Do SSDs Wear Out in a Couple of Years?

This is one of the myths in Flash Myths, Tested, where each common belief about SSDs gets checked against data, specs and papers. The belief here is that SSDs wear out in a couple of years, and the question is how fast a desktop drive really uses up its life.

My 4TB NVMe system drive has run for 10,726 power-on hours and written 42.9TB, about 96GB a day, and it reports 1% of its life used. Using up the ratings I checked in two years would take 275GB to 5,479GB of writes every day, so for desktop use the myth doesn’t hold.

Where the myth comes from

The worry rests on a true fact. Flash cells wear out as they are programmed and erased, and every drive carries a limit. In the Google fleet study the MLC drives had a limit of 3,000 program/eraseA program/erase cycle is one write of a block followed by its erase. Each cycle wears the oxide, and the cell type sets how many cycles it tolerates. cycles, and a number that small sounds like it could run out in a couple of years. Flash Foundations Part 9 explains how makers turn that limit into the TBWTerabytes written is the total host data a manufacturer rates a drive to absorb over its life under the warranty conditions. rating on the box.

The myth leaves out how many cycles real use takes. Below, my own SMART logs come first, then the ratings, then the field studies, and one caution runs through all of it. My data comes from desktop and bench use, and the field studies come from data centers.

What my own drives show

My system drive is a 4TB NVMe SSD that I use every day, and a read-only smartctl capture of its SMART log shows 83,853,719 data units written. A data unit is 512,000 bytes, so that is 42.9TB of host writes, about 10.7 times the drive’s capacity, over 10,726 power-on hours, which is 447 days. That works out to about 96GB a day across everything on the drive, and it reports a Percentage UsedPercentage Used is a vendor estimate of drive life consumed, based on writes and wear. It can pass 100%, and the drive may keep working. of 1%.

Percentage Used is the drive’s own estimate of how much of its endurance is gone. It’s a whole number, 100 means the estimated endurance is consumed, and the value can go above 100. Kingston’s Fury Renegade data sheet ties its warranty to that number: five years or a Percentage Used of 100. So the vendor’s estimate is the number that matters for a warranty claim, and mine reads 1%.

This is one machine, and 447 days isn’t two years. It gives me a rate, and the myth needs to be tested against a rate.

A second set of captures comes from a drive that held the operating system on my test bench. I read its SMART log 126 times across 1,229 power-on hours. Over that time its host writes rose from 17.55TB to 22.19TB, an increase of 4.64TB, about 91GB per 24 hours of power-on time. Its Percentage Used read 1% at every capture.

Figure 1 draws all 126 captures. The climb is steady and gentle, and the wear estimate never moved. I withhold the brands. Figure 1 shows the OS drive with the most captures, out of the four the bench went through.

What new drives show

New drives make up the third set, and when I review one I read its SMART log at the end of the test suite. For 83 new NAND drives, the median total was 16.45TB written, reached in a median of 42 power-on hours, and the range ran from 1.87TB to 57.77TB. I left out the two Optane drives and two records with no usable write count.

Figure 2 bins them. Of the 83 drives, 70 read a Percentage Used of 0% or 1% and the highest read 6%. This shows how hard a test suite writes, not wear over time, because every drive was new when I started. A median suite total of 16.45TB is what my system drive writes in about 171 days.

The drives that read more than 1% are a reminder that Percentage Used is measured against each drive’s own rating, and the ratings of these drives aren’t in my archive, so I can’t say what share of its rating each drive spent. I can say that none of them was close to 100.

What two used drives show

A fourth set comes from a second set of drives from my work bench. It gives me two used drives, both 2TB. I read them with CrystalDiskInfo, so the health figure here is the Health Status percentage that CrystalDiskInfo prints. It isn’t the NVMe Percentage Used counter that the other sets report, and I keep this set apart from them instead of pooling the numbers.

The first drive was read 48 times across 145 to 2,973 power-on hours, and its host writes rose from 7.2TB to 14.3TB, a rise of 7.0TB that averages about 60GB per 24 hours of power-on time. That average hides how it happened, because the writes arrived in steps between sessions on the bench. CrystalDiskInfo health read 100% until the capture at 2,920 power-on hours and 14.2TB written, and 99% at the last 4 captures.

A second drive was read 17 times across 248 to 582 power-on hours. It had 45.7TB written at the first capture and 48.1TB at the last, more than the 42.9TB on my system drive, and CrystalDiskInfo health read 97% at every capture.

This set is thin. It’s two drives. Neither drive’s rated TBW is in my data, and health isn’t Percentage Used, so I can’t say what share of its rating either drive spent. It adds one plain observation. A drive with 48.1TB of host writes still read 97% healthy, and a drive with 14.3TB read 99%. That fits the verdict, but two drives can’t prove it.

What wearing out in two years would take

Take two years as 730 days. A rating is a total, so the daily write that reaches it in two years is the rating divided by 730. TBW is worked out from the flash’s program/erase budget, the drive’s spare area and an assumed write amplification, so you compare it directly with the host writes in SMART, as Part 9 describes.

Figure 4 sets my measured 96GB a day beside three ratings from saved data sheets. Samsung rated the 512GB 950 PRO at 110GB a day for five years, which is about 200TB in total, and two years reaches that at 275GB a day. Samsung rates the 1TB 870 EVO at 600TBW, which is 822GB a day. Kingston rates the 4TB Fury Renegade at 4.0PB, which is 5,479GB a day. I don’t have a saved rating for my own drive, so the three ratings belong to other drives.

Ratings aren’t arbitrary. Kingston says its TBW is derived from the JEDEC client workload, JESD219A, and the SNIA endurance white paper says vendors rate TBW with the JESD219 trace for the drive’s class. SNIA also notes that most consumer use is write once, read many, and that only heavy content creators and power users regularly move tens to hundreds of GB.

What the field studies add

The two big field studies are data-center fleets, which usually write far more than a desktop, so they make a conservative case. Schroeder and colleagues studied drives in Google’s fleet for up to six years. MLC models had averaged 529 to 949 program/erase cycles against a limit of 3,000. The eMLC models averaged 377 to 607 against 10,000, and the SLC models 185 to 860 against 100,000.

Errors grew linearly with cycles, with no sudden spike past the limit in the range they studied, and age mattered on its own. The vendors’ accelerated tests showed a rapid rise only at about 3 times the limit, which led the authors to call the limits conservative. Meza and colleagues, studying Facebook’s fleet, found 14.8TB to 48.5TB written to flash per SSD over average ages of 0.5 to 2.4 years. They also found that failures do not rise steadily with wear, and that the data the operating system writes does not always match the wear on the flash, because of controller optimizations and buffering. That second point is why I read the Percentage Used field as well as the writes.

Where the myth could be true

A couple of years is plausible for one case. Those numbers show it. At 275GB a day, the 512GB drive in Figure 4 would reach its rating in two years, and that is inside the range SNIA calls heavy creators and power users. A small drive under a heavy creative workload can wear out that fast.

Two limits on what I can say: I have no data on drives with the lowest ratings, and I haven’t measured a creator’s workload. And wear is only one way a drive fails. Meza’s finding that failures do not climb steadily with wear is the subject of the myth that an SSD fails with no warning.

What to do

Look at your own drive instead of worrying about it. Read Data Units Written and multiply by 512,000 bytes, then read Percentage Used and Available SpareAvailable Spare is a health value showing the percentage of spare blocks remaining. The drive warns when it falls below a set threshold.. Divide your total by the days the drive has been powered to get a daily rate.

Multiply that rate by 730. If the result is a small fraction of the TBW on your drive’s data sheet, two years will not wear it out.

If you move hundreds of GB a day, check the rating before you buy instead of assuming a small drive will last. For ordinary desktop use, leave the drive alone. The same arithmetic answers the pagefile worry, and it keeps the page file where it is.