RAID Levels Reference: 0, 1, 5, 6, 10, 50, 60 Compared
What each RAID level actually does, its minimum disk count, fault tolerance, and where it fits — RAID 2, 3, and 4 are skipped here since they're defined by the standard but essentially unused in practice.
Comparison at a glance
| Level | Min disks | Fault tolerance | Capacity overhead | Read speed | Write speed |
|---|---|---|---|---|---|
| RAID 0 | 2 | None | None | Fastest | Fastest |
| RAID 1 | 2 | 1 disk | 50% | Fast | Same as 1 disk |
| RAID 5 | 3 | 1 disk | 1 disk's worth | Fast (N-1) | Slower (parity) |
| RAID 6 | 4 | 2 disks | 2 disks' worth | Fast (N-2) | Slower (dual parity) |
| RAID 10 | 4 | 1 per mirrored pair | 50% | Fastest of the redundant levels | Fast |
| RAID 50 | 6 | 1 per sub-array | 2 disks' worth | Fast | Slower (parity) |
| RAID 60 | 8 | 2 per sub-array | 4 disks' worth | Fast | Slower (dual parity) |
Run your own numbers for any of these levels with the RAID Calculator.
RAID 0 — striping, no redundancy
Data is split into blocks and written across all disks in turn, so reads and writes both benefit from every disk working in parallel — an N-disk RAID 0 array reads and writes roughly N times faster than a single disk. There is zero fault tolerance: lose any one disk and the entire array's data is gone. Use it only for data you can afford to lose entirely, or where the data itself is disposable (scratch/cache volumes, some render farms). Try the RAID 0 Calculator.
RAID 1 — mirroring
Two (or more) disks hold identical copies of all data. If one fails, the other has everything, uninterrupted. Write speed is limited to that of a single disk since every write has to hit every mirror; some implementations improve read speed by splitting read requests across the mirrors. Capacity overhead is a flat 50% — two 4 TB drives in RAID 1 give you 4 TB usable, not 8 TB. Simple, and a failed array's data can be read directly off any single surviving drive without needing to "recover" anything. Try the RAID 1 Calculator.
RAID 5 — striping with single parity
Data blocks are striped across all disks like RAID 0, but one parity block per stripe row is added, distributed across the drives rather than fixed to a single "parity disk". If one drive fails, its data is reconstructed on the fly from the remaining data and parity. Read speed scales with disk count similarly to RAID 0; write speed is held back because each write requires reading and recalculating the corresponding parity block first. Usable capacity is the array minus one disk's worth. RAID 5's practical weak point is rebuild time on large modern drives — see the RAID 6 usable capacity guide for why this pushes many admins to RAID 6 instead. Try the RAID 5 Calculator.
RAID 6 — striping with dual parity
Same idea as RAID 5, but with two independent parity calculations per stripe, tolerating any two simultaneous disk failures instead of one. Costs a second disk's worth of usable capacity compared to RAID 5, and write performance is slightly worse due to the extra parity calculation — but it survives the exact failure-during-rebuild scenario that makes RAID 5 risky on large arrays. Try the RAID 6 Calculator.
RAID 10 — mirrored stripes
A RAID 0 stripe built on top of mirrored pairs (or equivalently, mirrors of striped sets, depending on implementation). Read speed scales with total disk count like RAID 0; write speed is roughly half of that, since each write updates both sides of its mirror. Capacity overhead is a flat 50% regardless of array size — unlike RAID 5/6, where overhead shrinks as a percentage the more disks you add, RAID 10's overhead stays fixed, which makes it the most expensive option to scale to very large capacities, but the fastest and simplest to rebuild of the redundant levels — a rebuild only needs to copy from one surviving mirror partner, not read the entire rest of the array. Try the RAID 10 Calculator.
RAID 50 and 60 — striped parity arrays
RAID 50 combines multiple RAID 5 sub-arrays into an outer RAID 0 stripe; RAID 60 does the same with RAID 6 sub-arrays. Both need larger minimum disk counts (6 for RAID 50, 8 for RAID 60, assuming two sub-arrays) and are typically only available on higher-end hardware controllers. They scale better to large disk counts than a single wide RAID 5/6 array would, since rebuild impact is contained to one sub-array at a time, and — with some luck — can survive more simultaneous failures than a single RAID 5/6 array, as long as the failed disks land in different sub-arrays rather than the same one. Try the RAID 50 Calculator or the RAID 60 Calculator.
A note on RAID 1E and other variants
RAID 1E extends mirroring across an odd number of disks using striping, giving mirror-like redundancy without requiring an even disk count — a niche option mostly seen on specific hardware controllers rather than in general practice. A handful of other layouts exist (RAID 5E/5EE with integrated spare space, NetApp's RAID-DP) that are vendor-specific enough to be out of scope for a general comparison like this one.
ZFS RAIDZ1/2/3 — not hardware RAID, but the same tradeoff shape
ZFS implements its own parity scheme — RAIDZ1 (single parity), RAIDZ2 (double), RAIDZ3 (triple) — with usable capacity following the same N-minus-parity math as RAID 5/6/7-parity. The key practical difference isn't the math, it's that ZFS is copy-on-write, which eliminates the RAID 5/6 "write hole" risk entirely, at the cost of needing its own filesystem rather than sitting under an arbitrary one. Try the RAIDZ1, RAIDZ2, or RAIDZ3 Calculator.