drives that go into service on the same day accumulate the same hours. Both effects push failures toward each other in time, which is exactly the wrong shape: the moment you least want a second drive to die is during the rebuild caused by the first.

The failure model above assumes drives fail independently. Correlated drives break that assumption, and correlation is precisely what one manufacturing lot gives you. If a batch has a marginal head or a firmware bug, it does not affect one of your eight drives, it affects the batch. That converts an independent 40.7 percent chance of a single failure into a meaningful chance of several failures inside the same week.

Splitting the order across two retailers, or across two orders a week apart, usually lands you two different manufacturing lots for the same money. It costs nothing but a second checkout. Then burn them in before you trust them: our drive burn-in testing guide covers the full surface read and write pass that catches infant mortality while the array is still empty and the drives are still inside the return window. Finding a bad drive on day three is an inconvenience. Finding it in year two, during a rebuild, is a data loss event.

None of this makes the array a backup. RAID protects against a drive dying. It does nothing about accidental deletion, ransomware, a controller writing garbage to every member, theft, fire or flood, because all of those reach every drive at once. Size a real second and third copy with the 3-2-1 backup sizing calculator.

Who should not buy the Expert pick

The WD Red Pro 20TB is a genuinely excellent drive and most people reading this should not buy it. At $836.99 it is $41.85 per terabyte, which is worse value than the 8 TB and 16 TB models, and a single 20 TB member takes roughly 37 to 70 hours to resilver. That long rebuild window is only acceptable under double parity, so buying 20 TB drives commits you to RAID 6, RAIDZ2 or SHR-2, which costs you a second drive of capacity on top of the price.

Concretely: four 20 TB drives is $3,347.96 and gives 40 TB usable under double parity. Six 8 TB Red Plus drives is $2,123.94 and gives 32 TB usable under the same protection, with rebuilds that finish overnight instead of over a weekend. The 20 TB build costs $1,224 more for 8 TB more usable capacity, which works out at $153 per additional usable terabyte against a field average nearer $45.

Unless your bay count is genuinely the constraint, put the money into more medium drives rather than fewer enormous ones. Buy the 20 TB class when you have run out of bays and cannot add a chassis, when the chassis is in a rack where noise does not matter, and when you have already committed to double parity. Buy it for a two-bay NAS and you have spent $1,674 to get 20 TB usable with a rebuild window measured in days, which is the worst combination available at any price.

The same caution applies in the other direction. The WD Red Plus 4TB at $61.99 per terabyte is the most expensive storage in this field. It is the right answer only when the array is genuinely small, such as a two-bay unit holding documents and photos, and even then the Seagate IronWolf 4TB does the same job for $28.98 less per drive.

Related reading

• How to choose NAS drives for the full decision walkthrough, capacity by capacity
• NAS drive specification chart for the raw numbers without the commentary
• Drive reliability by model for published fleet failure statistics
• Intermediate home server setup for a complete build with these drives in it
• Best 8-bay NAS if you are sizing the chassis these drives go into

Deciding on a specific model? We review the Western Digital Red Pro 24TB NAS hard drive review, Seagate Exos 20TB enterprise hard drive review and Seagate IronWolf Pro 16TB review in full.