Which FAST M.2 NVMe 2TB SSD brand / model?

[Saracen999]

I always sort of factor that in when buying an SSD, but I also know in reality I doubt any of us here can really worry a modern SSD, unless it is one of the really bad ones. ...

In a desktop or laptop as a stadard drive, I'd agree. Not so much if it's a server/NAS cache, or a 'tiered' drive in that kind of environment. In a DT/LT use case, I'm more interested in performance than the ultimate in longevity, not least because it's the data more than the drive that's valuable, and I've typically got backups up the wazoo. Different criteria would (for me, certainly) apply to a cache drive. My NAS supports SSD cache (or direct storage, and tiering) but the current one tops out at about 1000M/s anyway. A "fast" SSD would be a bit pointless.

[Jonatron]

Quite a lot of comments on the subject here: https://news.ycombinator.com/item?id=34648149

[DanceswithUnix]

Not so much if it's a server/NAS cache, or a 'tiered' drive in that kind of environment.

Even then, it needs to be a really big server. A university home directory NAS, or something storing machine to machine comms.

We have a VM server at work which felt a bit lumpy in use. I added a WD Black NVMe SSD to it just as 250GB of swap. Made a huge difference to the usability of the server, but the actual amount of data written to it? 548GB spread over 3 years. Nothing, and that's basically running the company from manufacturing databases and fpga/software build services down to the intranet wiki. The main data is on spinning rust, and those old Toshiba drives don't collect such usage stats unfortunately (though they do say 41000 powered on hours, so they could), but we can't be writing that much as the machine would again be unresponsive writing to rotating media.

[AGTDenton]

Samsungs German division have come out and stated they're looking into the 990 problems.

Touch wood my 970 & 980 work for years to come, even with a 5 year warranty I'd prefer to not use it. Over the 4 months I've had them there's not been a firmware update.

https://www.techradar.com/news/samsu...apacity-issues

https://blocksandfiles.com/2023/02/0...health-issues/

[Saracen999]

Even then, it needs to be a really big server. A university home directory NAS, or something storing machine to machine comms.

We have a VM server at work which felt a bit lumpy in use. I added a WD Black NVMe SSD to it just as 250GB of swap. Made a huge difference to the usability of the server, but the actual amount of data written to it? 548GB spread over 3 years. Nothing, and that's basically running the company from manufacturing databases and fpga/software build services down to the intranet wiki. The main data is on spinning rust, and those old Toshiba drives don't collect such usage stats unfortunately (though they do say 41000 powered on hours, so they could), but we can't be writing that much as the machine would again be unresponsive writing to rotating media.

I was referring to the likelihood of needing higher endurance drives by "us here" though, which I took (rightly or wrongly) to mean mainly personal uses, i.e. home server or NAS or, at most, small business, not things like university uses. Though yeah, some of us here are in so-called 'big iron' environments (and that phrase dates me).

For instance, I've been running 'servers' for many years at home, as someone working from home (and most were probably well over-spec'd for home users) and, more recently, a 'prosumer NAS' meaning a bit pricey for most home users but even so, only 4 bay and Celeron based. In that environment, even when I'm chucking large volumes of video data back and forth, I doubt I'd seriously stress SSDs as a cache, to the point of needing specifically "higher endurance drives". There's certainly no point in high-performance ones in the NAS 'cos even if the SSD is fast, the extent to which my NAS can use it sure isn't.

Actually though, right now, I'm still playing and they're for storge not cache, and even if they were, I doubt I'd be caching the video stuff. The benefit to me would be on things like document read/write/update and I sure don't do enough of that to stress pretty much anything in the NAS, let alone SSDs.

Directly inside a PC, though .... that's where the raw grunt of a fast SSD will help me. Others? YMMV.

[DanceswithUnix]

I was referring to the likelihood of needing higher endurance drives by "us here" though, which I took (rightly or wrongly) to mean mainly personal uses, i.e. home server or NAS or, at most, small business, not things like university uses.

Sorry, I wasn't clear in what I meant.

I also have a home server, and have had one for decades, the first was a 386-40 back when that was fast which had a load of SCSI drives in it.

To flake out a modern SSD you have to write something like 300TB to it, and the actual figure goes up with the size of the drive so those with larger storage requirements will need to write more to wear out their drive. It's a huge number, and the official number isn't necessarily when it will flake out they can keep going for years after. I don't personally know anyone who has worn out an SSD yet.

[Saracen999]

...

I also have a home server, and have had one for decades, the first was a 386-40 back when that was fast which had a load of SCSI drives in it.

....

Similar. First was an old '386 re-purposed, running Netware 3, then 4. For reasons not worth explaining, it ended up as NT4, the Win 2000 Server. Then, I can't even remember what it was called but it ended up as TrueNas. FreeNAS? The hardware migrated a bit, but initially was a 396-16, IIRC, and ended up as a Celeron 550, but a dual-processor system. Kinda pathetic by todays standards but OK for my needs, back then. Of course, needs evolve and media (let alone video) serves weren't a thing then, Not for me, anyway. The dual Celeron had, yup, SCSI drives. I had two 3-bay SCSI cages (6 drives in total, 5 in a RAID-5 and a hot swap). Another spare cage and 3 more dives in a cupboard, just-in-case. It all ran from Adaptec boards and again, I can't remember all the variants but it went from 1540's up tom IIRC, 2940UW. Again, a spare or two in the cupboard, ready. I really need to clear most of that out, 'cos it's nearly all still sitting here.

Perhaps the most interesting (to us oldies) was a couple of 300MB-ish external Miniscribe (or was it Micropolis? Not sure now) "AV" drives. Back in the VERY early days of writing CDs, whwn the burner and software were about £3, and up, and blank CDR's about £10 EACH, I was feeding a Yamaha (CDR-100??) frm those AV drives, because they were heavily buffered to avoid ruining a disk if a conventional drive decided to do a "thermal recalibration" in the middle of a burn. At £10 for a wrecked silver-coloured plastic mug mat, that wasn't funny. But oh, were thse AV drives beautifully built. Expensive, but it showed.

Anyway, back on point.

I think we're actually pretty much on the same page. I see two factors :-

- the odds of using an SSD enough to "wear it out".
- the "cost" incurred of losing data if that happens.

In my case, while noting that the latter is pretty much impossible to entirely eliminate, whether SSD, HDD or evn optical media, you can certainly cut the risk right down. It then becomes a personal choice of balance between the extra marginal reduction in risk of the next level of precaution, versus cost and time/inconvenience.

I mean, of course, backup.

I have a fairly thorough backup process in place, though as the potential damage from loss decreased (retirement) so did my levels of precaution. That luckily coincided with the volumes of data going WAY up, here. So, no longer holding things like client data, or even accounting data HMRC might require from me, but I have digitised (as you probably remember) my CD collection, and then started working on my DVD collection. The latter amounts to something like 3000 DVDs, so digitising that little lot is going to require, firstly, quite a lot of disk space, and secondly, bucket loads of my time. It's the latter that justifies, in my mind, taking pretty strong precautions over backup. Each episode (typically 40-45 mins) of a TV show requires 600-800MB, and a film can hit 2GB though 1.5-1.6GB is probably more common. Those are M4V compressed but at decent quality. Fortunately, I don't bother with Bluray, which would jack up the necessary storage quite a bit.

So .... a "system" SSD, and that's one place where the fastest read/write I can get helps, is going to be reading/writing each of those quantities several times in the course of going source media to my NAS. Even that "300TB" can get eaten up pretty quickly doing that, hence quite a lot being done on pretty fast (like the Samsung T7 Touch) external, USB drives. They're easier to replace. But internal system drives are going to be handling at least some of it.

I do have some very important data that I first, don't want to lose and second, can want/need read/write to frequently. But, volumes of that are tiny, in comparison. Fr cache drives in a NAS, it'll take me decades to get anywhere near that 300TB read/write level. Or more. That, therefore, is where I see SSD caching (or tiering) of traditional HDDs coming into play. The audio and especially video data could, at a push, be re-ripped from source as I'll still have those but the time to do it?? Oh, boy, no. Uh uh, no bleeping way am I doing all that again. Life's too short. But that is certain;y where I MIGHT burn through an SSDs trustable life far quicker than I like. But provided I got the data through i onto the NAS, and then backed up, the most I'm going to lose if an SSD dies is the disk or two in process at the time. Once it's on the NAS, it's (as much as it can reasonably be) safe.

In short (-ish ), depending on which part of all that I'm doing, :-

- performance can be very important, or not
- the data can be very valuable, but small in quanity .... or not
- very large indeed in volume over a period but oly a relative small number of fairly big files at a time
- add 4k video from camera and/or drone, and that adds another layer of issue.

In all of that, firstly, my usage is probably fairly rare compared to the general public, but by no means that rare for enthusiasts or specific types of use case. But secndly, it gives me a spread of purposes n which sometimes performance is important but risk of SSD burnout minimal, and others where performance is much less important but I could conceivably burn through 300TB relatively quickly ... a lot depending on how I set up my workflow, how I structure data storage on the NAS, and what I cache and don't cache.

It's taken some exercising of "grey cells" to set up what I think works well (for me), minimises risk, maximises performance where it makes a difference, and doesn't bite me in the ass too badly if I missed a gotcha somewhere.

[MrJim]

- the odds of using an SSD enough to "wear it out".
- the "cost" incurred of losing data if that happens.

I've yet to get to the point of wearing out an SSD, but I assume it would simply refuse to be written to...but data already written could continue to be read?

[Mr_Jon]

I've yet to get to the point of wearing out an SSD, but I assume it would simply refuse to be written to...but data already written could continue to be read?

I had a Samsung 840 die recently after nine or so years of use - not sure if it counts as wearing out as there certainly wasn't 100s of tb written to it. A few dozen maybe. I put it down more to age and earlyish tech. It then did what I've seen from hdd's too when they die, just refused to show up anywhere as usable.

[DanceswithUnix]

- the odds of using an SSD enough to "wear it out".

Note that wear on an SSD is measurable and reported. The management tool from the manufacturer should tell you the percentage used so far, so you can predict when you will hit the design life of the drive, at which point it might fail or it might keep trucking for years.

I worry about random failure, but if it wears out then that's my negligence as I probably had a year of warning. Heck, a Linux box may even send me an email to report the SMART status change (I've not had that for an SSD, but have had impending failure warnings for hard drives in the past).

[DanceswithUnix]

Looks like the 990 Pro has another firmware update, leaving status of existing drives looking like they have had excessive wear:

https://arstechnica.com/gadgets/2023...-990-pro-ssds/

[CAT-THE-FIFTH]

Looks like the 990 Pro has another firmware update, leaving status of existing drives looking like they have had excessive wear:

https://arstechnica.com/gadgets/2023...-990-pro-ssds/

Its why I like the Seagate FireCuda series - the FireCuda 520 2TB had a TBW of 3600TB and the FireCuda 530 2TB a TBW of 2550TB. The Samsung 990 PRO 2TB only has a TBW of 1200TB. So at least in terms of TBW the Seagate IronCuda series seem pretty well provisioned in this regard at least.

[Output]

Looks like the 990 Pro has another firmware update, leaving status of existing drives looking like they have had excessive wear:

https://arstechnica.com/gadgets/2023...-990-pro-ssds/

From what was said, it seems that's because that while the new firmware fixes the issues from that point on, it can't restore the reported condition to what it should have been if there hadn't been the firmware issue, because any excessive wear damage the earlier firmware caused is still real.

EDIT: Although the title of that article pretty much says that in a much simpler single sentence anyway. Oops.

Those affected by the issue won't be happy it happened in the first place of course, but with a fix now out hopefully they'll be covered by Samsung's warranty to get a replacement or something.