Hard drive failure rates
Backblaze logs every drive that dies in its data centres and publishes the lot, and it's the best public reliability data there is for spinning rust. We've boiled it down to one failure rate per model, with an error bar, for the 36 models that have enough drives behind them.
- Drive models
- 36
- Drive-years
- 1,542,500
- Failures
- 21,804
- Median AFR
- 1.40%
Lifetime AFR by model
Lowest first. The dot's the measured rate and the bar is the 95% interval around it. A long bar means not many drives or not many years, and the real rate could be some way from the dot. Drives we list are in blue and link through to their pages.
Drives we listOther models in Backblaze's fleet
Median of these 36 models: 1.40%
Source: Backblaze. Backblaze Drive Stats, 2015 to 2026 Q2.
The numbers
Lifetime figures, plus 2026 Q2 where Backblaze had enough of the model running to say anything useful.
| Model | TB | Drives | Drive-years | Failures | AFR | 95% CI | 2026 Q2 AFR |
|---|---|---|---|---|---|---|---|
| ST8000NM000A | 8 | 237 | 840 | 3 | 0.36% | 0.07%–1.04% | 0.00% |
| Toshiba MG11 24TBTOSHIBA MG11ACA24TE | 24 | 9,597 | 4,642 | 23 | 0.50% | 0.31%–0.74% | 0.58% |
| WD Ultrastar DC HC550 16TBWDC WUH721816ALE6L4 | 16 | 26,208 | 84,197 | 452 | 0.54% | 0.49%–0.59% | 1.13% |
| WDC WUH721414ALE6L4 | 14 | 8,548 | 46,924 | 254 | 0.54% | 0.48%–0.61% | 1.07% |
| WD Ultrastar DC HC570 22TBWDC WUH722222ALE6L4 | 22 | 45,556 | 72,927 | 423 | 0.58% | 0.53%–0.64% | 0.56% |
| ST16000NM002J | 16 | 463 | 1,629 | 10 | 0.61% | 0.29%–1.13% | 3.47% |
| HGST HUH721212ALE600 | 12 | 2,591 | 17,401 | 112 | 0.64% | 0.53%–0.77% | 1.24% |
| Seagate Exos X16 16TBST16000NM001G | 16 | 34,998 | 129,390 | 877 | 0.68% | 0.63%–0.72% | 0.66% |
| Toshiba MG10 20TBTOSHIBA MG10ACA20TE | 20 | 25,171 | 25,546 | 201 | 0.79% | 0.68%–0.90% | 1.10% |
| WDC WUH722626ALE6L4 | 26 | 6,958 | 2,018 | 17 | 0.84% | 0.49%–1.35% | 0.95% |
| Seagate Exos X16 12TBST12000NM001G | 12 | 13,072 | 72,056 | 723 | 1.00% | 0.93%–1.08% | 1.35% |
| Toshiba MG07 14TBTOSHIBA MG07ACA14TA | 14 | 36,925 | 214,759 | 2,293 | 1.07% | 1.02%–1.11% | 1.94% |
| TOSHIBA MG08ACA16TA | 16 | 39,522 | 113,755 | 1,215 | 1.07% | 1.01%–1.13% | 1.03% |
| ST12000NM000J | 12 | 1,079 | 2,002 | 22 | 1.10% | 0.69%–1.66% | 0.00% |
| WDC WUH721816ALE6L0 | 16 | 2,941 | 13,519 | 165 | 1.22% | 1.04%–1.42% | 4.34% |
| Toshiba MG08 16TBTOSHIBA MG08ACA16TE | 16 | 6,228 | 27,707 | 341 | 1.23% | 1.10%–1.37% | 1.03% |
| TOSHIBA MG09ACA16TE | 16 | 893 | 465 | 6 | 1.29% | 0.47%–2.81% | 1.07% |
| Seagate Exos X16 14TBST14000NM001G | 14 | 10,455 | 57,467 | 795 | 1.38% | 1.29%–1.48% | 1.50% |
| HGST HUH728080ALE600 | 8 | 103 | 8,936 | 126 | 1.41% | 1.18%–1.68% | 3.85% |
| ST8000DM002 | 8 | 5,968 | 92,271 | 1,418 | 1.54% | 1.46%–1.62% | 2.46% |
| TOSHIBA MG07ACA14TEY | 14 | 982 | 3,630 | 64 | 1.76% | 1.36%–2.25% | 2.46% |
| HGST HUH721212ALE604 | 12 | 13,057 | 69,066 | 1,224 | 1.77% | 1.67%–1.87% | 3.37% |
| ST14000NM000J | 14 | 496 | 710 | 13 | 1.83% | 0.97%–3.13% | 0.00% |
| ST8000NM0055 | 8 | 12,874 | 126,874 | 2,516 | 1.98% | 1.91%–2.06% | 3.47% |
| ST12000NM0008 | 12 | 18,363 | 122,422 | 2,611 | 2.13% | 2.05%–2.22% | 3.17% |
| ST16000NM000J | 16 | 167 | 226 | 5 | 2.22% | 0.72%–5.17% | 2.57% |
| ST12000NM0007 | 12 | 957 | 102,502 | 2,307 | 2.25% | 2.16%–2.34% | 4.17% |
| TOSHIBA MG08ACA16TEY | 16 | 4,789 | 23,219 | 565 | 2.43% | 2.24%–2.64% | 3.82% |
| HGST HUH721212ALN604 | 12 | 9,397 | 76,099 | 1,858 | 2.44% | 2.33%–2.56% | 7.63% |
| Seagate Exos X24 24TBST24000NM002H | 24 | 9,482 | 9,975 | 302 | 3.03% | 2.70%–3.39% | 3.47% |
| Seagate Exos X18 18TBST18000NM000J | 18 | 66 | 347 | 11 | 3.17% | 1.58%–5.67% | – |
| ST10000NM0086 | 10 | 936 | 10,008 | 329 | 3.29% | 2.94%–3.66% | 9.75% |
| WD Ultrastar DC HC550 16TBWUH721816ALE6L4 | 16 | 531 | 448 | 22 | 4.91% | 3.08%–7.43% | 4.87% |
| HGST HUH728080ALE604 | 8 | 70 | 367 | 21 | 5.72% | 3.54%–8.75% | – |
| ST14000NM0138 | 14 | 1,211 | 7,885 | 463 | 5.87% | 5.35%–6.43% | 8.27% |
| ST14000NM0018 | 14 | 61 | 270 | 17 | 6.29% | 3.66%–10.07% | – |
Method
AFR = failures ÷ (drive-days ÷ 365) × 100. Every drive that runs for a year adds a drive-year to the total, whether it died or not.
Confidence interval. Failures are treated as a Poisson count, and the interval is the exact (Garwood) 95% one, which takes the χ² quantiles at 2.5% with 2k degrees of freedom and at 97.5% with 2k + 2, then halves them and divides by drive-years. With no failures at all the lower bound is 0, though the upper bound still tells you something.
What we publish. A model needs at least 60 drives and 50 drive-years behind it for a lifetime rate (25 for a single quarter) before it goes on here. Below that the interval's too wide to help anyone. We also leave out anything under 4 TB, which is mostly Backblaze's boot SSDs.
Matching to products. A Backblaze model string only gets matched to a drive we list when the normalised model number is identical, like TOSHIBA MG08ACA16TE to MG08ACA16TE. Close variants such as MG08ACA16TEY and MG08ACA16TA stay separate, because they're different part numbers.
Limits. Backblaze runs its drives in data-centre pods, 24 hours a day at a steady temperature, while a home NAS sees fewer hours and more spin-ups. The rates are good for comparing models. They won't tell you when your own drive's going to die. Backblaze also only reports on what it buys, and none of the 4 slower, 5,400 rpm-class drives we list turn up in its data.
Source and licence. Every figure here is derived from Backblaze Drive Stats. Backblaze asks anyone using the data to credit it, and we do on every page that shows it. We publish the derived rates and leave the raw data with Backblaze.