How to Use This RAID Calculator

Installed Drives ( $N$ ): Set the total number of physical drives, from 2 to 64. This is the number you will buy and install, spares included.
Capacity Per Drive ( $S$ ): Enter the labeled capacity of one drive and pick a unit: GB ( $10^9$ bytes), TB ( $10^{12}$ bytes), GiB ( $2^{30}$ bytes) or TiB ( $2^{40}$ bytes). Changing the unit reinterprets the number you typed; it does not convert it.
RAID Level: Choose RAID 0, RAID 1, RAID 5, RAID 6, RAID 10, RAIDZ1, RAIDZ2 or RAIDZ3.
Mode: Simple assumes every installed drive is an active member. Advanced adds a Hot Spares input; spares are subtracted from the active set before capacity is computed.
Read the results: The hero figure is usable capacity before filesystem overhead, shown with a dual readout such as 21.828 TiB · 24 TB. The grid below it reports raw capacity, parity/mirror overhead, hot-spare capacity, installed capacity efficiency, and guaranteed failure tolerance. The comparison matrix shows every layout for the same drives and flags layouts that are invalid for your drive count.

One-click presets

Preset	Configuration	Usable	Typical use
NAS	4 × 8 TB, RAID 5	24 TB (21.828 TiB)	Home or small-office NAS
Archive	6 × 12 TB, RAID 6	48 TB (43.656 TiB)	Media library, cold archive
Mirror	2 × 4 TB, RAID 1	4 TB (3.638 TiB)	Simple two-disk redundancy

RAID Capacity Formulas at a Glance

Let $N$ be the number of active drives (installed drives minus hot spares), $S$ the capacity of one drive, and $p$ the number of parity drives' worth of redundancy.

Layout	Min. drives	Usable capacity	Redundancy overhead	Guaranteed failures survived	Efficiency (no spares)
RAID 0	2	$N \times S$	0	0	100%
RAID 1 (n-way mirror)	2	$S$	$(N-1) \times S$	$N-1$	$1/N$
RAID 5	3	$(N-1) \times S$	$S$	1	$(N-1)/N$
RAID 6	4	$(N-2) \times S$	$2S$	2	$(N-2)/N$
RAID 10	4 (even)	$(N/2) \times S$	$(N/2) \times S$	1 (up to $N/2$ if in different pairs)	50%
RAIDZ1	2	$\approx (N-1) \times S$	$\approx S$	1	$\approx (N-1)/N$
RAIDZ2	3	$\approx (N-2) \times S$	$\approx 2S$	2	$\approx (N-2)/N$
RAIDZ3	4	$\approx (N-3) \times S$	$\approx 3S$	3	$\approx (N-3)/N$

The general parity form covers RAID 5, RAID 6 and RAIDZ1/2/3:

C_{\text{usable}} = (N - p) \times S

where $p = 1$ for RAID 5 and RAIDZ1, $p = 2$ for RAID 6 and RAIDZ2, and $p = 3$ for RAIDZ3.

With hot spares, the calculator separates installed drives from active drives:

N = N_{\text{installed}} - N_{\text{spares}}

C_{\text{raw}} = N_{\text{installed}} \times S \qquad C_{\text{spare}} = N_{\text{spares}} \times S

C_{\text{overhead}} = (N - N_{\text{data}}) \times S

\eta = 100 \times \frac{C_{\text{usable}}}{C_{\text{raw}}}

Efficiency $\eta$ is measured against installed capacity, so hot spares lower it. That is deliberate: you paid for those drives.

Worked Examples

Example 1: RAID 5 calculator, 4 × 8 TB

C_{\text{usable}} = (4 - 1) \times 8\ \text{TB} = 24\ \text{TB}

Raw capacity is 32 TB, overhead is 8 TB, and efficiency is $24/32 = 75%$ . Converted to binary units:

\frac{24 \times 10^{12}}{2^{40}} = \frac{24 \times 10^{12}}{1.0995 \times 10^{12}} \approx 21.828\ \text{TiB}

This is why a "24 TB" array shows up as roughly 21.8 TB in Windows Explorer, Linux df -h, or a NAS volume manager. Nothing is missing; the label uses decimal terabytes and the operating system uses binary ones (see below).

Example 2: RAID 6 disk calculator, 6 × 12 TB

C_{\text{usable}} = (6 - 2) \times 12\ \text{TB} = 48\ \text{TB} \approx 43.656\ \text{TiB}

Raw capacity is 72 TB, overhead is 24 TB, efficiency is 66.7%, and any two drives can fail without data loss.

Example 3: Hot spare reservation, 6 × 12 TB RAID 6 with 1 spare

Here $N = 6 - 1 = 5$ active drives:

C_{\text{usable}} = (5 - 2) \times 12\ \text{TB} = 36\ \text{TB}

Parity overhead is 24 TB, the spare holds 12 TB, and installed efficiency drops to $36/72 = 50%$ . Fault tolerance is still 2 drives. The spare does not add tolerance; it only shortens the time you spend degraded by starting a rebuild automatically.

Example 4: Comparing every layout, 8 × 8 TB

Layout	Usable	Efficiency	Guaranteed failures
RAID 0	64 TB	100%	0
RAID 1 (8-way)	8 TB	12.5%	7
RAID 5	56 TB	87.5%	1
RAID 6	48 TB	75%	2
RAID 10	32 TB	50%	1
RAIDZ1	≈ 56 TB	≈ 87.5%	1
RAIDZ2	≈ 48 TB	≈ 75%	2
RAIDZ3	≈ 40 TB	≈ 62.5%	3

The live comparison matrix in the calculator generates this table for any drive count and size you enter.

Decimal (TB) vs Binary (TiB): Why Your Array Looks Smaller

Drive makers label capacity in SI decimal units: 1 TB = $10^{12}$ bytes. Most operating systems report in binary units: 1 TiB = $2^{40}$ = 1,099,511,627,776 bytes. The IEC 80000-13 prefixes (kibi, mebi, gibi, tebi), adopted by NIST and IEEE, exist to separate the two. Some operating systems (Windows) display binary values with the decimal-looking label "TB", which is the source of most "missing space" confusion.

1\ \text{TB} = \frac{10^{12}}{2^{40}}\ \text{TiB} \approx 0.9095\ \text{TiB}

Labeled drive size	Decimal bytes	Binary equivalent
1 TB	$1 \times 10^{12}$	0.909 TiB
4 TB	$4 \times 10^{12}$	3.638 TiB
8 TB	$8 \times 10^{12}$	7.276 TiB
12 TB	$12 \times 10^{12}$	10.914 TiB
16 TB	$16 \times 10^{12}$	14.552 TiB
20 TB	$20 \times 10^{12}$	18.190 TiB
24 TB	$24 \times 10^{12}$	21.828 TiB

The gap is about 9.05% at the tera level, and it is applied once, not per RAID level. The calculator's dual readout shows both values so you can compare against either the box or the OS.

RAID Levels Explained

RAID 0 (Striping)

Data is split into stripes written across all drives. Capacity and sequential throughput scale with drive count, but a single drive failure destroys the entire array, and the failure probability grows with every drive you add. Use it only for scratch space, caches, and data that can be regenerated. The calculator disallows hot spares for RAID 0 because a spare cannot rescue an array with no redundancy.

RAID 1 (Mirror)

Every active drive holds an identical copy. Usable capacity equals one drive no matter how many mirrors you add, and the array survives $N-1$ failures. Two-way mirrors are the usual choice for boot volumes and small servers. Read performance can scale with mirror count; write performance is limited to the slowest member.

RAID 5 (Single Distributed Parity)

RAID 5 stores one drive's worth of parity, rotated across all members. It needs at least three drives, yields $(N-1) \times S$ , and tolerates one failure. It is the most capacity-efficient redundant level, and efficiency improves as $N$ grows: 75% at 4 drives, 87.5% at 8, 91.7% at 12.

Its weakness is the rebuild window. While degraded, RAID 5 has no redundancy left, and a rebuild must read every surviving sector. A single unrecoverable read error (URE) during that window can fail the rebuild.

A spec-sheet model makes the risk concrete. A consumer drive rated at one URE per $10^{14}$ bits read, in a 4 × 8 TB RAID 5 that loses a drive, must read 24 TB (about $1.92 \times 10^{14}$ bits) from the survivors:

P(\text{at least one URE}) = 1 - e^{-1.92} \approx 85\%

At the $10^{15}$ -bit rating common on enterprise and NAS-grade drives, the same rebuild gives $1 - e^{-0.192} \approx 17.5%$ . Real-world URE rates are usually far better than the worst-case spec, and many controllers and filesystems can continue past a bad sector, so treat these numbers as a ceiling on the problem and not a prediction. They do show why large-drive arrays move to dual parity.

RAID 6 (Dual Distributed Parity)

RAID 6 stores two independent parity blocks (P and Q) per stripe, so any two drives can fail simultaneously. It needs at least four drives and yields $(N-2) \times S$ . It is the standard recommendation for arrays of large drives (roughly 8 TB and up) or wide arrays, because a second failure or a URE during a rebuild is survivable. The costs are lower efficiency at small $N$ (50% at 4 drives) and a higher small-write penalty.

RAID 10 (Striped Mirrors)

RAID 10 stripes across two-drive mirror pairs. It requires an even number of active drives (minimum four) and yields $(N/2) \times S$ at a fixed 50% efficiency. It offers the best random write performance and the fastest rebuilds of the redundant levels, because a rebuild is a single-disk copy rather than a full-array parity reconstruction.

Fault tolerance is the subtle part. RAID 10 guarantees survival of exactly one arbitrary failure. It can survive up to $N/2$ failures only if each failed drive belongs to a different pair. Once one drive has failed, the chance that a second random failure hits its partner is:

P(\text{array lost on 2nd failure}) = \frac{1}{N-1}

That is 33% for a 4-drive array and about 14% for an 8-drive array. This is why the calculator reports "1" for RAID 10 and not $N/2$ . If you need guaranteed two-drive tolerance, RAID 6 or RAIDZ2 provide it; three-way mirrored pairs do too, at 33% efficiency.

If you enter an odd number of active drives for RAID 10, the calculator shows an error instead of a result. With a hot spare, make sure the active count (installed minus spares) is even.

RAIDZ1, RAIDZ2 and RAIDZ3 (OpenZFS)

RAIDZ is OpenZFS's parity implementation. It behaves like RAID 5 and RAID 6 in terms of tolerance (1, 2 or 3 failed drives per vdev), but it uses variable-width stripes, which removes the RAID 5 "write hole" and lets ZFS checksum and self-heal data on every read.

This calculator models a single vdev of equal-size disks and reports approximate capacity $(N - p) \times S$ . In practice, ZFS usable space differs from that figure because of:

Allocation padding. RAIDZ rounds allocations up to a multiple of $(p + 1)$ sectors, so small blocks and very small recordsize/volblocksize values waste more space. Large-record datasets approach the ideal ratio; zvols with small block sizes can lose several percent to tens of percent.
ashift. The sector-size exponent chosen at pool creation (for example ashift=12 for 4 KiB sectors) changes the padding granularity.
Slop space. ZFS reserves 1/32 (about 3.1%) of pool capacity by default as a safety margin.
Metadata and compression. Metadata consumes space; compression can return more than it costs.
Fill level. Performance degrades as pools fill, and keeping pools below roughly 80% full is common practice.

The calculator's layout minimums for RAIDZ1/2/3 are 2, 3 and 4 drives, but those are structural minimums, not recommendations. A two-drive RAIDZ1 has the capacity of a mirror, and narrow RAIDZ3 vdevs are inefficient. Common practice is to size RAIDZ vdevs wider than the minimum and to keep vdev width moderate (many guides suggest staying under about 12 to 16 disks for a single vdev) so resilver times stay manageable.

Multiple vdevs. A ZFS pool with several vdevs sums their capacities and stripes across them. To use this calculator for a multi-vdev pool, calculate one vdev, then multiply by the vdev count. For example, two 6-wide RAIDZ2 vdevs of 16 TB drives equal $2 \times (6-2) \times 16 = 128$ TB raw-usable. Note that losing any one vdev loses the pool.

RAIDZ expansion. OpenZFS 2.3 and later can add a disk to an existing RAIDZ vdev. Existing data keeps its original data-to-parity ratio until rewritten, so the real capacity gain can lag the ideal formula until data is rewritten.

Which RAID Level Should You Choose?

Scenario	Recommended layout	Why
Two-bay NAS or boot volume	RAID 1 (or ZFS mirror)	Simple, fast rebuild, 1-drive tolerance
4-bay home NAS, drives under 8 TB	RAID 5 / RAIDZ1	Best capacity per drive with 1-drive tolerance
4-bay NAS, drives 8 TB or larger	RAID 6 / RAIDZ2 (or RAID 10)	Survives a failure plus a rebuild-time error
6 to 12 bay archive or media server	RAID 6 / RAIDZ2	67% to 83% efficiency with 2-drive tolerance
Databases, VMs, write-heavy workloads	RAID 10 / ZFS striped mirrors	Low write penalty, fast rebuilds
Very wide arrays (12+ drives)	RAID 6 / RAIDZ2 or RAIDZ3 across multiple vdevs	Rebuild times and failure odds grow with width
Scratch, render cache, transcodes	RAID 0	Maximum speed and capacity, zero tolerance

Efficiency by Drive Count

Active drives	RAID 5 / RAIDZ1	RAID 6 / RAIDZ2	RAIDZ3	RAID 10
4	75.0%	50.0%	25.0%	50%
5	80.0%	60.0%	40.0%	n/a (odd)
6	83.3%	66.7%	50.0%	50%
8	87.5%	75.0%	62.5%	50%
10	90.0%	80.0%	70.0%	50%
12	91.7%	83.3%	75.0%	50%

Write Penalty and Performance

Level	Small random-write I/Os per host write	Notes
RAID 0	1	No redundancy
RAID 1 / RAID 10	2	Write to both mirror members
RAID 5	4	Read data + parity, write data + parity
RAID 6	6	Read data + P + Q, write data + P + Q

Full-stripe sequential writes avoid most of the penalty, which is why parity RAID is perfectly good for media and backup workloads and weaker for random-write databases. Write-back caches and ZFS's copy-on-write transaction groups alter these numbers in practice.

Hot Spares: What They Do and Do Not Do

A hot spare is an idle drive that the controller or software RAID layer automatically uses as the rebuild target when an active member fails. In the calculator:

Spares are subtracted from the active drive count before capacity is computed.
Spares count toward raw (installed) capacity and lower installed efficiency.
Spares do not raise the number of simultaneous failures the array can survive. A RAID 5 with a hot spare still tolerates exactly one failure at a time; the spare just starts the rebuild without waiting for you to swap hardware.
RAID 0 does not accept spares.

A spare is most valuable in unattended or remote systems. In a NAS you check daily, a cold spare on a shelf plus a quick swap may be a better use of a bay. For large arrays, a hot spare plus RAID 6 meaningfully reduces your time spent with reduced redundancy.

Rebuild Time Estimate

Rebuild time is bounded by the capacity of the replacement drive and the sustained rate:

t_{\text{rebuild}} \approx \frac{S}{r_{\text{sustained}}}

An 8 TB drive at 150 MB/s sustained is $8 \times 10^{12} / (150 \times 10^6) \approx 53{,}000$ seconds, about 14.8 hours, in the ideal case. Under production load, or in parity arrays that throttle rebuilds, expect one to several days for large drives. Longer rebuilds mean a longer window of reduced redundancy.

Sizing in Reverse: How Many Drives Do You Need?

To hit a target usable capacity $T$ with drives of size $S$ , first compute the number of data drives:

N_{\text{data}} = \left\lceil \frac{T}{S} \right\rceil

Then add redundancy:

Layout	Drives required
RAID 5 / RAIDZ1	$N_{\text{data}} + 1$
RAID 6 / RAIDZ2	$N_{\text{data}} + 2$
RAIDZ3	$N_{\text{data}} + 3$
RAID 10	$2 \times N_{\text{data}}$
RAID 1	$\max(2,\ \text{copies})$ for $T \le S$

Add any hot spares on top. Example: you need 40 TiB usable using 16 TB drives. Convert first: $40\ \text{TiB} \times 1.0995 \approx 43.98\ \text{TB}$ , so $N_{\text{data}} = \lceil 43.98 / 16 \rceil = 3$ . RAID 5 needs 4 drives, RAID 6 needs 5, RAID 10 needs 6. Because the result is only about 4% under the capacity of three data drives (48 TB), add headroom for filesystem overhead and growth; plan for 20 to 30% free space, and for ZFS stay under about 80% full.

Use the calculator to confirm: enter the candidate drive count and size, then check the efficiency and comparison matrix.

Synology NAS Storage Planning & SHR Comparison

Many people searching for a Synology calculator or RAID calculator for Synology are choosing between standard RAID levels and Synology Hybrid RAID (SHR).

Equal-size drives: SHR-1 and SHR-2 give the same usable capacity and fault tolerance as RAID 5 and RAID 6 respectively, so this calculator's RAID 5 and RAID 6 results apply.
Mixed-size drives: SHR's advantage is that it can use the extra capacity of larger drives that RAID 5/6 would waste. This calculator does not model SHR or any mixed-size slicing. For mixed drives, standard RAID is limited by the smallest drive: usable capacity is based on $S_{\min}$ , so enter the smallest drive size to see what classic RAID would deliver.
What the NAS reports: DSM and most NAS operating systems display binary sizes (often with a decimal-style label), then subtract filesystem metadata (Btrfs or ext4) and system partitions. Expect the volume you create to be somewhat smaller than the result shown here. The TB and TiB dual readout lets you separate the unit gap from real overhead.
Other vendors: QNAP, TrueNAS, Unraid and Netgear ReadyNAS have their own pooling models. Standard RAID 1/5/6/10 and ZFS RAIDZ numbers from this tool apply to the standard modes; vendor-specific pool types (like Unraid's parity array with mixed sizes) do not follow these formulas.

Is RAID a Backup? No.

RAID protects against drive hardware failure. It does not protect against file deletion, ransomware, filesystem corruption, controller failure, fire, theft, or user error, because all of those are replicated or applied across every member instantly. Treat RAID as an availability feature, and keep independent backups, ideally following the 3-2-1 rule: three copies of your data, on two different media types, with one copy offsite. The calculator shows a persistent advisory about this because it is the most common and most costly RAID misunderstanding.

What the Calculator Does and Does Not Model

It models:

Equal-size disks in a single RAID array or single ZFS vdev.
Pure mathematical usable capacity: raw parity and mirror arithmetic.
Hot-spare reservation and its effect on active drives and efficiency.
Guaranteed arbitrary failure tolerance per layout.

It does not model:

Filesystem overhead from ext4, XFS, NTFS, APFS, Btrfs or ZFS, such as inode tables, journals, reserved blocks, and metadata.
ZFS specifics like allocation padding, slop space, compression ratio, and dataset reservations.
Mixed-size drives or vendor technologies such as SHR/SHR-2.
Nested RAID beyond RAID 10, such as RAID 50 and RAID 60, or multi-vdev ZFS pools (calculate one group and multiply).
Hardware controller reserved areas (some controllers hold back a small amount of each drive for metadata or to normalize slightly different drive sizes).

Quick Answers

How do I calculate usable capacity for RAID 5?

Usable capacity is $(N-1) \times S$ . Four 8 TB drives give $3 \times 8 = 24$ TB (21.828 TiB). RAID 5 needs at least three drives and survives one drive failure.

How do I calculate usable capacity for RAID 6?

Usable capacity is $(N-2) \times S$ . Six 12 TB drives give $4 \times 12 = 48$ TB. RAID 6 needs at least four drives and survives two simultaneous failures.

What is the difference between a RAID calculator and a RAIDZ calculator?

The capacity math is the same shape: RAIDZ1/2/3 mirror RAID 5/6 plus a third parity level, with $(N-p) \times S$ capacity. ZFS adds allocation padding, slop space, and metadata that reduce real usable space, so RAIDZ results are approximate. The calculator labels them that way.

Why does my 24 TB RAID show only about 21.8 TB?

Drive makers use decimal terabytes ( $10^{12}$ bytes) and your OS reports binary tebibytes ( $2^{40}$ bytes). $24\ \text{TB} \approx 21.828\ \text{TiB}$ . Filesystem formatting then takes a further small amount.

How many drives can fail in RAID 10?

One failure is guaranteed to be survivable. More failures are survivable only if no mirror pair loses both drives. After one failure in an $N$ -drive array, the odds that the next random failure kills the array are $1/(N-1)$ .

Does a hot spare increase fault tolerance?

No. A hot spare lets the array start rebuilding automatically, which shortens the degraded window, but the number of simultaneous failures the array can withstand is unchanged. If a second drive fails before the rebuild finishes in a RAID 5, you lose the data.

Can I mix drive sizes in RAID?

Standard RAID treats every member as if it were the size of the smallest drive, so the extra capacity on larger drives is wasted. To estimate classic RAID capacity for mixed drives, enter the smallest drive size. Synology SHR and similar technologies can use the difference, but they are outside this tool's model.

What is the minimum number of drives for each RAID level?

RAID 0: 2. RAID 1: 2. RAID 5: 3. RAID 6: 4. RAID 10: 4 (even number). RAIDZ1: 2. RAIDZ2: 3. RAIDZ3: 4. The calculator enforces these minimums and flags layouts that are invalid for your drive count in the comparison matrix.

RAID 5 or RAID 6 for a 4-bay NAS?

With four drives, RAID 5 delivers 75% efficiency and RAID 6 delivers 50%, the same as RAID 10. If the drives are large (8 TB or more) and the data matters, RAID 6 or RAID 10 gives more safety at the cost of capacity. If you have solid backups and smaller drives, RAID 5 is a common balance. Compare them directly in the matrix.

How can I check the result against a real system?

Compare the calculator's TiB figure with what your OS or NAS reports, then expect the real figure to be slightly lower because of filesystem and system-partition overhead. For ZFS, run zpool list for raw pool size and zfs list for the space available after parity, padding, and slop.

Standards and Terminology Used

RAID levels follow the definitions in the SNIA Dictionary and the original Berkeley RAID taxonomy. RAID 10 (1+0) here means a stripe over two-drive mirrors.
RAIDZ1/2/3 follow OpenZFS documentation for single, double, and triple parity vdevs.
Capacity units: decimal prefixes (kB, MB, GB, TB) are SI; binary prefixes (KiB, MiB, GiB, TiB) are defined in IEC 80000-13 and adopted by NIST and IEEE.
Failure tolerance in this tool means guaranteed simultaneous arbitrary drive failures survivable without data loss, not the best case.

Related Planning Tips

Mix drive batches and ages when possible; same-batch drives can share failure modes.
Monitor SMART attributes and schedule periodic scrubs (ZFS scrub, md check, controller patrol reads) so latent errors are found before a rebuild.
Prefer CMR drives for parity RAID and ZFS; heavy rebuild and resilver workloads can behave poorly on some SMR drives.
Test restores from your backups, not just the array's health indicator.
Re-run this RAID calculator whenever you change drive size or count; moving from 4 to 6 drives often changes the best layout.

Calculation Methodology Documented

Read the formula definitions, supported input ranges, and limitations before using this estimate.

Review Methodology

Author Debjeet Bhowmick AWS Certified Solutions Architect • Cloud & DevOps Consultant

AWS Certified Solutions Architect Professional and Senior Cloud & DevOps Consultant with over 14 years of hands-on expertise architecting resilient cloud infrastructure across AWS, GCP, and Azure. Specializes in infrastructure as code, continuous delivery automation, and telemetry across Terraform, Ansible, Docker, Kubernetes, ELK, Python, and SQL. Creator of OrangeCalc, delivering client-side high-precision computational tools.