How common is a catastrophic failure of an SSD?

[Bobbysixjp]

My Samsung 980 has suddenly died (code 10 error in device manager and is not recognised in disk manager, etc).

So this got me wondering how common total failures like this are and what can be done to prevent them? 

[Kilrah]

Sudden failures can and will happen as with anything - not much you can do about it, just bad luck. 

[Oshino Shinobu]

Depends on multiple factors, such as the conditions the drive is in, in terms of temperature, moisture levels etc. The age of the drive and the number of total writes performed on the drive.

 

Typically, without some external factor like a voltage spike or short, drives (both HDDs and SSDs) tend to fail either very early in their lifetime or towards the end, which for HDDs tends to be powered on hours and movement of the actuation arm/read head and for SSDs, tends to be the total number of writes to the drive, as well as age. Between that, catastrophic failure is very uncommon (think less than 1%) without some external cause, though they do happen, which is basically just bad luck.

 

There's nothing you can really do to prevent those kinds of failures, so instead you should focus on having a good backup solution in place for important data you don't want to lose, as the drives are just a method of storing data, some of which cannot be replaced, while the drive can be. Just a note here that RAID or other similar redundancy solutions are not a backup, nor is just copying the data to another drive that you have constant write access to. Either use a cold backup, where the drive is disconnected from the system after the backup completes, or by using a dedicated backup software that locks the drive to any writes outside of backups.

 

Some recommend doing a sort of burn in test on new drives to basically 0 the drive (write a 0 to every byte on the drive) to weed out early failures before putting any data on them.

[Chris Pratt]

Not common.

 

Let's start by saying any drive can fail at any time. That's the #1 mantra of data retention. Anything you want to keep should be in three places, one of them off-site/in the cloud.

 

That said, aside from some sort of manufacturing defect, a drive should last many years. SSDs are more volatile than HDDs. They need trickle power to maintain their data, while HDDs can be completely removed from a computer and put in a safe or something. SSDs also degrade over time. The more you write to them, the shorter their lifespan. 

 

However, the Samsung 980 carries a 5 year warranty and is rated for up to 600 TBW (terabytes written). As such, it should have a decently long life in virtually any normalish usage scenario. Now, if you were using it to mine Chia or something, then yeah, you could expect a drive failure within a month.

[LWM723]

I've had several SSD's fail. I just gave up and went back to reliable HDD's which have very low failure rates and literally can outlast any SSD.

[wseaton]

I have yet to lose a data center grade SSD. Typically the main difference between these and consumer class is the server SSDs have added power protection. 

 

 

[Rocketdog2112]

13 hours ago, wseaton said:

I have yet to lose a data center grade SSD. Typically the main difference between these and consumer class is the server SSDs have added power protection. 

 

 

Do you have examples of these data center grade SSD's?

 

Do they operate at the same speeds as consumer models?

[wseaton]

I stick to Intel or Kingston data center SSDs for server builds requiring SATA and my reputation is on the line. In fact I no longer use RAID with smaller servers because the MTF ratings are much higher for the SSDs than any RAID controller I'm aware of. Proper backups fill in the gaps. Again, data center grade SSDs typically have extra power protection to buffer abrupt power loss. Optane has a solid industry rep.

 

Consumer NVMe is a different issue. Not enough scale experience with those. Storage devices using multiple NVMe sticks are fairly proprietary and likely not the same animal as an enthusiast MB with all kings of devices sharing the power bus.