Solved Recommendation for a Reliable SSD

Crucial SSDs are not as reported to be good. Two crucial SSDs (BX@150GB & MX@1GB) have failed within the space of two months and they have been purchased and used for 1.5years.

They won't show glaring errors. I only saw 2.45MB checksum error in the last failed SSD. I needed to change the drive position on board to see it. Before then, it would be zpool status unavail or faulted. I changed power cable too and position it; at the time the SSD would be back and active in the pool but only for a day or two before reporting error again.
SmartMon/Smartctl gave no early warning.

Is daily scrubbing not ideal? It was recommended somewhere here though.

Reviews should that Samsung EVO is more expense because of its brand; Seagate Barracuda is on the high side too.
 
I have 3 Samsung SSDs. Well, one is nvme so not sure that counts. Yes they are expensive but from my reading have good reliability. The 2 actual SSDs are an EVO and an EVO Pro. Both have had countless operating systems installed and formatted and reinstalled. They are around 5 years old and according to what I understand from using smartctl, they have a lot of life left. I have to caveat that with the fact I am not a smartctl expert. I would provide data but the drives are in a box in the closet right now…

the nvme drive has Windows 10 on it and is only a few months old so can’t speak to how it is doing. The Samsung magician software is installed on that system so I’ll check next time I boot it up.
 
Crucial SSDs are not as reported to be good. Two crucial SSDs (BX@150GB & MX@1GB) have failed within the space of two months and they have been purchased and used for 1.5years.
I don't know which reports You rely on. I recently did a market scan for SSD, and Crucial were clearly not among the good ones.
I remember writing an article about it, but somehow it seems it did not make it to a posting?
 
Trying to recap: the only thing I am interested in is the lifespan. And that is also the one thing many manufacturers are reluctant to tell - more about that below.

There are generally two different kinds of SSD available: consumer use and datacenter use. Datacenter use is twice the price (or more). These may have better power failure protection, and may have better sustained-write behaviour.
On the consumer side the sustained-write behaviour is crappy: the device will read data at full speed for a second or two, and then it will stall a while for housekeeping.

So, as this is probably all the same crap, lets focus on the lifespan. And lets normalize all values to the number of PE-cycles. There are currently three different quality bands (plus a forth):
  1. 180-350 cycles, QLC
  2. 220-400 cycles, TLC
  3. 500-800 cycles, TLC
And all of them cost nearly the same.
The fourth band is somewhere around 1250 cycles, which is the same as datacenter/mostly-read, i.e. 0.7 DWPD over 5 years. And these cost usually twice as much.
If I remember correctly, both of the Crucial (BX and MX) belong to band 2.

In band 3 we have, among others, the HP S700, Verbatim Vi550, Lexar NS100, Samsung 860EVO, and certain Seagate Barracuda (but not the barracuda Q). In band 4 is e.g. Seagate IronWolf or Samsung 860PRO. (These are NOT recommendations, I didn't test them all.)

How to derive the normalized PE-cycles: for the more professional devices there may be a DWPD value, then you have to figure out the warranty-timespan, and multiply the DWPD with the number of days. For others there may be a TBW value, then divide this by the capacity. But beware: often the TBW value is not given for the actual device, but for the biggest-capacity device in the series!

But often nothing at all is given. Then there is still a way, because manufacturers do obviousely consider their customers stupid, but they themselves are not stupid when it comes to the money. So, look for the warranty specs. The warranty specs usually point to some fine-print exclusions - another document. These exclusions may then mention a limit of TBW, and point to some value given in the product specs.

In the case of Crucial, a few months ago when I checked this, there was a link to Full Specifications, but no such value given. And I went searching around in circles and couldn`t find the mentioned product specs document. Finally with google I did find the document. On the webpage it was practically impossible to find.

Now checking again, they have fixed that. The value is now in the Specifications, and also my local reseller shows it on their product page. (If that were there earlier, I would not have bothered to search the website.)
Anyway, that search adventure left an impression, so I for my part wouldn't buy that brand.
 
For what it’s worth, I’ve got a Crucial SSD (2 TB NVMe) that is working perfectly fine. I’m using it as part of a mirror together with a Samsung SSD, because the rule of thumb says that you should use different vendors for mirrors, in order to lower the risk that they fail at the same time.

Apart from that, most of my SSDs are Samsung, from various points on the price / speed range (EVO, QVO, PRO, both SATA and NVMe). The oldest Samsung SSD I have is an 830 series one (128 GB SATA) that still works fine.

My oldest SSD is an OCZ Vertex-2 (SATA). It also still works without issues.

Usually, the SSD lifetime specified by vendors (in TBW units) is trustworthy. At least that has been my personal experience, and this is confirmed by an extensive test published by the German c’t magazine recently. Some vendors’ SSDs even exceed their TBW values considerably (by an order of magnitude). However, there are some conditions that can reduce the lifetime of an SSD, for example high temperatures.
 
I've had decent luck with Smasung 850/870 Evo/Evo Pro in machines that are general purpose desktops, not heavy database activity, not streaming content.
 
Usually, the SSD lifetime specified by vendors (in TBW units) is trustworthy.
Yes, thats why I recommend looking at the warranty specs, if these values aren't to be found anywhere else. (e.g. for the Adata brand they were published only in some excel sheet that could be found via the warranty specs.) Because these are warranty clauses, they go into the insurance calculations of the manufacturer, so they must be quite solid.
And, just as with warranty periods, on average a device will work quite a while longer.
 
any reputable brand will do, just don't buy the cheapest consumer versions
don't fill them to 100% the lower fill factor the better
don't make to many snapshots of rapidly changing data
 
I bought a Kingston 120GB A400 in May 2019 and I have been using FreeBSD as my desktop xfce4,firefox,libreoffice,etc

juan@fbsd12-i3:~ $ df -H
Filesystem Size Used Avail Capacity Mounted on
/dev/label/ssdrootfs 116G 24G 83G 22% /


And gsmartcontrol reports SSD Life Left at 93% at this rate it will fall to 30% in 18 years. I hope I will have a bigger and better drive or a new PC long before that.

Also I have installed Windows in many AData SSD (mostly 256gb) and haven't had a problem yet. Maybe you just had bad luck with the Crucial drives.
 
I recently switched a workstation to SK Hynix Gold S31 -- I'd read good reviews and it was a few dollars cheaper than Samsung 870EVO. Does anyone have actual data on the durability of the Hynix ssd?

(Hynix is mainly known as a Korean memory manufacturer, for example supplying DRAM memory to Apple).
 
I recently switched a workstation to SK Hynix Gold S31 -- I'd read good reviews and it was a few dollars cheaper than Samsung 870EVO. Does anyone have actual data on the durability of the Hynix ssd?

That looks to be difficult again. Probably one needs to speak korean to find the warranty fine-print. What I find in english looks somehow like cut&paste.
But, on the Amazon sales page there is a statement "best-in-class 600 TBW". And that is the same for all of the models, 250, 500 1000 GB. So yes, this is again those weird statements as I described above.
But then, assuming 1TB is indeed the biggest in the series, this would equate to 600 cycles, same as the Samsung 870EVO.
 
Where is the Url?
remember writing an article about it, but somehow it seems it did not make it to a posting?
In
The fourth band is somewhere around 1250 cycles, which is the same as datacenter/mostly-read, i.e. 0.7 DWPD over 5 years. And these cost usually twice as much.
Intel DC,......
What other names?
 
That looks to be difficult again. Probably one needs to speak korean to find the warranty fine-print. What I find in english looks somehow like cut&paste.
But, on the Amazon sales page there is a statement "best-in-class 600 TBW". And that is the same for all of the models, 250, 500 1000 GB. So yes, this is again those weird statements as I described above.
But then, assuming 1TB is indeed the biggest in the series, this would equate to 600 cycles, same as the Samsung 870EVO.
And SK Hynix SSD not in the US market evident from. eBay. The company makes good ECC RAM, maybe the SSD too.
 
We can rule out temperature but not heavy writing (& reading)- several jails, big data, poudrière, etc.
I'm waiting for a brand to use next so as to mix them.
 
Where is the Url?
lost
Intel DC,......
What other names?
For the 0.7 DWPD class - Seagate IronWulf or Samsung PRO. But I don't know if these are to recommend. I don't need such, they would loose value quicker than I could use them.

This is also to consider: given a consumer class device would probably endure 4 years, and a datacenter class 8 years - then it is not worth the premium, because in 4 years these things will likely be cheaper. (I suppose you run them mirrored anyway, and have backups anyway, so when one dies, it's no issue to replace.)

What I am using: two A400: nice, cheap, easy to obtain, good service, but only 330 cycles officially. HP S700: 570 cycles officially, internal smart data rather points to 1000 cycles.
And the first I bought was a Sandisk 64G , which died within warranty, and I got 120GB replacement, I don't know what exactly. Sandisk seems to have a policy to not really publish PE-cycles, the smart data is just as unintellegible, but it has near 500 cycles accumulated and still counting.
 
If you're buying consumer gear you're also getting that kind of quality because everything is made to a price.
If you're looking for SATA ones I've had good experience with ones based on the Marvell 88SS1074 (these are usually TLC based) from Kingston, Pioneer (Liteon) and Sandisk/Western Digital in various applications ranging from gaming consoles to PCs. Might not the the fastest on paper but they to work well in general. Samsung Evo (I'd avoid QVO like the plague) are supposedly good but they're usually much more expensive than Marvell based ones for no apparent reason other than you're paying for the brand. Phison seems rather dodgy when it comes to SATA units from what I can tell and I'm not all the confident in Silicon Image either.

As for NVME I'd take a look at SSDs using Realtek RTS5762 (not RTS5763), Micron, Toshiba / Kioxia (Exceria Plus or better) and Sandisk controllers depending on pricing and requirements.

As for SK Hynix, they did have a few SSD that got great reviews in US however I expect that they're backordered for quite some time due to the current situation.
 
We can rule out temperature but not heavy writing (& reading)- several jails, big data, poudrière, etc.
I'm waiting for a brand to use next so as to mix them.
Aye, then probably Datacenter class is right for You...
 
In my machines I run mostly smaller Innodisk or Advantech SSD. 16GB in my firewall to 64GB SATA DOM on my storage machines.
They are costly but are very reliable drives. Some are SLC.
My big SSDs are 1.92TB NVMe. Samsung PM983.
My laptops I use SanDisk X300 X400 series 64GB to 128GB.
My mini NFS fileservers are using Micron M500DC and M510DC Datacenter 1TB SATA drives.
 
I'm happy with ADATA SSD's that I bought on Amazon. It's what I have my FreeBSD install on. Generally, I would stick with major brands like Samsung, Kingston, SanDisk, Seagate, Western Digital... I did have a Crucial SSD back in 2005, but back then, I did not have the compatible connectors inside the PC, and I messed things up by buying a converter (SATA to IDE) that was poorly made. But I would not have a problem buying a Crucial SSD... Since 2012, I don't use HDD to install an OS, but I still have a couple 2 TB HDD's for my animes - prices for 2 TB SSD's are not where I want them to be yet, but they are coming down.
 
You didn't post the TBW for the drives, so even if they were relatively new, they could be worn out. The MX should be pretty good. The BX is kind of typical consumer-y.

I would only trust three vendors for SSDs: Samsung (Non-QLC), Crucial (only the highest end product lines), and Intel if I really had to (usually for mSATA devices) The SK Hynix Gold has very positive reviews for being somehow reliable and cheap, but I haven't heard any long term impressions about it yet. All the others are hit and miss sort of nonsense, it seems from reliability/durability reports.
 
You didn't post the TBW for the drives, so even if they were relatively new, they could be worn out.
Concerning the OP, it may indeed be that these drives have been used to the limit.

I recommend to do a periodic logging of smartctl -x (and partitioning data), from /etc/monthly or /etc/weekly[1]. (Here is some sample script, fix it up to your needs.) Write the output into /var/backup with the month or week in the filename, so they rotate annually.
This gives you four features:
  1. In case of a disaster, you have (hopefully) a backup, but usually you don't have the partitioning schemes. With this, you can grab these files with recent partitioning info out of the backup.
  2. When a disk fails, it is probably too late to look at the smart values, because the disk may no longer answer to them. With this, you have a smart output from a recent time when it was still operative, that can be analyzed for irregularities.
  3. You always have a history of smart data over the last year, which makes it easier to detect when some values suddenly start to grow. (Scripts to do this automatically, or draw charts, could be easily written.)
  4. If the disk includes some "lifetime writes" data in the smart output (most of them do), you have some data showing the actual demand of your usecase per month/week, to calculate which product is suitable.
[1] Supposed that the machine runs 24/7.
 
Last edited:
Back
Top