RAID Capacity and Rebuild Risk

Updated October 2026

How much of an array you can actually use at RAID 0, 1, 5, 6 and 10, how many drives a project needs, how long a rebuild takes, and what the drive's read-error rating means for a RAID 5 rebuild.

12 min readIntermediate

RAID 5 vs RAID 6
Usable capacity
Rebuild time
URE math

On this page

Usable capacity by RAID level

A RAID array spends some of its drives on redundancy. Mirroring stores every block twice; parity stores enough extra data to rebuild a missing drive. With n drives of size d, the space left for footage is:

Usable capacity

RAID 0  = n × d
RAID 1  = d          (every drive holds the same data)
RAID 5  = (n − 1) × d
RAID 6  = (n − 2) × d
RAID 10 = n ÷ 2 × d   (n even)

Drive sizes are decimal terabytes, as printed on the drive. Format the array and the operating system takes a little for the file system, and some systems then show the result in binary units, which look about 9% smaller.

Where

n: Number of drives in the array
d: Capacity of one drive (use the smallest if they differ)(TB)

Example: Eight 16 TB drives

Calculation:: RAID 5: 7 × 16 = 112 TB; RAID 6: 6 × 16 = 96 TB; RAID 10: 4 × 16 = 64 TB
Result:: RAID 6 keeps 75% of the 128 TB raw, RAID 10 keeps 50%

The free RAID capacity calculator compares every level for your drives in your browser.

Two details trip people up. Parity levels need a minimum number of drives: three for RAID 5, four for RAID 6. RAID 10 needs at least four drives and an even number, because it stripes across mirrored pairs. And the parity overhead is fixed in drives, not in percent, so it shrinks as the array grows: RAID 6 loses half of a four-drive array but only a sixth of a twelve-drive one.

Capacity for common arrays

Usable space for 16 TB drives in the enclosure sizes most edit suites and DIT carts use. The percentage is the share of raw capacity left for data.

16 TB drives: usable TB (share of raw)

Drives	Raw	RAID 5	RAID 6	RAID 10
4	64 TB	48 TB (75%)	32 TB (50%)	32 TB (50%)
6	96 TB	80 TB (83.3%)	64 TB (66.7%)	48 TB (50%)
8	128 TB	112 TB (87.5%)	96 TB (75%)	64 TB (50%)
12	192 TB	176 TB (91.7%)	160 TB (83.3%)	96 TB (50%)

RAID 0 is the raw figure with no protection at all. RAID 1 is one drive's worth, 16 TB, however many drives the mirror spans; that is how the CineMath Storage/RAID calculator treats it, so an eight-drive mirror shows 16 TB and survives seven failures.

How many drives you need

Work backwards from the space the project needs. Divide by the drive size, round up, then add the drives the level spends on redundancy.

Drives for a target capacity

Data drives = ⌈ Needed ÷ d ⌉
RAID 5 drives = data drives + 1
RAID 6 drives = data drives + 2
RAID 10 drives = data drives × 2

Example: 100 TB of usable space on 16 TB drives

Calculation:: ⌈100 ÷ 16⌉ = 7 data drives; RAID 6: 7 + 2 = 9 drives
Result:: 9 drives, 112 TB usable. On 20 TB drives: 5 + 2 = 7 drives, exactly 100 TB with no headroom

The 20 TB answer shows why the target should include headroom. An array filled to the last terabyte has nowhere for renders, cache or the reshoot nobody planned, so size the “needed” figure from the footage, proxies and working files together, as the storage planning guide does, and then add your own margin.

What each level survives

Fault tolerance (SNIA definitions)

Level	Minimum drives	Survives
RAID 0	2	Nothing: one failed drive loses the array
RAID 1	2	All but one drive of the mirror
RAID 5	3	Any one drive
RAID 6	4	Any two drives
RAID 10	4, even	One drive in each mirrored pair

RAID 10's tolerance depends on which drives fail. After the first failure, the array dies only if the next failure is that drive's mirror partner. If the second failure is equally likely to be any of the remaining drives, the chance it hits the partner is 1 in (n − 1): 1 in 7 (14%) for eight drives, 1 in 3 for four. RAID 6 survives any second failure, which is why it is the safer choice for large parity arrays.

Rebuild time

When a drive fails, the replacement has to be written end to end from the survivors. A drive cannot be written faster than its sustained transfer rate, which gives a floor for the rebuild:

Shortest possible rebuild

Rebuild (s) ≥ Drive capacity (bytes) ÷ Max sustained rate (bytes per second)

Data sheets quote the maximum sustained rate at the outer edge of the platters (OD). The inner tracks are slower, and an array that is still serving an edit while it rebuilds is slower again, so real rebuilds take longer than this.

Example: 16 TB Seagate IronWolf, 240 MB/s (data sheet)

Calculation:: 16 × 10¹² ÷ (240 × 10⁶) = 66,667 s
Result:: At least 18.5 hours. A 20 TB Exos X20 at 285 MB/s: 70,175 s, at least 19.5 hours

For that whole time a RAID 5 array has no redundancy left, and every surviving drive is being read from end to end. That is the window the next section is about.

Read errors during a rebuild

Hard drive data sheets give a maximum rate of non-recoverable read errors (UREs): sectors the drive cannot read back. Seagate rates its 7,200 rpm IronWolf drives at no more than 1 per 10¹⁵ bits read and the 5,400 rpm models at 1 per 10¹⁴. A RAID 5 rebuild reads every surviving drive in full, so you can estimate the chance of meeting at least one error at the rated maximum:

Chance of at least one URE in a rebuild

Bits read = (n − 1) × d × 8 × 10¹²
Expected errors = Bits read ÷ R
P(at least one) ≈ 1 − e^(−Expected errors)

Where

n: Drives in the array, including the failed one
d: Drive capacity(TB)
R: Rated bits read per error, e.g. 10¹⁵

Example: Eight 16 TB drives in RAID 5, one failed

Calculation:: 7 × 16 TB = 112 TB = 8.96 × 10¹⁴ bits; at 10¹⁵: 0.896 expected errors
Result:: 1 − e^−0.896 = 59%. At 10¹⁴ the expectation is 8.96 errors, effectively certain

Chance of at least one URE while rebuilding RAID 5, at the rated maximum

Array	Data read	At 1 per 10¹⁴	At 1 per 10¹⁵
4 × 4 TB	12 TB	62%	9%
6 × 20 TB	100 TB	> 99.9%	55%
8 × 16 TB	112 TB	> 99.9%	59%
8 × 20 TB	140 TB	> 99.9%	67%

Read these as a worst case, not a forecast. The data sheet figure is a maximum, and drives usually do better. What a URE costs depends on the RAID controller or software: at best one unreadable stripe, which damages a file; at worst a rebuild that stops. Either way the arithmetic explains the usual advice: with today's large drives, use RAID 6, which still has a second parity to correct a bad sector while it rebuilds one failed drive, and keep RAID 5 for small arrays of smaller drives.

Choosing a level for video

RAID 0 for scratch and cache you can regenerate, such as render files, proxies you can rebuild, a transcode target. Never for the only copy of camera originals.
RAID 5 for small arrays (four or five drives of modest size), where the rebuild reads little and finishes quickly.
RAID 6 for the edit-suite or DIT array of large drives: two-drive tolerance at the cost of two drives.
RAID 10 when you can afford to give up half the raw space for simpler, faster rebuilds: a failed drive is copied from its mirror instead of recomputed from every drive.

Case study

Sizing the edit array for a documentary series

Scenario

Six episodes, about 70 TB of camera originals and proxies expected, edited from an eight-bay enclosure.

Challenge

Pick a drive size and RAID level that holds the project with room to spare and survives a failed drive during a long rebuild.

Solution

Rule out RAID 5: a rebuild of eight large drives reads more than 100 TB, so a URE is likely at a 10¹⁵ rating.
Try RAID 6 on 16 TB drives: 6 data drives give 96 TB.
Check the headroom against the 70 TB estimate.

The numbers

(8 − 2) × 16 = 96 TB usable
96 − 70 = 26 TB of headroom
Rebuild floor: 16 × 10¹² ÷ (240 × 10⁶) = 66,667 s = 18.5 h

Result

RAID 6 on eight 16 TB drives: 96 TB usable, 26 TB of headroom, and any two drives can fail.

Key takeaways

Enter the 70 TB as the target in CineMath's Storage/RAID calculator and it marks RAID 6 as the safest level that fits, with 26 TB of headroom.
The array is still one copy: the originals also go to shuttle drives or tape.

RAID is not a backup

Redundancy covers one thing: a drive that stops working. It does nothing about a deleted bin, a corrupted file written to every drive at once, a failed enclosure, theft, fire or water. Count the array as one copy in a 3-2-1 plan, and keep the other copies on separate media, at least one of them somewhere else. For long-term copies of finished projects, see LTO tape archiving.

Next step: put your drive count, drive size and the space you need into the Storage/RAID calculator, in the app or the free RAID capacity calculator in your browser. It shows every level side by side, marks the ones that fit, and names the safest one.