I recently had to RMA a Samsung 840 EVO that was less than 3 years old, and had only seen light usage (but no extended offline time). It just suddenly stopped responding to ATA commands, returning errors. All the SMART attributes were fine and the drive even gave itself a clean bill of health. Distressingly, even as it was returning read/write errors like crazy, the SMART read/write error attributes stayed at 0. It was running the latest firmware. I could not even do an ATA secure erase to reset it to clean condition, as that command returned an error as well.

I sent it out for RMA, they returned it with the work description "firmware upgraded" and no further information. It's the same drive, but it's been zeroed/factory-reset.

I have backups, so I'm not too concerned about losing data, but this episode didn't exactly make me confident in the quality of SSD firmware, especially since samsung is supposed to be one of the better manufacturers out there.

Just one anecdote, make of it what you will. Of course, I've had countless hard drives fail over the years as well, but the failure mode is very different, there's no partial read possible and it just happened all of a sudden with no indication.

Close to 100% of the SSD failures I've seen have not been from being left off for a few years, nor from actual flash media failure, but from firmware bugs or, occasionally, mechanical failure.

(I've had very small sample size of SSDs not being powered up for a few years, but the few I have powered up after not powering on for years have had all their data retained. Ask me in another 15 years or so, though.)

Part of what you count as firmware bugs may have been cases of data/firmware corruption caused by unexpected power loss.

That's a firmware bug, then.

I've written a few flash storage systems (small ones, for consumer devices). If your data structures are such that you can't do an atomic commit, you shouldn't be in the marketplace.

Sudden loss of power is not an excuse.

There was some widely circulated study few years ago which found that most SSD models they tested had problems on power loss. Some even bricked. They didn't publish concrete names, unfortunately.

But yeah, I agree it's a bug and shouldn't happen. So what :)

edit:

There we go: https://www.usenix.org/system/files/conference/fast13/fast13...

Some more: http://lkcl.net/reports/ssd_analysis.html

Note that the 320 recommended by the second article had similar issues, supposedly fixed in firmware 4PC10362, but some users reported problems even on that version: https://communities.intel.com/thread/24339?start=15 (many of those complaints look like different problems, but not all).

I had corruption issues with an OCZ drive on a machine with unstable power. Since adding a UPS the drive has been rock solid.

We had corruption / sudden death issues with OCZ even on machines with redundant power supplies. I do not recommend them.

My "favorite" were the OCZ Vertex 1 drives which almost certainly had issues with the firmware eating paste and overwriting some internal metadata - periodically the drives would start throwing read/write errors on swathes of LBAs, and the only way to mitigate this would be to throw the jumper on to put the drive in "recovery" mode and do a destructive firmware reflash that blew away all the contents.

And then it would work perfectly well again...for a while.

I agree with the downstream commenter about this still qualifying as a FW bug, but to be more clear:

* there's the story I posted about the OCZ V1 drives eating paste periodically while behind redundant power.

* there was a fun problem of some OCZ Deneva 2 drives deciding that they would simply drop off the port and not come back without a cold power cycle, behind redundant power (and a related story about the drives being in such a state, and the AHCI init code on the BIOS getting confused and hanging forever, which was what actually necessitated the cold cycle).

* there was the really fun case someone I worked with came up with where they found a way to reproducibly brick a particular family of SSDs using only about a TB of IO on a 480 GB SSD. (And by "brick", I mean "call the manufacturer, it's not coming back.")

* there was the SandForce bug once upon a time which caused the affected SSDs to not show up correctly as disks under Windows when on SATL because the firmware reported an empty "version" field, and while all the other OSes tested were content with this, and Windows when not using SATL was content with this, Windows when using SATL would report the disk as an "Unknown Device" for this.

> mechanical failure.

How does that happen?

I had my first computer fire, last year. One of the contacts in an SSD power connector delaminated, causing a short. The SSD was toast. Or at least, the power connector had melted. The power supply was toast. But with a new power supply, the machine itself is fine :)

Crushing the SSD disk or whole laptop :)

Connectors are the weak point?

There are brands of SSD that I will not have in our datacenter (medium scale, several thousand machines). Kingston and OCZ are the ones that I kill on sight. I found some new-in-box drives in a corner a week or two ago, and threw them away. They're poison.

For developer workstations, we get good results from Samsung drives. We run Intel enterprise class drives in servers and they're fine, don't even have to think about them.

We've pretty much stopped buying spinning drives, with a few exceptions where we need lots of capacity and don't care about performance.

> have you seen case of SSD data loss due as result of not powering up for couple years?

In normal circumstances an SSD is not going to lose data if left powered off for two years. [1]

[1] http://www.anandtech.com/show/9248/the-truth-about-ssd-data-...

All in all, there is absolutely zero reason to worry about SSD data retention in typical client environment. Remember that the figures presented here are for a drive that has already passed its endurance rating, so for new drives the data retention is considerably higher, typically over ten years for MLC NAND based SSDs. If you buy a drive today and stash it away, the drive itself will become totally obsolete quicker than it will lose its data. Besides, given the cost of SSDs, it's not cost efficient to use them for cold storage anyway, so if you're looking to archive data I would recommend going with hard drives for cost reasons alone.

> is SSD reliable?

We've returned 25% of the laptops of a particular model because the Kingston SSDs had failed. That's across a sample in the hundreds.

I built a new server for myself six months or so ago. I put in 6 x 4 TB WD Red HDDs and a pair of Samsung 850 Pro 256 GB SSDs for the OS/boot partitions.

I really like those SSDs! I replaced the (about three year old) Crucial M500 480 GB SSD in my ThinkPad W530 with a pair of them and have since put them in every machine I've got (with the exception of my new MacBook Pro).

Personal anecdote, I have had trouble with OCZ. I have had no trouble with Samsung EVO. Now, I go with Samsung EVO for all my drives - they are not much more expensive than the cheap drives and the fact that they just work is worth it to me.

Most problems come from sudden powerloss.