In short. Usable capacity depends on the level: RAID 5 gives you every drive minus one, RAID 6 minus two, RAID 10 exactly half. Eight 4 TB drives yield 28 TB in RAID 5, 24 TB in RAID 6 and 16 TB in RAID 10. The second number that matters is the write penalty: ×4 on RAID 5 and ×6 on RAID 6 against ×2 on a mirror — and that is what decides whether the array can carry a database.
Usable capacity is shown in TB as the drives are labelled and in TiB as the operating system reports it. Rebuild time assumes 80 MB/s: a loaded controller delivers less, an idle one more.
How usable capacity is derived
RAID 0 simply adds the drives with no redundancy: all capacity is usable, but a single failure destroys the array. RAID 1 and RAID 10 return half, because data is written in full to a pair of drives. RAID 5 spends one drive worth of space on parity, RAID 6 two, regardless of the total count. RAID 50 and 60 are built from several groups of five or six, so they lose one or two drives per group.
There is a second loss, in units. A drive labelled 4 TB holds 4,000,000,000,000 bytes, while the operating system divides by 1024 and reports 3.64 TiB. That is neither a defect nor controller reserve: it is how decimal terabytes convert to binary ones. The calculator shows both figures.
What the write penalty means
Every write on RAID 5 takes four operations: read the old data, read the old parity, write the new data, write the new parity. RAID 6 needs six, because there are two parities. A mirror needs two writes. That is the penalty: of the 1000 IOPS the drives physically deliver, roughly 250 reach the application on writes under RAID 5.
For a file server or an archive this does not matter; for 1C and databases it decides everything. The working rule: databases and virtualisation with heavy writes go on RAID 10; files, backups and video archives on RAID 6; RAID 5 is left for small arrays of drives up to 4 TB.
Why RAID 5 is risky on large drives
A rebuild reads every remaining drive end to end. An 8 TB drive at 80 MB/s takes around 28 hours, and for all of that time the array runs without redundancy. A second failure within that day means data loss. On top of that, reading tens of terabytes raises the odds of hitting an unrecoverable read error. So from 4 TB upwards the choice is RAID 6, with a hot spare as a separate drive so that the rebuild starts without anyone present.
Next steps
RAID is not a backup: it protects against hardware failure, not against a deleted file or ransomware. To pick drives and a controller: server SSD and HDD drives, RAID controllers. PERC and Smart Array models with cache and power-loss protection are compared in the controller reference, prices and compatibility in the drives and controllers guide.