RAID Level Comparison Chart: Every Layout, 2 to 16 Drives

Every layout, every sensible drive count, one page. RAID 5, RAIDZ1 and SHR all return (n minus 1) times the smallest drive and survive one failure. RAID 6, RAIDZ2 and SHR-2 return (n minus 2) and survive two. RAID 10 returns half of raw and guarantees one. Those three sentences cover most of the decision. The tables below give the exact figure for every combination, plus the two columns that actually decide a layout: what percentage of your money goes to parity, and how much data has to be read back to recover from a failure.

Capacity here is expressed as a multiple of one drive, so the table is correct at any drive size. Two 8 TB drives in RAID 1 give 1x, which is 8 TB. Eight 16 TB drives in RAID 6 give 6x, which is 96 TB. Multiply the multiple by your drive size and you have your answer.

What is the usable capacity of each RAID level?

Usable capacity is the space the filesystem can address after redundancy is taken out, and for every layout except SHR it is a fixed multiple of the smallest drive in the set. That last clause matters more than it looks: classic RAID levels and every ZFS RAIDZ vdev truncate all members down to the size of the smallest one, so a single 4 TB drive in a set of 12 TB drives throws away 8 TB on each of the others.

Usable capacity as a multiple of one drive, computed with the same engine behind the RAID capacity calculator. A dash means the layout cannot be built at that drive count: RAID 6 and RAID 10 need four, RAIDZ3 needs five, and RAID 10 needs an even number. SHR figures assume identical drives, where SHR is exactly RAID 5 and SHR-2 is exactly RAID 6.

Read the RAID 1 row carefully, because it is the one people misread. An n-way mirror gives you one drive of usable space no matter how many drives you put in it. Four drives in RAID 1 is not 2x, it is 1x with three copies. That is occasionally what you want on a boot pool and almost never what you want for bulk storage.

The other row worth noting is JBOD, which returns the sum of every drive because it does not truncate anything and does not protect anything. It is the only layout in the table where mixed drive sizes cost you nothing, and the only one where losing a drive loses only what was on that drive rather than the whole set.

What is the parity overhead and rebuild cost of each layout?

Parity overhead is the fraction of what you bought that you cannot store data on, and rebuild reads is how much data must be read off the surviving drives to reconstruct one failed member. The second column is the one that changes decisions, because it is what turns an abstract "one drive of redundancy" into "the array will read 140 TB over the next three days while unprotected".

Usable and rebuild reads are multiples of one drive. "Can fail" is the guaranteed number of simultaneous drive failures the layout survives and "best case" is the luckiest arrangement, which differs only for RAID 10 and n-way mirrors. The last three columns work the multiple through at 8 TB drives, with the binary TiB figure your NAS will print alongside. See the TB against TiB chart for why those two differ by 9.05 percent.

The rebuild column is the argument for double parity, and it is arithmetic rather than opinion. Single parity across eight 8 TB drives has to read 56 TB off seven surviving members with no redundancy left. The same array under double parity still has one drive of protection while it reads. Manufacturers rate NAS drives at one unrecoverable read error per 10^14 or 10^15 bits, which is a warranty floor rather than a measurement, and both ZFS and mdadm survive a single read error rather than aborting. Use the figure to rank layouts, not as a forecast. Ranked that way, single parity across eight large drives loses.

How do I read the formula for each layout?

Every layout in this table reduces to one line of arithmetic. The reference card below is the version to memorise, along with the minimum drive count and the kind of array each one belongs in. The overhead column shows what each layout costs at eight drives, which is the fairest single point of comparison because it is a bay count all of them can reach.

Minimum drive counts are the practical ones the calculator enforces. Some implementations will let you build a two-drive RAID 5 or a four-drive RAIDZ3, and neither is a sensible thing to do. For the mechanics of how each layout writes and reads data, see RAID levels explained.

Is SHR the same as RAID 5?

With identical drives, yes, exactly. Not approximately and not nearly: Synology SHR on a set of matched drives collapses into a single parity group and yields (n minus 1) times the drive size, which is the RAID 5 formula. SHR-2 collapses to RAID 6 in the same way. Anyone who tells you SHR gives more usable space on four identical drives is wrong, and you can see it in the matrix above where the rows are identical.

The difference appears the moment the drives are not the same size. SHR slices the drives into horizontal bands at every distinct drive size and gives each band its own parity group. A band sitting on k drives with p drives of parity yields (k minus p) times the band height when k is greater than p, and is stranded entirely when it is not. Classic RAID does something much cruder: truncate everything to the smallest member and throw the rest away.

Five drives totalling 42 TB. Stranded capacity is space you paid for that the array cannot address. Efficiency is usable divided by the total physical capacity you bought, which is the honest way to compare layouts on a mismatched set. At single parity SHR returns 14 TB more than RAID 5, and at double parity SHR-2 returns 6 TB more than RAID 6.

That gap is the entire commercial argument for Synology, and it is worth being precise about when it applies. If you buy five matched drives today, SHR gains you nothing. It gains when one drive in the set is a survivor from an older machine, and it gains again in two or three years when you replace drives one at a time and want the extra space before the last one has been swapped. See SHR against RAID 5 for the band-by-band walkthrough, and run your own drive set through the RAID capacity calculator, which is the only one that puts SHR, RAIDZ and standard RAID side by side on the same drives.

Why does parity get cheaper as you add bays?

Parity costs a fixed number of drives, not a fixed percentage. One drive of parity across four members is 25 percent of what you bought. The same one drive across twelve members is 8.3 percent. This is why the layout advice changes with bay count rather than being a universal rule, and it is the strongest argument for buying a wider chassis than you think you need.

Efficiency is usable capacity divided by raw capacity. Rebuild reads are multiples of one drive. Notice that efficiency and rebuild cost move in opposite directions: wider arrays waste less on parity and read more during recovery, which is why very wide single-parity arrays are the worst of both worlds.

The two columns on the right are the counterweight. A wider array is more efficient and takes longer to recover, because there are more surviving members to read. At four drives a RAID 5 rebuild reads three drives. At sixteen it reads fifteen. That is the reason ZFS practice caps a RAIDZ2 vdev at around eight to twelve drives and reaches for RAIDZ3 beyond it, rather than building one enormous vdev.

Which layout should I choose at each bay count?

A short version of everything above, in the order the decision actually gets made.

• Two bays. RAID 1, and accept that half your money buys redundancy. SHR on two drives is also single parity and behaves the same way. The alternative, two independent drives with one backing up the other, is worth considering because it protects against deletion as well as failure.
• Four bays. RAID 5, RAIDZ1 or SHR with drives up to about 8 to 10 TB, giving 3x usable. Past that size, RAID 6 or SHR-2 at 2x, which is a heavy 50 percent tax and still the right call on 16 TB and larger drives. See best 4-bay NAS.
• Six bays. The crossover point. RAID 6 costs 33 percent here rather than 50, so double parity stops feeling expensive. Single parity across six large drives is difficult to defend.
• Eight bays and up. RAID 6, RAIDZ2 or SHR-2, always. Overhead is down to 25 percent and a single-parity rebuild would read seven full drives with nothing protecting the array. See best 8-bay NAS.
• Twelve and beyond. RAIDZ3, or split into two vdevs of RAIDZ2. A rebuild at this width takes long enough that a third parity drive is cheap insurance rather than paranoia.

What does each layout cost in drives and dollars?

Parity is not an abstraction, it is drives you buy and do not store anything on. At current verified prices a Seagate IronWolf 8TB NAS HDD (ST8000VN004) is $289.99 and a Seagate IronWolf Pro 16TB NAS HDD (ST16000NT001) is $579.99, so the second parity drive in a RAID 6 array is a real line item rather than a checkbox. Four 16 TB drives under RAID 6 give 32 TB usable for the price of 64 TB of drives.

That is the correct way to feel the cost, and it is also the argument for solving the problem with bay count rather than with layout. Eight 8 TB drives under RAID 6 give 48 TB usable. Four 16 TB drives under RAID 6 give 32 TB for almost the same money in drives. The wider array wins on capacity, loses on enclosure cost and power, and takes longer to rebuild. Work the trade with the cost per terabyte chart, which prices exactly this comparison.

Whatever the table says, parity is not a backup. Every layout on this page protects against a drive dying. None of them protects against a deleted folder, a ransomware run, a failed power supply taking the whole enclosure, a fire or a theft. A large external drive holding a second copy covers a class of failure that no amount of parity touches. Size it with the backup sizing calculator and read the 3-2-1 rule guide.

Related reading

• RAID levels explained for how each layout actually writes and reads data
• Usable capacity chart for the same maths at real drive sizes with prices
• RAID rebuild time chart for how long the rebuild column takes in hours
• RAID 5 against RAID 6 for the single decision this page exists to inform
• RAID 10 against RAID 6 for the random-workload case
• RAID capacity calculator for your own drives, mixed sizes included