Compare RAID levels
Capacity patterns use the smallest drive inside each striped or redundant group. The write column is the classic backend operation count for one small random write. It is not an IOPS forecast.
| Layout | Minimum drives | Theoretical usable | Guaranteed tolerance | Small random write |
|---|---|---|---|---|
| RAID 0 | 2 | N × smallest | None | 1 |
| RAID 1 | 2 | Smallest | Any N - 1 | N mirror writes |
| RAID 5 | 3 | (N - 1) × smallest | Any 1 | 4 |
| RAID 6 | 4 | (N - 2) × smallest | Any 2 | 6 |
| RAID 10 | 4, even | Sum of mirror pair minima | Any 1 | 2 |
| RAID 50 | 6 | Sum of RAID 5 groups | Any 1 | 4 per group |
| RAID 60 | 8 | Sum of RAID 6 groups | Any 2 | 6 per group |
| Linear | 1 | Sum of all drives | None | 1 |
| JBOD | 1 | Sum of all drives | None in this model | 1 |
The smallest drive sets the size of every group
Striped and parity layouts write fixed-width stripes across every drive at the same offset, so each drive can contribute only as much as the smallest one in its group holds. Pair a 4 TB drive with an 8 TB drive in a mirror and you get 4 TB of protected space. The other 4 TB stays visible above as size mismatch because it is real capacity you paid for and could recover by changing the layout or the drives.
Linear and JBOD are the exceptions. They concatenate members rather than striping across them, so every supplied byte counts.
Nested layouts apply the rule per group, not across the whole array. In RAID 10 each mirror pair is capped by its own smaller drive, so the order you enter drives changes the answer. For four drives of 4, 4, 8, and 8 TB, pairing 4 + 4 and 8 + 8 yields 12 TB; pairing each 4 TB member with an 8 TB member yields 8 TB. RAID 50 and RAID 60 behave the same way inside each parity group.
Linux md can use unequal devices in RAID 0 with layout-specific remainder zones, and that behavior has changed across kernel versions. This calculator uses the portable smallest-member model rather than emulating those layouts.
Failure tolerance belongs to the topology, not to one number
A single tolerance number hides the thing that matters in nested layouts. RAID 10 always survives one drive failure. It may survive one failure in every mirror pair, which on eight drives means four simultaneous failures. It never survives both members of the same pair. Those situations differ entirely because of which drives fail.
RAID 50 and RAID 60 work the same way one level up. Each RAID 5 group tolerates one failure, each RAID 6 group tolerates two, and the stripe across groups tolerates the loss of no complete group. Lose one group and the array is gone.
That is why the topology above is part of the result rather than decoration. A capacity figure without the grouping is only half the answer.
The number stops before the filesystem starts
This is array-level theoretical capacity. It is the size of the block device the array presents before anything is written to it.
Not included: array metadata, partition alignment, filesystem metadata, reserved blocks, snapshots, and application-level allocation. On a NAS appliance, also account for the system partition and swap the operating system takes from every drive.
Also not included: rebuild time, IOPS, throughput, or the probability of data loss. Those depend on hardware, workload, and drive health. A capacity calculator that invents them would be guessing.
Drives are sold in decimal terabytes, where 1 TB is 1,000,000,000,000 bytes. Operating systems often report binary tebibytes, where 1 TiB is 1,099,511,627,776 bytes. That is where roughly 9% of the numerical difference appears before RAID takes any capacity. Use the result unit control to see both.
What should I verify before choosing a RAID layout?
Check that the intended controller or operating system supports the exact layout and grouping. A portable capacity formula doesn't reproduce every vendor's allocation or recovery behavior. Linux MD documentation is one implementation reference, not a promise that another controller behaves identically.
In this calculator, Linear and JBOD both add the supplied capacities without array-level redundancy. Some vendors use JBOD to mean individually exposed disks instead of concatenation. Verify the meaning before assuming one combined volume.
For large parity arrays, consider the degraded period as well as the initial capacity. The rebuild calculator lets you examine a read-volume and time scenario; it doesn't predict whether your rebuild will succeed.
RAID is not a backup. It provides redundancy for specific drive failures, not protection from deletion, ransomware, theft, or every form of corruption. Keep an independent recovery copy and test that it restores.
Does this send my drive list anywhere?
The calculation runs in your browser without a separate calculation request. After you change an input, the versioned state is placed in the URL so the layout can be restored. Anyone you share that URL with can read the encoded drive sizes.