Hard Drive Reliability: What the Ratings Mean and What Actually Predicts Failure

Hard drive reliability ratings: what MTBF and AFR actually claim

A data sheet's reliability rating usually comes as two numbers: mean time between failures (MTBF) in hours, and an annualised failure rate (AFR) as a percentage. They read like measurements of the drive in front of you. They are not. They are the maker's projection for a population of drives running under conditions the maker sets out, and those conditions are printed beside the number, often in a footnote.

Here is what three enterprise data sheets state, and the conditions each one attaches to its figures.

Three makers, three drive lines, one rating: 2.5 million hours and 0.35%. The agreement tells you less than it seems to. The Toshiba and WD figures belong to a drive kept at or below 40°C and given a set amount of work, and both makers say what happens outside that envelope: WD derates the rating, and Toshiba says sustained operation above 40°C may degrade reliability. Seagate prints the Exos X24's AFR as a rating at full 24x7 operation; we have not found a workload condition on that sheet.

WD's Red Pro sheet is the most open about method. It prints the Red Pro's MTBF as a projected value, and says that when final, MTBF figures 'are based on a sample population and are estimated by statistical measurements and acceleration algorithms' under a workload of 220 TB a year at a drive temperature of 40°C. Above those parameters the MTBF is derated, up to 550 TB of writes a year, and the footnote ends: 'MTBF do not predict an individual drive's reliability and do not constitute a warranty.' The HC580 sheet carries the same two caveats for both its MTBF and its AFR.

Two consequences follow. A rating is a statement about many drives, so it cannot tell you whether your drive will fail. And a rating is only as relevant as the match between the maker's conditions and yours: a drive run hotter or harder than its sheet assumes is outside the number.

Workload rating: the condition most buyers skip

Workload is the amount of data a drive handles in a year, and it counts more than writes. Toshiba defines it as 'the amount of data written, read or verified by commands from host system'. WD's Red Pro sheet gives the arithmetic: annualised workload rate equals terabytes transferred multiplied by 8,760, divided by the recorded power-on hours. A media server that streams every evening is doing rated work even when nothing new is being stored.

The formula lets you check a drive you already own. Say it has been powered on for 4,380 hours, half a year, and has transferred 150 TB: 150 multiplied by 8,760 and divided by 4,380 gives an annualised 300 TB a year. Against the Red Pro sheet, that is above the 220 TB a year WD uses as the typical condition behind its MTBF, and below the 550 TB a year up to which it derates.

The enterprise sheets work on the same basis. The MG10's rating assumes 550 TB a year, and the HC580's MTBF and AFR derate above their parameters up to 550 TB a year. Whether a ceiling of that size binds depends on your own annual figure, which the formula above gives you.

Hard drive reliability ratings by part number: the IronWolf Pro 20TB pair

Seagate's IronWolf Pro 20TB data sheet shows why a rating belongs to a part number rather than to a capacity or a family name. It prints two 20 TB drives side by side, with different ratings.

Same maker, same capacity, same family name, and an MTBF that differs by more than a factor of two. The 18 TB pair on the same sheet repeats the pattern: 550 against 300, and 2,500,000 against 1,200,000 hours. An Amazon.co.uk listing titled only IronWolf Pro 20TB has not told you which of the two ratings you are buying. The part number in the title, or on the drive's label once it arrives, does.

How to read an AFR, and what 0.35% means in practice

Backblaze gives the plainest reading of the percentage: a drive model with a 1% annualised failure rate means you can expect one drive out of 100 to fail in a year.

Apply that to the makers' 0.35% and a fleet of 1,000 drives inside the rated conditions would expect three or four failures a year. MTBF says the same thing another way: 8,760 hours in a year divided by 2,500,000 hours is 0.0035, which is the 0.35% AFR printed beside it. The two numbers on the sheet are one rating written twice.

For four drives rated at 0.35% and run inside the rated conditions, the same arithmetic gives about 0.014 expected failures a year, a number that says almost nothing about whether one of your four will fail this year. That is the point of WD's warning that MTBF does not predict an individual drive's reliability.

A maker's AFR is a rating. Backblaze's AFRs are measured: failures counted in its own fleet, scaled to a year of drive running time, from a snapshot it takes of each operational drive in its data centres every day. The two kinds of number can be compared only loosely, because the drives, the workloads and the years all differ.

Hard drive reliability stats from Backblaze: what SMART catches and misses

SMART is the drive's own record of trouble, and Backblaze has tested how far it predicts failure in its fleet. It records five SMART statistics against drive failure, and in a 2016 analysis, posted on 6 October 2016, 76.7% of its failed drives had one or more of those five above zero.

The other side of the same result is the more useful one: in that 2016 analysis, 23.3% of failed drives showed no warning from the SMART statistics Backblaze records. SMART is a partial predictor. A non-zero count is a reason to act; a clean report is not a guarantee.

In practice that gives two rules. Treat any tracked statistic moving above zero as a reason to check your backups and plan a replacement, and do not treat a clean SMART report as proof of health.

The analysis dates from 2016 and describes the drives in Backblaze's fleet at that time, so treat its percentages as one fleet's evidence from that period rather than a current rate for every drive.

Temperature: what Backblaze found and what the makers rate

Here the sources differ in emphasis, so this page prints both. In a post dated 12 May 2014, Backblaze's analysis of its drives concluded: 'Overall, there is not a correlation between operating temperature and failure rates.' It added that as long as drives run well within their allowed range of operating temperatures, keeping them cooler does not matter.

The makers draw their line at 40°C. Toshiba says continual or sustained operation with the HDA surface above 40°C may degrade product reliability, and WD derates the HC580's MTBF and AFR above 40°C device-reported temperature, up to 60°C. The statements fit together better than they first appear: Backblaze's conclusion is about drives inside their allowed range, and the makers' ratings describe what that range assumes.

The temperature study is from 2014, so it speaks for the drives of its time. The practical reading holds either way: keep drives well within the operating range on their sheet, and do not expect extra cooling below that range to buy extra life.

Age and the bathtub curve

Age gives the clearest signal in Backblaze's data. In its post of 15 October 2025 revisiting the bathtub curve, Backblaze reports from its 2025 data that drives between zero and one year old barely crack 1.30% AFR, and that its peak is 4.25% at 10 years and 3 months.

Two readings follow. A first-year rate near 1.3% sits well below the peak, so in this fleet early failures are not the dominant risk. And the peak arrives at about ten years, long after the five-year limited warranty printed on, for example, the Exos X24 sheet has run out.

Backblaze's fleet is its own: its drives, its workloads, its data centres. Ages, rates and peaks from one operator are evidence about how drives age, not a forecast for the drive in your server.

Hard drive reliability by brand: why this page does not rank

Searches for hard drive reliability by brand expect a league table. The ratings on this page are per part number, and the IronWolf Pro pair shows a single family carrying two different ratings at one capacity, so a brand-level figure would blur exactly the detail a buyer needs.

Measured rates have the same problem in a different form. Backblaze's numbers come from the drives in its own data centres, recorded day by day, so how a brand looks depends on which of its models are in the fleet and how old they are. A brand average built from those numbers would describe a purchasing history as much as a maker.

Backblaze's measured rows by model, each with its period named, are the subject of hard drive failure rates, and the latest quarter's headline figures and report mechanics that of Backblaze drive stats. The ranking question itself belongs to most reliable enterprise hard drive.

Hard drive reliability vs SSD

In a 2021 comparison, posted on 30 September 2021, Backblaze set SSDs against hard drives used as boot drives in its own servers. Its conclusion was cautious: 'using the failure rate as a factor in your decision is questionable'. It also flagged a structural limit, that the SSD and HDD cohorts sat at very different points in their lifecycle, and age, as above, moves failure rates.

So on Backblaze's 2021 evidence, failure rate is a weak reason to choose between an SSD and a hard drive. Capacity, workload and price per terabyte are the stronger reasons, and the workload rating on the sheet matters for either kind of drive.

Using reliability data when you buy a hard drive in the UK

• Match the rated workload to your use. Work out your annual terabytes with WD's formula and choose a drive rated above it; at 20 TB the IronWolf Pro line alone offers both a 550 and a 300 rating.
• Read the part number, not the family name, because the rating belongs to the part number.
• Keep the drive inside the temperature on its sheet: the Toshiba and WD 0.35% ratings assume 40°C or less.
• Watch the SMART statistics Backblaze tracks, and act on a rising count rather than waiting for a failure.
• Treat every Backblaze figure as one fleet's evidence, with its period named.
• Do not read a warranty as a reliability promise. WD states that MTBF does not constitute a warranty, and the five-year term on sheets such as the Exos X24 is the maker's limited warranty.

This guide lists no prices. The drives whose sheets it quotes are priced from Amazon.co.uk, in pounds inc VAT, on the Exos X24, Toshiba MG10 and Ultrastar DC HC580 pages, and enterprise hard drives explains how the enterprise class differs from NAS and desktop drives. Service life and replacement timing are the subject of the English drive lifespan guide.