Pro Evo-lution shocker: Samsung SSDs focus on endurance over capacity Samsung's new 860 Pro and Evo SSDs have virtually the same capacity and performance as the 850 line, but with much higher endurance. The 850 drives were launched in 2014 and boasted 128GB to 2TB capacity in 2.5-inch form factors with a 6Gbit/s SATA interface. They used 32-layer V-NAND, Samsung's version of 3D NAND – 2bit/cell …
I thought it was already well established that endurance, in terms of number of write-cycles is already well in excess of what consumer devices would ever reach (I thought there was an El Reg article on that, can't find it, but see e.g. Grueling endurance test blows away SSD durability fears.
Okay, you can't have too much of a good thing, and maybe these things will find themselves in some enterprise workloads. But all the SSD failures I've seen, which is several, including the 1TB Samsung in my own laptop, have not been because the drive got within a million miles of its rated write cycles, but because they just decided to stop working one day. "Boot device not found". Game over.
So by all means work on the endurance, Samsung, but if you could make it so they don't just keel over and die for no apparent reason as well, that'd be great!
(Note: I'm not saying Samsung has any kind of reliability problem, quite probably their failure rate is good, and we just got unlucky. Just pointing out the obvious, that non-wear-related failures are a thing.)
> "they just decided to stop working one day. "Boot device not found". Game over."
Momentary power loss? Power loss at an inopportune moment (i.e. writing page table) can really ruin an SSD owner's day. Some drives have capacitors to try to help avoid this (most Crucial's), others don't (at least some Samsungs).
Endurance and reliability are both within or exceeding the ranges that non-SSD drives do.
Speed too.
Capacity? Nope.
Price? Nope.
So why they would not bring down prices, or raise capacities, but give more endurance and speed? I can't fathom. Just STOP making traditional HDDs except for high-end server stuff if you really need to.
And, as an anecdote, I bought the cheapest, most useless SSD I could find for work machines. It's a Crucial thing that cost a pittance and just large enough to cover our base image. It makes all our machines FLY if they have them in there. Literally a bigger speed impact than double-RAM or five-years-newer processors. We bought them for machines that can't go above 4Gb because of motherboard restrictions (even though 64-bit Windows). They make a bigger difference that the ones we upgraded to 8Gb and beyond.
According to the Crucial Storage Executive software, the SSD in my IT-Office, always-on, only machine I use every day, remote-desktop-into-from home, everything-installed machine, lowest-of-the-low (so people can't say I'm using better kit than they are) reads thus:
Power On Hours Count: 2172 Hours (the disk has been in there a year, the machine is on 24/7, it's only been "powered on" for 3 months, by that number so obviously it powers down out of hours and in idle times).
Reallocated NAND Blocks 0 NAND Blocks
Percentage Lifetime Remaining: 97 Percent
Available Reserved Space: 100 % Spare Blocks Remaining
Total Bytes Written: 9.87 TB
It's 500MB/s read and write. Warrantied for three years (which they class as 80TB Total Bytes
Written for that drive). At current usage, it should give me.... another 8 years. I don't expect it to, it was cheap-as-chips, but it should, on average, overall, across my users. I don't think there's an hard drive that I'd trust for that length of time in active service.
P.S. I did nothing more than image the drive from the previous HDD - didn't change swap settings (and it's only 4Gb RAM), didn't put on over-provisioning, ignored all the software recommendations for caching, etc. so technically I'm really abusing it as an SSD and could get much more life out of it.
Sure, I wouldn't want to use it as a 24/7 CCTV-recording NAS or something, but that's more than adequate. We haven't had one fail. When we do, they're really cheap to replace.
But what I'd really like is something much, much, much larger even if that meant it came in 3.5" format. Cracking one open (they aren't hermetically sealed like HDD's) reveals that it's a little aluminium tin can with a tiny half-populated circuit board taking up about 1/3rd of a 2.5" drive on it. With no further effort, they could easily multiply capacity by six without having to even use different storage chips (maybe a different controller chip). My Samsung 850 EVO at home is the same. That's 1Tb but you could easily make it 4-6Tb in the same container. In a 3.5" drive? You could easily have a RAM-stick-like arrangement and put many dozens of Terabyte chips vertically on an horizontal controller board.
But it's the one area we don't seem to see SSDs growing in - actual capacity, or cost-per-capacity. Which, given that even commodity Windows PCs now come with SSD options, sometimes even by default, I can't fathom.
Stop faffing around and start making larger versions of what you have. I couldn't care less if it was even slightly slower than the current SSDs so long as it was lots faster than an HDD. And was affordable in the multi-Tbyte range.
That's my point...
Increasing speed is pointless.
Increasing durability is pointless.
We need them to do no more than produce a standard, small module, in massive quantities, and bring the price down. They can sell that module singly in a small drive, or thousands together in a humungous one. But until the cost per module comes down, they aren't practical and wasting money on FASTER controllers/chips isn't helping.
I'd be more than happy with whatever chip is in that several-year-old SSD, multiplied up to fill the box, and sold at a decent price. A 1TB Samsung cost me £300 several years ago. By the same token, a 15Tb one should cost £4500 max. In actual fact, it costs £7000 ($10k).
If they can't get the modularity, mass production, and scale correct, we are never going to get affordable SSDs. And I'd be happy for them to abandon HDD production to do so.
@tony72
With regards to Samsungs reliability, well we sell Samsung, Crucial, Hynix, and SanDisk SSDs and the return rate on just Samsung drives is higher than the other three put together. Performance wise yeah, the Samsungs do beat the others, but for most people and oganisations reliability is the most important aspect of storage.
We also build small all flash arrays, and because we mostly build for reliability over performance we’ve stopped using Samsung drives for this unless the customer specifically requests them for performance reasons. And even then we’re upfront about the situation.
Anon - because I have to on this
With regards to Samsungs reliability, well we sell Samsung, Crucial, Hynix, and SanDisk SSDs and the return rate on just Samsung drives is higher than the other three put together.
Interesting. I moved to Samsung's exclusively after having a bunch of SanDisk's fail within their first year. Have not had a Samsung fail yet. Have not tried Crucial or Intel. SanDisk replaced them, and we have nightly backups so data loss was minimal, but the experience was disconcerting.
They could have found their endurance specs were conservative after enough long term real world data. Or more likely determined that so few people actually wrote enough to their device that bumping the TBW warranty would cost them little but provide a marketing advantage over the competition.
I think most people seriously overestimate the amount of writing they do, and worry needlessly about wearing out their SSDs. I invite everyone to check their SMART data and see what the "media wearout indicator" shows. That's what tells you how close you are (lowering from 100 down to 0) to hitting the rated TBW of your drive. At the rate I'm wearing out my SSDs they may live longer than I do :)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
