Unraid Parity Drive: Size, Speed and Why Used Enterprise Works

Your Unraid parity drive must be at least as large as the largest data drive, and its speed sets the ceiling on array write speed. Buy a 7200 rpm CMR enterprise drive at the capacity you expect to grow into, add a second parity drive past six or eight disks, and burn it in before trusting it.

How Unraid parity works

Unraid does not stripe data across drives. Each data drive holds a complete filesystem on its own, and one or two dedicated parity drives hold the information needed to rebuild any single (or any two) failed drives. Parity is calculated bit by bit across the same position on every data drive, which is why the parity drive has to cover every address on the largest data drive.

This design has consequences that differ from RAID or ZFS. Drives can be different sizes. Losing more drives than you have parity loses only the data on those drives, not the whole array. And parity drives do no work on reads, so array read speed is the speed of whichever single drive holds the file. Writes are the exception, which brings us to the parity drive's job.

Size: parity must be the largest

The rule is simple and Unraid enforces it: no data drive can be larger than the smallest parity drive. If you have an 18 TB parity drive, you can add data drives of any size up to 18 TB. To add a 22 TB data drive, you first replace parity with a 22 TB or larger drive and let it rebuild.

The practical advice is to buy parity one step larger than you need today. If your data drives are 14 TB, an 18 TB parity drive costs a little more now and saves a parity rebuild later. Used enterprise drives make this affordable; an 18 TB Ultrastar HC550 or Exos X18 datacenter pull at Standard tier is often cheaper than a new 14 TB NAS drive.

Speed: parity sets the write ceiling

In Unraid's default write mode, every write to a data drive triggers four operations: read the old data block, read the old parity block, write the new data, write the new parity. The parity drive is involved in every write to every data drive, so it is the bottleneck. A slow parity drive makes the whole array slow to write, regardless of how fast the data drives are.

The alternative "reconstruct write" mode (often called turbo write) reads from all data drives at once and writes parity in one pass, which is faster but spins up every disk. Either way, parity drive throughput is the limiting number. This is the argument for a 7200 rpm enterprise drive in the parity slot even if your data drives are slower.

Drive class	Typical sustained transfer	Spindle	Parity suitability
Enterprise capacity (Exos X, Ultrastar DC HC)	Around 250 to 290 MB/s	7200 rpm	Best choice
Pro NAS (IronWolf Pro, WD Gold)	Around 240 to 270 MB/s	7200 rpm	Good
Standard NAS (IronWolf, 5400 class)	Around 180 to 210 MB/s	5400 to 5900 rpm	Works, slower writes
Shingled (SMR) drives	Collapses under sustained writes	Varies	Avoid

Transfer figures are approximate outer-track maximums; every drive slows toward its inner tracks. SMR drives belong nowhere in an Unraid array and especially not in the parity slot, because parity rebuilds are exactly the sustained random-write workload that makes shingled drives crawl.

Dual parity

Unraid supports two parity drives. With one, you survive one drive failure. With two, you survive any two. The case for dual parity grows with array size and drive capacity: rebuilding an 18 TB drive takes most of a day during which the array is reading every other drive end to end, and that is the moment a second marginal drive tends to fail.

A reasonable rule is single parity up to six data drives, dual parity beyond that or whenever drives exceed 14 TB. The second parity drive has the same size requirement as the first. It does not need to be the same model; mixing an Exos and an Ultrastar of the same capacity is fine.

Enterprise versus NAS drives for parity

NAS drives like the IronWolf Pro and WD Gold are designed for this duty and work well. Used enterprise drives are usually the better value for a parity slot for three reasons. They are 7200 rpm across the board, so there is no slow variant to accidentally buy. They carry a 550 TB per year workload rating, which parity rebuilds will never approach. And datacenter pulls price below new NAS drives at the same capacity.

The trade-offs are noise and power. Enterprise drives are louder on seeks and draw a watt or two more at idle than consumer NAS drives. In a closet server neither matters. Our post on used enterprise drives covers the reliability question in more depth, and the SATA HDD category lists current stock, since Unraid arrays are overwhelmingly SATA.

Burn in before you trust it

A new parity drive, used or new, should be tested before the array depends on it. The Unraid community has long used the preclear process, which writes zeros across the whole drive, reads it back, and reports SMART changes. Current Unraid versions can also clear a drive when it is added. The goal is the same: force every sector to be written and read once so that a weak spot shows up now rather than during a rebuild.

For an 18 TB drive, a full preclear cycle takes well over a day. If the drive arrives with a published test report showing a completed SMART extended self-test, a full-surface write from sanitization and a read-back verification, you have already had that burn-in done. Run a short SMART self-test on arrival to confirm nothing happened in shipping, check that attributes 5, 197 and 198 are still zero, and add it to the array. The Unraid documentation covers the parity assignment and rebuild steps.

FAQ

Can the parity drive be larger than every data drive?

Yes, and it is a good idea. The extra space goes unused until you add a larger data drive.

Can I use a SAS drive for parity?

Yes, with a SAS HBA. Unraid supports SAS drives fine. SAS datacenter pulls are often the cheapest route to a large, fast parity drive.

Does the parity drive wear out faster?

It sees more writes than any single data drive, since it participates in every write. Under a 550 TB per year rating that is still not a concern for home use. It does spin up more often, which is why the enterprise duty cycle rating matters.

Should parity be a different brand from the data drives?

Not required. Some people prefer it to avoid a shared firmware issue. Capacity and speed matter more than brand.

What we do at Atomic Racks

Every used drive we sell gets a SMART extended self-test, a fio benchmark and NIST SP 800-88 sanitization with read-back verification, which means every sector has been written and read before the drive ships. Hours are read from the drive and never reset, and each unit is priced into a power-on-hour tier. The QR tamper seal on each certified unit opens its public test report. 90-day warranty on standard items, up to 2 years on certified tiers, 30-day returns and free shipping to the lower 48.

Further reading: Building a 100TB+ homelab array on used enterprise drives.