I’m shocked at how long I’ve been using this desktop computer! The SSD that I originally installed in it has 50,512 power-on hours count. The HDD that I added just a bit later has 50,197 hours and has already exhibited signs of failure and had some bad sectors sequestered.
The SSD is a Samsung 850 Pro, which unlike the more common EVO models uses MLC (two bits per cell) technology rather than the cheaper TLC or QLC. This is my critical system drive. Of course I have backups but I’m interested in opinions about the lifespan of an MLC type SSD.
That’s difficult to answer without knowing how much data has been written to the SSD. An SSD typically lasts for about 5 to 10 years, with your drive probably towards the high end of that. 50,000-ish hours is a bit over 5 years. So if your PC has seen average use, you are probably about halfway through its expected life. If your PC has seen above average use (at least with respect to the amount of data written to the drive) then your expected life will be less.
Always have backups. The drive could already be dead, and you just don’t know it yet.
The more important field to look at is “Percentage_used.” It might be called something else in the tool you’re using. That’s what it’s called when I use nvme-cli to look. ETA: That is the percent of expected write capacity over the drive’s lifetime, not the space used on the drive.
My most worn drive is at 65,000 hours, 181TB written, and 67% used. I plan to keep using it until it fails, or gives warnings failure is imminent. I have backups, and I know how to use them.
So to answer the questions, it’s hard to know from just power-on hours, but unless you’re seeing signs of failure, the drive probably has lots of life left. I know you measure computer lifetime in decades, so maybe not quite that much.
Thanks for that. In fact, almost all of the data-intensive activity on this computer has been directed to the HDD. Most of the activity on the system SSD has been whatever Windows feels like doing.
IMHO, the power-on time is much less important than the total amount of data written. Unless you have a way to estimate that, the condition of the SSD is unknowable.
That should be fine. CHKDSK operates at the filesystem level, not the drive level. A filesystem can become corrupt while the drive is still perfectly fine. The filesystem becoming corrupt might mean the drive is failing, but can be caused by other things like unclean shutdowns or bugs on the OS.
Reading an SSD does not cause wear the same way that writing to it does, run CHKDSK all you want. My recommendation is to just use the SSD, and don’t worry about it. Unless you have a very edge case of usage (like a security camera DVR), it should be fine, you’ll get many years out of it.
Every single SSD I’ve had that has failed has been premature. I’ve never, ever had one fail because it was “used up”.
As alluded to above, the dominant determinant of lifetime is the number of writes. Enterprise level SSDs tend to have significantly more provisioning of extra space, which can significantly improve performance of the wear levelling system. But eventually you run out. If the OS virtual memory system needs to swap to your SSD you can burn through writes. Mostly such activity is a thing of the past, with enough RAM. But you can be caught out. Same with logging activity. Depending on system config there can be a lot more system disk activity, and actual data writing, than one might imagine.
Samsung have their own proprietary software tool that can tell you a lot about the drive: Magician. The description unhelpfully only lists a few 900 series drives as supported, but the actual release notes include the 850 Pro. So probably a good step.
Magician, once installed, wants to load at system start and maintain continuous diagnostics of your Samsung storage, but it’s not really required if you just want to use it manually once in a while. Automatic startup can be configured “off”.
It’s actually erasing and not writing that causes wear on flash memory. It’s kind of a pedantic point since a block must be erased before it can be written, although it’s not one-to-one because usually the erase block size is larger than the write block size. For example, I worked with a flash chip in which an erase operation would erase a minimum of 128K, while a write operation could overwrite as little as 16K of erased space.
I’m decommissioning some old server hardware this afternoon, and came across a 256GB Samsung 850 Pro drive, so that caused me to look at this a bit more deeply. The drive claims its Power_On_Hours at either 099 (value) or 158,000 (raw). Neither of which are possible to be a measure of “hours”. The drive has been turned on since I bought it, possibly in 2015, but that is considerably less than 18 years.
So 65,000 might not be an accurate representation of how long you’re drive has been turned on. Fortunately power on hours is not an important detail for an SSD.
The attribute of importance is something called “Wear leveling count”, which a search tells me starts at 100 and goes to 0. My current value is 66. I think, if I’m doing my math right, the drive has had 236TB written to it. The drive did contain a mail spool for many people, so that is lots of writes.
This article I found shows that in a durability test an 850 Pro managed 9100 TB before failing.
I think the conclusion is that an 850 Pro might not last until the sun expands to swallow the earth, but it might last for the rest of our lifetimes.
Very informative post, thank you. In fact I was just recalling earlier that the wear leveling count is available through any number of disk monitoring utilitiies (I use HD Tune). But I don’t believe your description of it is correct. It’s defined as the maximum number of erase operations performed on a single flash memory block. I seem to remember it being 0 when the drive was first installed and then gradually climbing. It currently sits at 237. Since AIUI the internal electronics is always wear-balancing, this means that pretty much each cell has been erased around 237 times. (ETA: I’m not certain but I believe wear-leveling occurs across the entire drive, regardless of what data is on it – the data is just moved around to different cells.)
Total data written is 1,253,591,262, which I take to be just under 1.254 TB. That seems a bit low but it might be right. The SSD is primarily a system drive and there is a big mechanical drive for data-intensive stuff. A bunch of other parameters indicating different kinds of error count are all 0.
That is possible, and we both can be right, because each entry has a “VALUE” and a “RAW_VALUE”. Supposedly the VALUE portion is a count down from 100, and the RAW_VALUE counts up. My RAW_VALUE is at 2018.
Flash memory is specified for a minimum number of erase/write cycles that still leave the cell with data retention of 10 years. If you only care that the data is retained for 1 year, then the number of cycles increases dramatically. If your definition of failure is immediate data loss after write, then you can get a tremendous number of cycles, but just be aware that long before that number is achieved, the drive will be pretty unusable.
A paper that discusses some strategies for increasing the number of cycles by relaxing retention:
Those labels vary with the particular tool you’re using. Under the “Health” tab, HD Tune lists 5 columns labeled “Current”, “Worst”, “Threshold”, “Data”, and “Status”. I have never understood what any of those columns are supposed to mean except “Data”, which contains the actual information you need to know. Based on the amount of data written to your SSD, a wear leveling count of 2018 is quite plausible.
Just to nitpick further on this point, a “countdown” from 100 seems to imply that the drive knows exactly when it’s going to fail. And furthermore, what does “100” mean? “100” what?
In my equally mysterious HD Tune statistics, I have values of 96 in both the “Current” and “Worst” columns. I consider these to be totally meaningless. The “237” as the number of erase cycles in the “Data” column makes sense. If your 850 Pro has 2018 cycles and was still working fine, that’s very reassuring, along with the aforementioned 9100 TB write durability.
I believe my question has been answered!
Though I’m not sure I’m necessarily overjoyed at the thought that my Samsung 850 Pro may outlive me!
ETA: I have various brands of SSDs on all three of my current laptops, in addition to the Samsung Pro on my main desktop and some cheap SSD on a secondary desktop. All of them are working just fine. The one on my oldest laptop was creating intermittent problems on boot, which got fixed with a firmware upgrade to the SSD. I have the impression that concerns with wear on SSDs are generally overwrought, since other things, like entirely new computers, will almost certainly happen first.
I have no idea if I’ll look more into this later, but I will say I have yet to have an SSD fail on me. Early on I was very careful and had perfect backups on an HDD, but I’ve not gotten around to that. I do have stuff backed up on a cheap cloud backup, but with a couple computers, I’m not as fussed about needing to be able to boot a particular one.
I have a Samsung drive and use their management software (Magician). One thing it does is offer over provisioning. When you setup the drive you can specify a certain percentage to be blocked off and remain unused (defaults to 10% but you can change it to whatever you want).
The drive will then start using the unused portion of your drive that you blocked off and is, basically, as good as new as bits of the used portion of the drive go bad.
This will extend the life of your drive a fair bit and happens behind the scenes. No input from you beyond the initial setup (which is basically pressing a button).
I do not know if other manufacturers have this and/or if there is drive management software you can get that does this. Worth a look though.
My wife mentioned some of the tills at the store where she works had started to go die - the SSD’s failed in less than a year. Tech support did not specify exactly, but blamed it on the till software which rewrote continuously. I was trying to imagine just how much data one would have to write how fast to kill an SSD within a year. Seems more like they got a bad batch of SSD’s.
