RAID 10 vs RAID 6: Speed Against Capacity
RAID 10 gives n divided by 2 times the smallest drive and guarantees survival of one failure. RAID 6 gives (n minus 2) times the smallest drive and guarantees two. At four drives they give identical usable capacity, so RAID 6 is strictly better there unless the workload is random writes. At eight drives RAID 6 gives six drives of capacity against RAID 10's four, and RAID 10's real advantage is a rebuild that copies one drive instead of reading every survivor.
This is the performance against capacity comparison, and it has one result that surprises almost everybody. At four drives, RAID 10 and RAID 6 give exactly the same usable capacity, and RAID 6 guarantees survival of two failures against RAID 10's one. At eight drives the picture reverses on capacity: RAID 6 returns six drives of space against RAID 10's four. RAID 10's genuine advantage is not capacity and it is not sequential speed. It is random writes, and a rebuild that copies one drive rather than reading every surviving member of the array.
What is the difference between RAID 10 and RAID 6?
RAID 10 is a stripe across mirrored pairs: every drive has an exact twin, and data is spread across the pairs. Usable capacity is n divided by 2 times the smallest drive, and it needs an even number of drives. There is no parity computation anywhere, so a write is just two writes.
RAID 6 stripes data across every member and dedicates two drives worth of space to parity syndromes. Usable capacity is (n minus 2) times the smallest drive, and any two members can fail without loss. A write that does not cover a full stripe has to read the old data and both parity syndromes, then write all three back.
Both truncate every member to the size of the smallest drive, so mixed drive sizes strand capacity in either layout. The mechanics of every level are collected in RAID levels explained, and the parity count decision that sits underneath this one is worked through in RAID 5 against RAID 6.
How much usable capacity does each give?
The crossover is the whole story, and it happens at six drives. Below that the two layouts tie on capacity, above it RAID 6 pulls away and never comes back.
| Drives | Drive size | Raw | RAID 10 usable | RAID 6 usable | RAID 10 efficiency | RAID 6 efficiency | Capacity winner |
|---|---|---|---|---|---|---|---|
| 4 | 8 TB | 32 TB | 16 TB | 16 TB | 50% | 50% | Identical capacity |
| 4 | 12 TB | 48 TB | 24 TB | 24 TB | 50% | 50% | Identical capacity |
| 4 | 20 TB | 80 TB | 40 TB | 40 TB | 50% | 50% | Identical capacity |
| 6 | 8 TB | 48 TB | 24 TB | 32 TB | 50% | 67% | RAID 6 by 8 TB |
| 6 | 12 TB | 72 TB | 36 TB | 48 TB | 50% | 67% | RAID 6 by 12 TB |
| 6 | 20 TB | 120 TB | 60 TB | 80 TB | 50% | 67% | RAID 6 by 20 TB |
| 8 | 8 TB | 64 TB | 32 TB | 48 TB | 50% | 75% | RAID 6 by 16 TB |
| 8 | 12 TB | 96 TB | 48 TB | 72 TB | 50% | 75% | RAID 6 by 24 TB |
| 8 | 20 TB | 160 TB | 80 TB | 120 TB | 50% | 75% | RAID 6 by 40 TB |
| 10 | 8 TB | 80 TB | 40 TB | 64 TB | 50% | 80% | RAID 6 by 24 TB |
| 10 | 12 TB | 120 TB | 60 TB | 96 TB | 50% | 80% | RAID 6 by 36 TB |
| 10 | 20 TB | 200 TB | 100 TB | 160 TB | 50% | 80% | RAID 6 by 60 TB |
| 12 | 8 TB | 96 TB | 48 TB | 80 TB | 50% | 83% | RAID 6 by 32 TB |
| 12 | 12 TB | 144 TB | 72 TB | 120 TB | 50% | 83% | RAID 6 by 48 TB |
| 12 | 20 TB | 240 TB | 120 TB | 200 TB | 50% | 83% | RAID 6 by 80 TB |
Computed with the same functions behind the RAID capacity calculator. Figures are decimal TB from the drive label. Your operating system reports each about 9.05 percent smaller in binary tebibytes, so 72 TB usable reads as 65.5 TiB, which is covered on the TB against TiB chart.
RAID 10 sits at 50 percent efficiency forever, by construction, because half of every array is a copy. RAID 6 starts at 50 percent on four drives and climbs: 67 percent at six, 75 at eight, 83 at twelve. That is the entire capacity argument, and it is decided by how many bays you own rather than by anything about the workload.
How many drives can each layout actually lose?
RAID 6 tolerates two failures, guaranteed, in every configuration. Any two members, in any order, with no luck involved. RAID 10 tolerates one guaranteed and up to n divided by 2 in the best case, and the gap between those two numbers is where people get into trouble.
After the first drive fails in a RAID 10, exactly one of the surviving members is its mirror partner, and losing that one ends the array. With n drives there are (n minus 1) survivors, so a second random failure is fatal one time in (n minus 1).
| Drives | Mirror pairs | RAID 10 guaranteed tolerance | RAID 10 best case | Second failure survived | Second failure fatal | RAID 6 guaranteed tolerance |
|---|---|---|---|---|---|---|
| 4 | 2 | 1 | 2 | 66.7% | 33.3% | 2 |
| 6 | 3 | 1 | 3 | 80% | 20% | 2 |
| 8 | 4 | 1 | 4 | 85.7% | 14.3% | 2 |
| 10 | 5 | 1 | 5 | 88.9% | 11.1% | 2 |
| 12 | 6 | 1 | 6 | 90.9% | 9.1% | 2 |
The two RAID 10 percentage columns assume a second failure hits a uniformly random surviving member. Real failures are not perfectly independent, because drives bought in one order share a manufacturing window and running hours, which pushes the fatal case slightly higher than the arithmetic suggests.
Read the four-drive row. A four-drive RAID 10 loses everything one time in three if a second drive fails during the rebuild window, while a four-drive RAID 6 with the same usable capacity loses nothing at all. Anyone recommending RAID 10 on a four-bay NAS for resilience reasons has the argument backwards. The reason to run it there is write performance, and only that.
Why does RAID 10 rebuild so much more gently?
Because it copies rather than reconstructs. Replacing a failed RAID 10 member reads one drive, its mirror partner, and writes a straight copy onto the replacement. Nothing else in the array is touched. RAID 6 has to read every surviving member end to end and recompute the missing blocks from the parity syndromes.
| Drives | Drive size | RAID 10 rebuild reads | RAID 6 rebuild reads | Read volume ratio | Drives read, RAID 10 | Drives read, RAID 6 | Hours at 150 MB/s |
|---|---|---|---|---|---|---|---|
| 4 | 8 TB | 8 TB | 16 TB | 2x | 1 | 3 | 15 h |
| 8 | 8 TB | 8 TB | 48 TB | 6x | 1 | 7 | 15 h |
| 12 | 8 TB | 8 TB | 80 TB | 10x | 1 | 11 | 15 h |
| 4 | 12 TB | 12 TB | 24 TB | 2x | 1 | 3 | 22 h |
| 8 | 12 TB | 12 TB | 72 TB | 6x | 1 | 7 | 22 h |
| 12 | 12 TB | 12 TB | 120 TB | 10x | 1 | 11 | 22 h |
| 4 | 20 TB | 20 TB | 40 TB | 2x | 1 | 3 | 37 h |
| 8 | 20 TB | 20 TB | 120 TB | 6x | 1 | 7 | 37 h |
| 12 | 20 TB | 20 TB | 200 TB | 10x | 1 | 11 | 37 h |
Duration tracks the size of one member in both layouts, because RAID 6 survivors read in parallel while the replacement is written. The difference is not the clock, it is how many drives are working flat out while it runs. Precomputed durations are on the rebuild time chart.
This is RAID 10's strongest genuine argument and it deserves to be stated properly. A RAID 6 rebuild puts every surviving drive under a full-surface sequential read for a day or more, at the exact moment those drives are the only copy of your data. A RAID 10 rebuild puts one drive under that load. If the surviving drives are old, or came from the same manufacturing batch, or have been throwing SMART reallocations, that difference is real.
The catch is which drive gets stressed. RAID 10 reads only one member during the rebuild, and it is the single drive whose loss would destroy the array. RAID 6 spreads the load, and can lose any one of the drives doing that reading and still complete. Gentler on the array as a whole, harsher on the one member that matters most. Neither is unambiguously safer, which is why the guaranteed tolerance number is the one to decide on.
How much faster is RAID 10 for random writes?
Roughly three times, and the reason is the write penalty. A RAID 10 write becomes two writes, one to each half of a mirror. A RAID 5 write that does not cover a full stripe becomes four operations, and a RAID 6 write becomes six, because both parity syndromes have to be read and rewritten.
A 7200 rpm 3.5 inch drive sustains roughly 100 random operations per second. The table applies the write penalties to that figure. Random reads are unaffected: every layout can read from every drive at once.
| Drives | Random read IOPS, any layout | RAID 10 write IOPS | RAID 5 write IOPS | RAID 6 write IOPS | RAID 10 advantage over RAID 6 |
|---|---|---|---|---|---|
| 4 | 400 | 200 | 100 | 67 | 3x |
| 6 | 600 | 300 | 150 | 100 | 3x |
| 8 | 800 | 400 | 200 | 133 | 3x |
| 10 | 1000 | 500 | 250 | 167 | 3x |
| 12 | 1200 | 600 | 300 | 200 | 3x |
Estimated from the standard write penalties, 2 for a mirror, 4 for single parity and 6 for double parity, applied to a nominal 100 random IOPS per 7200 rpm drive. These are arithmetic from published behaviour rather than measured results, and a real array with a write cache and a filesystem that batches writes will do better than the raw figures in bursts.
Put those numbers in context before acting on them. An eight-drive RAID 6 at about 133 random write IOPS is slow for a database and irrelevant for a file server, because copying a large file is sequential and the network link caps it anyway. The question to ask is what fraction of your writes are small and scattered.
Does any of this matter on a home network?
For sequential transfers, no. A gigabit link tops out around 113 MB per second and a 2.5GbE link around 280, while a single 7200 rpm drive already sustains 150 to 260 MB per second. Any four-drive array of either type fills both links, so the layout you choose makes no difference to how fast a film copies across. The full set of ceilings is in 2.5GbE against 10GbE.
For random operations, it does matter, and here is where it shows up: virtual machine disk images, a database behind an application you self-host, container storage that writes logs continuously, a photo catalogue doing thumbnail generation, and video editing scratch space. All of those are lots of small scattered writes rather than one big sequential one.
There is a third option worth naming before you spend a whole array on the problem. Putting the random workload on flash and the bulk data on parity is usually better than choosing a layout that compromises on both. A pair of NAS SATA SSDs mirrored as a separate pool, or an NVMe drive where the platform allows storage pools on M.2, gives you tens of thousands of IOPS rather than a few hundred, at a cost per terabyte covered in HDD against SSD for a NAS.
What does the capacity gap cost in money?
The drives are the purchase, so the honest way to price this decision is cost per usable terabyte at real drive prices rather than as a percentage of raw.
| Drives | Drive | Drive spend | RAID 10 usable | RAID 6 usable | Per usable TB, RAID 10 | Per usable TB, RAID 6 | RAID 10 premium per TB |
|---|---|---|---|---|---|---|---|
| 4 | Seagate IronWolf 8TB NAS HDD (ST8000VN004) | $1,160 | 16 TB | 16 TB | $72.50 | $72.50 | $0.00 |
| 4 | Seagate IronWolf Pro 12TB NAS HDD (ST12000NT001) | $1,752 | 24 TB | 24 TB | $73.00 | $73.00 | $0.00 |
| 6 | Seagate IronWolf Pro 12TB NAS HDD (ST12000NT001) | $2,628 | 36 TB | 48 TB | $73.00 | $54.75 | $18.25 |
| 8 | Seagate IronWolf Pro 16TB NAS HDD (ST16000NT001) | $4,640 | 64 TB | 96 TB | $72.50 | $48.33 | $24.17 |
| 8 | WD Red Pro 20TB NAS HDD (WD202KFGX) | $6,696 | 80 TB | 120 TB | $83.70 | $55.80 | $27.90 |
| 12 | WD Red Pro 20TB NAS HDD (WD202KFGX) | $10,044 | 120 TB | 200 TB | $83.70 | $50.22 | $33.48 |
Prices are the verified figures used across this site and move often. The premium column is what each usable terabyte costs extra under RAID 10, which is zero at four drives and grows with every pair you add.
Look at the first two rows. At four drives the premium is zero, so RAID 10 there is a free choice on money and a paid one on fault tolerance. By eight drives you are paying roughly a third more per usable terabyte for the mirror layout, and by twelve you are paying two thirds more. That is a lot of money for random write performance you could buy far more cheaply with a small flash pool.
Which should you choose, by reader type?
Five readers, five winners.
- Four-bay NAS holding files, photos or media. Winner: RAID 6. Identical usable capacity to RAID 10, 24 TB on 12 TB drives either way, and it guarantees two failures against one. You give up random write speed you were never going to notice while copying files over a network link that caps at 113 or 280 MB per second.
- Four-bay NAS running virtual machines or a database. Winner: RAID 10. Same capacity, roughly three times the random write throughput, and a rebuild that copies one drive instead of reading three. You give up guaranteed two-drive tolerance, and one time in three a second failure during a rebuild takes the pool, so back it up properly.
- Six to eight bays for bulk storage. Winner: RAID 6, decisively. At eight drives it returns 72 TB against 48 TB on 12 TB drives, a whole 24 TB more, while also guaranteeing two failures rather than one. There is no reading of this where RAID 10 wins.
- Eight or more bays with a genuinely write-heavy workload. Winner: split the array. Run RAID 6 for bulk and a separate mirrored flash pool for the hot data. Spending eight mechanical drives on RAID 10 to get a few hundred IOPS is the expensive way to solve a problem two SSDs solve outright.
- Twelve bays or more. Winner: RAID 6, or RAIDZ3 if the pool is an archive. RAID 10 at this width costs 72 TB of the 144 TB you bought, still guarantees only one failure, and gains you nothing that matters for archival data.
The general rule underneath all five: choose RAID 10 for latency, choose RAID 6 for capacity and guaranteed tolerance, and remember that flash solves the latency problem far more cheaply than mirrors do. Whichever you choose, neither is a backup. Parity and mirrors both protect against a drive dying and neither protects against deletion, ransomware or fire, which is what the 3-2-1 backup rule is for.
How do I check these numbers for my own drives?
The arithmetic is short enough to do on paper. RAID 10 usable capacity is the smallest drive times the number of drives divided by two. RAID 6 usable capacity is the smallest drive times the drive count minus two. Rebuild reads is one drive size for RAID 10 and (n minus 2) times the drive size for RAID 6. Random write throughput is drives times per-drive IOPS divided by the write penalty, 2 for a mirror and 6 for double parity.
Run your exact set, including mixed drive sizes, through the RAID capacity calculator, which puts RAID 10, RAID 6, RAIDZ2, SHR-2 and the rest side by side on the same drives. It is the only calculator we know of that is not owned by a hardware vendor, which is why it will happily tell you that your favourite layout is the wrong one for your bay count.
Related reading
- RAID 5 against RAID 6 for the parity count decision underneath this one
- RAID levels explained for every layout in one place
- HDD against SSD for a NAS for the cheaper way to buy IOPS
- Best 4-bay NAS for the hardware where this choice is usually made
- Usable capacity chart for precomputed figures at every bay count
Deciding on a specific model? We review the Western Digital Red Pro 24TB NAS hard drive review in full.
Frequently asked questions
Is RAID 10 better than RAID 6?
For random writes and for rebuild gentleness, yes. For capacity and for guaranteed fault tolerance, no. At four drives they give identical usable capacity and RAID 6 tolerates two failures guaranteed against RAID 10 one, so RAID 6 is strictly better there unless the workload is random writes. At eight drives RAID 6 gives six drives of capacity against four, which is a 50 percent difference on the same hardware.
How many drives can RAID 10 lose?
One guaranteed, and up to half the array if you are lucky. RAID 10 is a stripe across mirrored pairs, so it survives any failure that does not take both halves of the same pair. After the first drive dies, one of the remaining members is its partner, so a second random failure kills the array roughly one time in (n minus 1). At four drives that is a one in three chance, which is not a comfortable number.
Why does RAID 10 rebuild so much faster?
Because it copies rather than reconstructs. Replacing a failed RAID 10 member means reading one surviving drive, its mirror partner, and writing a straight copy. RAID 6 has to read every surviving member end to end and recompute the missing blocks. On an eight-bay array of 20 TB drives that is 20 TB of reads from one drive against 120 TB of reads spread across seven, which is a completely different load on the rest of the array.
Should I use RAID 10 on a four-bay NAS?
Only for a random write workload such as virtual machine disks or a database. At four drives RAID 10 and RAID 6 both give you two drives of usable capacity, so RAID 10 costs nothing in space and buys write performance, but it drops guaranteed fault tolerance from two drives to one. For a file server, a media library or photo storage, RAID 6 is the better trade at the same capacity.
Does RAID 10 performance matter on a home network?
For sequential file transfers, no. A gigabit link caps at about 113 MB per second and a 2.5GbE link at about 280, and any four-drive array of either type fills both. The RAID 10 advantage is in random operations per second, which show up as responsiveness in virtual machines, databases and busy metadata workloads, not as faster copies of large files across the network.
What about RAID 10 with more than two drives per mirror?
A three-way mirror in each group makes any single mirror survive two failures, which removes the RAID 10 gamble entirely, and costs two thirds of your raw capacity. Six drives configured that way give you two drives of usable capacity against four under RAID 6. It is a defensible layout for a small very high value pool, and an expensive one for bulk storage.
How we choose: we compare published manufacturer specifications, drive datasheets, published reliability statistics and verified owner reviews. We do not test hardware in person, and we are not tied to any NAS vendor. Capacity and power figures here are researched guidance, not a warranty. RAID protects against drive failure, not against deletion, ransomware, fire or theft, so keep verified backups regardless of what any calculator tells you.
Working out your own cost per usable terabyte? The Home Server Build Planner is the paid version of these pages: 8 printable worksheets you fill in with your own numbers, plus the full PDF, $29.