RAID capacity explained: why your NAS shows less space than the drives promise

You buy four 8TB drives, drop them into a NAS, expect somewhere close to 32TB, and the array reports 21TB usable — sometimes less once the operating system finishes rounding. Nothing is broken. Two separate effects are stacking on top of each other: the gap between how manufacturers count a terabyte and how your OS counts one, and the capacity the RAID level itself sets aside for redundancy. Understanding both means you can size a NAS correctly the first time instead of discovering the shortfall after the drives are already installed.

Two different terabytes

Drive manufacturers count in decimal: 1 TB is 1,000,000,000,000 bytes, because base-10 multiples are easier to print on a box and, not coincidentally, produce a bigger-looking number. Operating systems count in binary: what Windows or a NAS's web UI calls a "TB" is really a tebibyte (TiB), equal to 2 to the 40th power, or 1,099,511,627,776 bytes. Divide the two and a manufacturer's terabyte is only about 0.909 of an operating system's terabyte. A drive advertised as 4TB reports as roughly 3.64 TiB before a single byte of RAID overhead is applied — and every array you build compounds that same 9-ish percent haircut on top of whatever the RAID level itself takes.

None of this is a defect. It is a decades-old naming collision that manufacturers have little incentive to fix, since it makes their drives look larger on the shelf. Budget for it: whatever number a RAID calculator gives you in TB, multiply by about 0.909 to see what the OS will actually display.

RAID levels trade capacity for fault tolerance

Redundancy is not free. Every RAID level that survives a drive failure does so by writing your data more than once, or by writing extra parity information alongside it — and both cost usable space.

RAID 0 — striping, no redundancy

RAID 0 spreads data across every drive with no duplication, so usable capacity equals the full raw total: n drives times capacity, no discount. It is also the only level with zero fault tolerance — lose one drive and the entire array is gone, because every file is split across all of them. Use it only for scratch space or cache you can afford to lose.

RAID 1 — mirroring

RAID 1 writes every byte to two drives at once. Usable capacity is one drive's worth per mirrored pair, so an array of n drives yields floor(n/2) times the single-drive capacity — with a minimum of two drives. It is simple and it survives a failed drive in the pair, but you pay for it with fifty percent of your raw capacity.

RAID 5 — single parity

RAID 5 needs at least three drives and reserves the equivalent of one drive's worth of space for parity data, spread across all the drives rather than concentrated on one. Usable capacity is (n minus 1) times the per-drive capacity, and the array survives any single drive failure. It is the most common "good default" for small NAS boxes because it gives back most of the raw capacity while still tolerating one failure.

RAID 6 — dual parity

RAID 6 needs at least four drives and reserves two drives' worth of space for parity, giving (n minus 2) times the per-drive capacity as usable. In exchange, it survives two simultaneous drive failures — which matters more than it sounds, because rebuilding a large array after one failure puts heavy read load on the remaining drives for many hours, and a second failure during that rebuild window is a genuinely common way RAID 5 arrays lose data.

RAID 10 — mirror plus stripe

RAID 10 needs an even number of drives, at least four, and mirrors pairs before striping across them. Usable capacity is (n divided by 2) times the per-drive capacity — the same fifty percent cost as RAID 1 — but you get RAID 0's performance on top. It tolerates at least one failure per mirrored pair, which is why larger RAID 10 arrays can often survive more than one dead drive, as long as no mirrored pair loses both members at once.

Worked example: four drives in RAID 5

Take four 4TB drives in RAID 5, with 10 percent set aside for filesystem and metadata overhead — a realistic default for most NAS operating systems once you count the file system journal, snapshot reserves, and reserved blocks.

• Raw capacity: 4 drives times 4TB equals 16TB.
• RAID 5 usable: (4 minus 1) times 4TB equals 12TB — one drive's worth is spent on parity.
• After 10 percent formatting overhead: 12TB times 0.9 equals 10.8TB.
• What the OS actually shows: 10.8TB times 0.909 equals about 9.82 TiB.

Three separate deductions turned a 16TB sticker into a 9.82 TiB reality: one drive lost to parity, another slice lost to formatting overhead, and a further nine percent lost to the TB-versus-TiB conversion. None of the three is hidden or unusual — they are simply invisible until you add them up, which is exactly what trips up first-time NAS buyers who size an array off the raw drive specs alone.

RAID is not a backup

It is worth saying plainly, because it is the single most common misunderstanding about redundant storage: RAID protects against a hardware failure, not against a mistake. A parity or mirror array will happily and instantly replicate a ransomware encryption pass, an accidental "delete all," a corrupted file, or a botched firmware update across every drive in the array — because that is exactly the job RAID is built to do. It also does nothing for a fire, flood, theft, or a failed RAID controller that takes the whole array down at once. RAID buys you uptime when a drive dies; it does not substitute for an actual backup copy stored on separate media, ideally somewhere else entirely.

The parity write penalty is worth knowing about too

Capacity is not the only thing RAID 5 and RAID 6 trade away. Every write to a parity array requires reading the old data and parity, computing new parity, and writing both back out — several operations for what looks like a single write from the application's side. Mirrored levels like RAID 1 and RAID 10 do not pay this penalty, since a write is just a duplicate write with no parity math involved. For a home media server this rarely matters; for a NAS handling a database, a virtual machine datastore, or heavy simultaneous small-file writes, the parity penalty is a real reason to prefer RAID 10 over RAID 5 or 6 even at the same drive count, despite the lower usable capacity.

Picking a level for your NAS

If you only need scratch space or you already back the data up elsewhere, RAID 0 maximizes capacity. For a small two-bay NAS, RAID 1 is usually the only redundant option available. For four or more bays holding data you cannot easily replace, RAID 5 is the standard starting point, and RAID 6 is worth the extra parity drive once the array gets large enough that a rebuild takes the better part of a day. RAID 10 is the pick when you need both redundancy and the read/write performance of striping, and you have the drive bays to spare for the fifty percent capacity cost.

Hot spares and rebuild time

Buying one extra drive as a hot spare is worth budgeting for on any array of five or more bays. When a drive fails, the array runs in a degraded state — still serving data, but with no remaining fault tolerance — until a replacement is rebuilt from parity. On a large RAID 5 array, that rebuild can take the better part of a day, reading every remaining drive under sustained load the whole time, which is precisely when a second drive is statistically most likely to fail. A hot spare lets the rebuild start automatically the moment a failure is detected, instead of waiting on you to notice and physically swap a drive, shrinking the window where the array is running without redundancy.

Run your own numbers before you buy drives, not after. Our NAS / RAID usable capacity calculator takes your drive count, per-drive capacity, RAID level, and a formatting-overhead percentage, and returns the raw capacity, the usable capacity, and the fault tolerance in one pass — plus the TiB figure your operating system will actually display. Size the array to what you will really get, and keep a real backup running regardless of which RAID level you choose.