So I upgraded my Sandisk sata m.2 to a 1tb 960 EVO and my 500GB 860 EVO to a 1TB 860 EVO in my laptop and man… Nvme drive speeds are no joke. Using swapspace becomes almost a reasonable thing to do
Anyway, quite apart from this thing being f*cking awesome I did notice one thing: It appears that the 860 EVO has a 5 year warranty and a 2.4PB write limit while the 960 EVO has a 400TB write limit and a 3 year warranty.
How worried about this should I be? Will the drive actually just fail at 400TB? I’ve read some tests that suggest that a 256GB 960 EVO can survive up to 6PB writes. I don’t really know if that is a silicon lottery thing though.
Back in 2015, some goons did a non-scientific year, and found the nand chips on consumer SSD’s were pretty remarkable.
They were probably MLC rather than TLC, so your mileage may vary, but write endurance (while finite) isn’t really the thing that will fail you
Yeah, I’m mostly just worried about the drive suiciding on me when it hits the write threshold on the box. But it appears that they don’t do that so I’ll probably be fine.
Yeah, [Edit flash chips wearing out] not likely to happen, especially during the warranty period, if used as an actual storage drive, where one typically stores less than 20GB/day (like one large steam game a week, plus a couple of movies, and then some random files)
I would only really worry about endurance in either cache drives, or scratch drives for video editing.
More likely the controller will give way unexpectedly, but hopefully not…
I have a couple of old Vertex 3 drive from like 5? Years ago, and they are slowly chugging along well after their 2 year warranty. Actually, I should check their realloc count…
I have an SSD from 2013 that’s still working fine.
If you want you can try to figure out how much you write on average and algebra it out to see how long it will take you to hit that mark.
They probably base the write endurance on three standard deviations or something, so even the “bad” drives should be able to hit that write endurance mark.
I seem to have written about 800GB in 4 days. So I assume then that I should be fine, I was more worried about the 400TB being some kind of hard limit after which the drive would self-destruct. It appears that they only do that if they actually run completely out of spare capacity. I’ll just keep an eye on it and replace the drive when the spare capacity reaches 15% or so.
The writing is less of an issue than the erasing.
Obviously one would write more to it at first, setting it up and installing all of Steam on it.
I would only pay attention if I erase and write a lot to it, butting doesn’t hurt to keep an eye on it over a few months to see how it goes.
Samsung magician is pretty straightforward for showing the basic info, and has a couple of useful tools for reconditioning, over-provisioning etc.
If I understand correctly, once the flash memory on The SSD reaches the point at which it is worn out, it functions as read-only. The data that is already written will remain intact, but no new data can be written to the drive.
I’m pretty sure that’s how it works, and I think it does it on a page/block level.
So the wear levelling should make each block wear mostly evenly, and then stop using sections as they wear out too much…
usually it is about 600-900 writes pr. cell. but most these drives intelligently sort the near death cells out by moving the data located in dying cells to a reserved area somewhere else on the disc.
so unless you’re defragmenting your ssd on a daily basis i wouldn’t worry, they’re quite durable, to the point where spinning rust now a days proberly die faster.
That’s how it’s supposed to work, but in reality every drive handles it differently. Some just stop working entirely and you lose all your data. So don’t rely on that.
The 840 (non-pro) in that test is TLC, and it hit 800+TB just fine. To wear that NAND out inside the 3 year warranty period you’d have to write 730GB per day.
The 840PRO in that test has a 5y or 73TBW warranty. It lasted 2.4 PB.
They had 2 Kingston HyperX 3K 240GB drives. Those are rated for 203TBW, one lasted 800TB, the other made it past 2PB
Those were much older drives than what’s available now.
So yeah, I wouldn’t worry too much about a 400TB warranty. You’re not likely to hit 400TBW during the 3 year warranty period (unless you write more than 365GB/day) , and it’s nowhere near what your drive’s actual point of failure will be. You’ll most likely upgrade to a larger one long before the NAND on this one is anywhere near half-worn.
I believe some vendors (Samsung?) are offering up to 5 (or is it 10) year warranty on TLC nand these days, at some absolutely colossal re-write rate.
The short version is that if you buy a decent brand SSD these days, it will likely be too small or too slow for you well before it does die to write endurance issues in an end user desktop or laptop. It may suffer some other form of failure (as always, don’t trust any storage media exclusively with stuff you can’t afford to lose), but NAND cells wearing out is just not likely.
Basically you won’t be able to generate enough data to destroy it, and you’re likely not on a fast enough internet connection to download enough that fast either.
edit:
As per the link above, there has been SSD torture tests performed to deliberately try to kill previous generation SSDs and they lasted many months of constant full re-writes, 24/7… even in a typical server environment you’re just not likely to be doing that.
Don’t worry and have a tested backup strategy in place for important data.
My 2 Samsung NVMe SM951 drives have been in continuous use for almost 3 years and both have only clocked up 4500+ hrs of power on time and 17 TB of writes. I was a bit cautious in the first few weeks but quickly forgot about the whole write life and just used them just as I would use a HDD.