How the RAID calculator works
You have a pile of drives and a question with two halves: how much space do I actually get, and what happens when one dies. Those halves fight each other, which is the whole story of RAID. Every level is a different settlement between capacity and safety, and the arithmetic for each one is simple enough to do on a napkin, but the napkin will not tell you that a second failure during a rebuild loses everything, or that your 8 TB drive counts as 4 TB next to a smaller partner. This page does the napkin math and then tells you the parts the napkin leaves out.
Enter your drives and pick a level for the full result, but the comparison table at the bottom of every answer may be the more useful half: all six arrangements at your exact drive count, side by side, with usable space, efficiency, and what each one survives. We cover the levels people actually build in 2026: RAID 0, 1, 5, 6, 10, and JBOD. RAID 2, 3, and 4 exist the way steam cars exist: real, historically important, and not something you should buy. Nested levels like RAID 50 and 60 (stripes built out of parity arrays) are honest tools on big controllers, and honestly out of this page's scope; the per-array math here still governs each leg of them.
The formula
RAID 1: S usable regardless of N, survives N − 1 (minimum 2)
RAID 5: (N − 1) × S, survives 1 (minimum 3)
RAID 6: (N − 2) × S, survives any 2 (minimum 4)
RAID 10: (N ÷ 2) × S, survives 1 guaranteed, up to N ÷ 2 (minimum 4, even N)
JBOD: sum of all drives, survives 0 (mixed sizes fully used)
N is the number of drives and S is the size of the smallest drive, because standard RAID sizes every member down to it. TiB conversion: 1 marketed TB is 1012 bytes, which is 1012 ÷ 240, about 0.9095 TiB. Rebuild exposure for the parity levels: replacing a failed drive reads (N − 1) × S of surviving data, and the expected number of unrecoverable read errors in that read is its size in bits divided by the drive's error-rate spec.
Worked example
Six 12 TB drives in RAID 6. Raw disk: 6 × 12 = 72 TB. Usable: (6 − 2) × 12 = 48 TB, which your operating system will report as about 43.66 TiB. Efficiency: 66.7%; two drives' worth of space holds the double parity. Fault tolerance: any two drives can fail, in any order, and the data survives. Rebuilding one failed drive reads the 5 survivors in full, 60 TB or 480 trillion bits: about 4.8 expected unreadable sectors at the consumer error spec, or 0.48 at the enterprise spec, and either way the second parity is there to absorb them.
The same six drives, every other way: RAID 0 gives 72 TB (100%) and survives nothing. RAID 1 gives 12 TB (16.7%) and survives five failures. RAID 5 gives 60 TB (83.3%) and survives one, with the rebuild as its danger window. RAID 10 gives 36 TB (50%), guaranteed to survive one failure and up to three if they land in different mirror pairs. JBOD gives 72 TB (100%) and survives nothing. RAID 6 sits in the middle of that table on space, and at the top of it on sleep quality.
What each level actually does
RAID 0 stripes. Data is chopped into chunks and dealt across the drives like cards, so reads and writes hit all spindles at once. It is fast and it wastes nothing, and every byte exists in exactly one place, so one dead drive destroys the whole deck. The zero is how much data survives a failure. Use it for scratch space you can regenerate, never for anything you would miss.
RAID 1 mirrors. Every drive holds a complete, identical copy. Usable space is one drive's worth no matter how many mirrors you stack, and the array survives until the last copy dies. Recovery is the simplest of any level: the surviving drive is just a normal drive with your data on it.
RAID 5 and 6 use parity, and parity is just arithmetic: XOR all the data blocks in a stripe together and store the result, so any one missing block can be recomputed from the ones still standing. RAID 5 keeps one parity block per stripe, spending one drive's worth of space to survive one failure. RAID 6 keeps two, computed differently from each other, spending two drives' worth to survive any two failures. That second parity is not paranoia; it is what turns an unreadable sector during a rebuild from a disaster into a log entry.
RAID 10 stripes across mirrors. Drives are paired into mirrors first, then a stripe runs across the pairs: RAID 1 for safety, RAID 0 for speed, half your raw space for the privilege. Rebuilds are fast and gentle because they copy one partner drive instead of reading the whole array. The fine print is in the pairing: a second failure in a different pair is survivable, but losing both halves of one pair loses everything.
JBOD spans. No stripes, no parity, no arithmetic at all: the drives are glued end to end into one big volume. It has exactly one real virtue, and it is a genuine one: mixed sizes are fully used, every terabyte of them. It also has RAID 0's failure mode without RAID 0's speed, since one dead drive typically takes the volume with it.
Mixed drive sizes: the smallest drive is the boss
Standard RAID needs every member to contribute the same amount of space, so it quietly sizes the whole array to the smallest drive. A 4 TB and an 8 TB drive in a mirror make a 4 TB mirror; the extra 4 TB is decoration. Six drives where the smallest is 6 TB build a RAID 6 of six 6 TB members: (6 − 2) × 6 = 24 TB usable, no matter that the other five are 12s. If your drives are not all the same size, put the smallest one's size in the optional field and the calculator will do the math honestly and name the waste. The two ways out: upgrade the smallest drive (most arrays will then grow into the new floor), or use JBOD, the one arrangement that wastes nothing and protects nothing.
The RAID 5 rebuild question, answered honestly
In 2007 a widely shared article predicted that RAID 5 would "stop working" in 2009, and the arithmetic behind it was real. When a drive fails, rebuilding the replacement means reading every bit of every surviving drive, and consumer drives are specced at one unrecoverable read error (URE) per 1014 bits, which is only 12.5 marketed TB of reading. Take eight 12 TB consumer drives in RAID 5: the rebuild reads 7 × 12 = 84 TB, which is 672 trillion bits, and at that spec the read expects about 6.72 unreadable sectors. On paper, the rebuild is nearly certain to hit at least one, and on a classic RAID 5 controller one unreadable sector is enough to fail the whole rebuild.
Now the other direction, because the argument is often quoted with the honesty left out. The spec is a worst-case-flavored floor, not a measurement: real drives routinely run far better than their published rate, and modern controllers and filesystems handle a bad sector more gracefully than the 2009-era assumption of instant death. Swap the same math onto enterprise drives at one URE per 1015 bits and the 84 TB rebuild expects about 0.67 errors: roughly a coin flip of a clean pass on paper, and much better than that in practice. So RAID 5 is not dead. But the shape of the conclusion survived: big arrays of big consumer drives on single parity are a genuinely bad idea, and that arithmetic, not fashion, is why RAID 6 and mirrors won the last fifteen years of storage design.
A note on hot spares, because they are often sold as the fix: a spare drive standing by shortens the danger window, the days between failure and rebuild that would otherwise be spent waiting for a delivery. That is worth having. But the spare changes nothing about the rebuild itself: the same terabytes get read, the same errors are expected. A hot spare gets you to the lottery drawing faster; it does not change the odds.
RAID is not a backup, and the 3-2-1 rule
This is the sentence the result box repeats on purpose: RAID is availability, not backup. It exists so a dead drive does not take your Saturday, your file server, or your business offline. It does nothing about the failures that actually claim most data: you delete the wrong folder and the array deletes it everywhere instantly; ransomware encrypts your files and the array mirrors the encryption with perfect fidelity; the controller corrupts data and the corruption is written redundantly; a power surge, a flood, or a thief takes all the drives at once, parity included. Against every one of those, a six-drive RAID 6 offers exactly the same protection as a single drive from a clearance bin: none. The fix is boring and beats every RAID level ever designed: the 3-2-1 rule. Three copies of anything you care about, on two different kinds of media, one of them somewhere else. The array can be one of the three. It cannot be all of them.
Decimal terabytes, binary tebibytes, and where your space went
A marketed 12 TB drive holds 12 × 1012 bytes, and the label is honest: that is what tera means. But operating systems count in binary units, where a tebibyte (TiB) is 240 bytes, about 10% bigger than a decimal terabyte, so the same drive reports as about 10.91 TiB. Windows then labels that binary number "TB" anyway, which is how a 12 TB drive comes to show 10.9 and a support forum comes to exist. Nothing was stolen; two measuring sticks are in the room. This calculator gives usable capacity in both, so the number on the box and the number in your file manager can finally be introduced. Filesystem overhead (metadata, journals, reserved blocks) shaves a further few percent off, varies by filesystem, and is not modeled here.