Samsung 870 EVO - Beware, certain batches prone to failure!

[Blaeza]

So I previously enquired about my 870 EVO's. I also must be having the dumb...

 

[ahmadmob]

Hello guys.

So reading this topic and some topics on Reddit it looks like no one has yet to report any issues with drives manufactured Nov2021 or after. I won't take the date thing as a solid indicator but that's the only thing we have to differentiate between these drives to know if they are good or not.

Anyway, I ordered a 4TB 870 EVO from Amazon a few days ago and will get it in a few days, I wish I knew about these issues beforehand. Wouldn't be an easy process to cancel it now as it entered the shipping process. I just hope I get one that's made anywhere in 2022.

 

[chrcoluk]

Hello guys.

So reading this topic and some topics on Reddit it looks like no one has yet to report any issues with drives manufactured Nov2021 or after. I won't take the date thing as a solid indicator but that's the only thing we have to differentiate between these drives to know if they are good or not.

Anyway, I ordered a 4TB 870 EVO from Amazon a few days ago and will get it in a few days, I wish I knew about these issues beforehand. Wouldn't be an easy process to cancel it now as it entered the shipping process. I just hope I get one that's made anywhere in 2022.

I think we need more time, its kind of like how it was originally thought to be no later than March, it could simply be that those drives are still too new to trigger the problems.

Sadly with Samsung been silent we all left guessing, they think they increasing confidence in their product by staying silent, but if they released a statement like this, I would have no issue buying a new 870 EVO "We have discovered a defect in the 870 EVO SKU, and wish to apologise to those customers affected, this is now fixed in batches produced from DD/MM/YY so you can rest assured on buying a 870 EVO".

 

[ahmadmob]

I think we need more time, its kind of like how it was originally thought to be no later than March, it could simply be that those drives are still too new to trigger the problems.

Sadly with Samsung been silent we all left guessing, they think they increasing confidence in their product by staying silent, but if they released a statement like this, I would have no issue buying a new 870 EVO "We have discovered a defect in the 870 EVO SKU, and wish to apologise to those customers affected, this is now fixed in batches produced from DD/MM/YY so you can rest assured on buying a 870 EVO".

True, Samsung should definitely issue a statement like this and it would be much better for them than to stay silent! What you said makes a lot of sense about needing more time to be sure about the new drives, however I asked a guy on reddit today about his 8 months replacement drive and he answered that it's working good so far after 8 months of use (didn't ask him about the date of the replacement drive though). So fingers crossed about the new drive batches. Here's the thread where I asked him: https://www.reddit.com/r/unRAID/comments/r4rcqg/_/hmrw47g
Edit: I was able to cancel my 870 EVO 4TB order, as the more I read about these 870 EVO drives the more I find horror stories about them.

Do you guys recommend the Crucial MX500 4TB? I saw some people here (or can't remember exactly where) saying that it had issues with some batches regarding firmware or build quality, is this still true or was that only with the lower capacity early revisions?
I was also looking at the WD Blue SSD 4TB, which one would you recommend out of these two drives?

 

[exxos]

I just found this thread as my 870 QVO 1TB died out of the blue yesterday. Dated 2021.07.

I've got 2 of them. My main C: drive and the now failed backup drive. Though its totally died. Even the bios no longer sees it. I've done all the usual swap cables and tried on my laptop. No luck. The other 870 still seems fine but now thinking I don't trust it anymore. Its not like the back up drive had much use even.

 

[CiTay]

870 QVO problems are unrelated to what we see with the 870 EVO. And sudden death is not consistent with the symptoms we have seen here. So i would say, that is very unfortunate, but it doesn't seem to have anything to do with these potential problems of the early 870 EVO batches.

 

[Mussels]

True, Samsung should definitely issue a statement like this and it would be much better for them than to stay silent! What you said makes a lot of sense about needing more time to be sure about the new drives, however I asked a guy on reddit today about his 8 months replacement drive and he answered that it's working good so far after 8 months of use (didn't ask him about the date of the replacement drive though). So fingers crossed about the new drive batches. Here's the thread where I asked him: https://www.reddit.com/r/unRAID/comments/r4rcqg/_/hmrw47g
Edit: I was able to cancel my 870 EVO 4TB order, as the more I read about these 870 EVO drives the more I find horror stories about them.

Do you guys recommend the Crucial MX500 4TB? I saw some people here (or can't remember exactly where) saying that it had issues with some batches regarding firmware or build quality, is this still true or was that only with the lower capacity early revisions?
I was also looking at the WD Blue SSD 4TB, which one would you recommend out of these two drives?

The MX500's have pretty low TBW values, like the QVO drives.
WD greens are atrocious.

People bag out the QVO's for the lower writes and lifespan, so keep that in mind here

QVO:

MX500 - similar life to the QVO
700 vs 720, close enough IMO.

WD blues makes the MX500 look good
720 to 500? Nah.

WD actually hide the green drives TBW and don't list it, but reviews and news websites have it before they hid the values
So you know, these make even the garbage blues look good.

TL;DR: These drives make QVO's look good.

 

[80251]

My old Samsung 850 Pro SSD's are only rated to 150 TBW although after years of service the most written to SSD only has 5.6 TBW.

After someone here on TPU wrote that controller failures account for most SSD failures I have to wonder how relevant TBW figures are.

 

[nomdeplume]

My old Samsung 850 Pro SSD's are only rated to 150 TBW although after years of service the most written to SSD only has 5.6 TBW.

After someone here on TPU wrote that controller failures account for most SSD failures I have to wonder how relevant TBW figures are.

Due to ongoing conflicts I will make mention alone of a constant write test performed on a few hundred SSD make/model to explore TBW. In the case of a 500GB Samsung 850 Pro it overperformed by better than an order of magnitude. The next gen of Samsung NVMe were better yet in the TBW category. 840 SSD showed only slightly less endurance than the 850. Obviously drive memory size matters in relation to performance capability and this was a solitary data point. 2TB does appear to have multiple advantages over 256GB.

Over the full course of this test it became apparent that a known high quality controller, good programming/firmware, and build quality could be undone by a bad pairing of memory to controller. Net effect being sufficient memory and memory quality were often still abundant at point of failure. A controller that intially appeared to be performing well increasingly got its signals crossed and left a very short window of reporting to notice it in a monitoring program. Taking some memory etc down with it that did show up as a reduction of drive life. Sometimes cataclysmically, but more often very subtly with only a few percentage points of life dropping within a short timeframe.

NVMe do appear to have better reporting capability. Better insight into corrections taken and actual signs they are struggling. The only SSD that reported stairstepping failures down to red level warnings on Crystal Disk actually had lower quality memory that was dying.

 

[ajsmsg78]

I had installed one of these in a clients machine in May and four months later (a few days ago) they were getting all sorts of blue screens and random windows crashes. The drive was slowly starting to die. The date on it was 2021.11 so these issues go further than just the batches from early to mid 2021.

 

[80251]

I had been considering buying a 4 TiB 870 EVO, but went with the much slower 4 TiB 870 QVO. The 870 EVO seems to be the most failure prone SSD Samsung has ever manufactured and I'm surprised Samsung has kept completely silent about it.

 

[ReachJuggernog]

Thought I'd make an account to add to the growing list of failing drives here, I bought a Samsung 870 EVO 4TB in October 2021, and it's failing bad.
I used it to store all my games, so nothing important was stored on it, but since I don't play much of them these days I didn't notice how bad the failures were until I stumbled across Event Viewer basically spamming the logs with "this drive has a bad block".

Tried scanning it with Samsung Magician, a process that should have taken around 3-4 hours with a 4TB drive, but it was still scanning 12 hours later and had only completed around 20% of the drive because of all the failures.

Attached is some drive failure p_rn for all of you, using Samsung Magician's Diagnostic Scan
(talking to support now to get some kind of compensation, preferably a refund)

 

[CiTay]

When was the drive made? It's written on the sticker.

 

[ReachJuggernog]

When was the drive made? It's written on the sticker.

I'm in the process of making sure I've gotten everything off that's remotely "important" off that drive (some game configs etc), so I haven't taken the drive out yet.
Based on my serial number and this site: https://technastic.com/check-manufacturing-date-samsung-devices/
I'm putting my date of manufacture at July 2021. My serial number being S6BCNJ0R700----
R7 being the key 8 and 9th digit in that which would put my drive on that date.
Applying the same method to another person's serial number earlier in this thread (S62BNJ0NC62) and them reporting their date of manufacture in December 2020, "NC" (according to that site) would confirm that the drive was manufactured in December 2020.

Some more evidence of it failing, "Uncorrectable Error Count" and "ECC Error Rate" are critical according to Samsung Magician.

Model Name	Samsung SSD 870 EVO 4TB
Serial Number	S6BCNJ0R700----
Drive Type	SATA
Result	ID	Description	Threshold	Current Value	Worst Value	Raw Data
	5​	Reallocated Sector Count	10​	43​	43​	2706​	OK
	9​	Power-on Hours	0​	98​	98​	5522​	OK
	12​	Power-on Count	0​	99​	99​	360​	OK
	177​	Wear Leveling Count	0​	99​	99​	7​	OK
	179​	Used Reserved Block Count (total)	10​	43​	43​	2706​	OK
	181​	Program Fail Count (total)	10​	100​	100​	0​	OK
	182​	Erase Fail Count (total)	10​	100​	100​	0​	OK
	183​	Runtime Bad Count (total)	10​	43​	43​	2706​	OK
	187​	Uncorrectable Error Count	0​	54​	54​	459507​	CRITICAL
	190​	Airflow Temperature	0​	79​	45​	21​	OK
	195​	ECC Error Rate	0​	199​	199​	459507​	CRITICAL
	199​	CRC Error Count	0​	100​	100​	0​	OK
	235​	POR Recovery Count	0​	99​	99​	2​	OK
	241​	Total LBAs Written	0​	99​	99​	32146165156​	OK

 

[CiTay]

Yep, as i mentioned in another reply before, that is an epic amount of ECC errors now, but it's to be expected. The drive is retrying over and over again to read out the bad sectors - due to your full drive scan - and stalling for seconds on each sector it can't read. The SSD basically tried half a million times to read the almost 3000 bad sectors, unsuccessfully.

That's not one of the earliest manufacturing dates, but still heavily affected...

 

[ReachJuggernog]

Yep, as i mentioned in another reply before, that is an epic amount of ECC errors now, but it's to be expected. The drive is retrying over and over again to read out the bad sectors - due to your full drive scan - and stalling for seconds on each sector it can't read. The SSD basically tried half a million times to read the almost 3000 bad sectors, unsuccessfully.

That's not one of the earliest manufacturing dates, but still heavily affected...

Crazy that Samsung has seemingly offered no warning about this? I posted this in a Discord server with a few friends and one of them picked up the same drive not long after mine, so he's gunna check he's also not having the same issue.

 

[CiTay]

Funny thing though: I did not RMA my drive (the failed one from the first post) yet, to see how it develops, and so far, it hasn't developed any new bad sectors. Maybe the newest firmware SVT02B6Q prevents that somehow. I filled the drive to the brim a couple times (0 bytes free space), but it seems that once you have deleted the files which contained the bad sectors, those bad sectors are indeed mapped to intact ones and are not written to again. This is how i never got another bad file to this date, and the bad sector count remained the same at 329 for that drive.

I will RMA it eventually, but i thought this is an interesting observation. I purposely use this drive more heavily than before (of course only for files i can easily download again), but ever since i deleted the bad files, the drive is behaving. Bad sectors seem reallocated to good ones and no new bad ones developing.

 

[chrcoluk]

I've got speculations in my head as to whats possibly going on, but ultimately the question would be why ditch the 860 EVO skew for a new skew that has no performance/marketing advantage over the older one? Answer possibly to increase margins due to cheaper components. Is the nand the same binning quality?

 

[E.S]

and so far, it hasn't developed any new bad sectors. Maybe the newest firmware SVT02B6Q prevents that somehow.

SVT02B6Q may not have solved the problem.

I updated to SVT02B6Q after having the same problem as you guys.
After that I had no problems for 9 months.

But today, the value of the ID "FC" in CDI increased. This "FC" was added in SVT02B6Q.

This is described as "Vendor Specific", but after looking it up, it appears to be S.M.A.R.T 252 "Newly Added Bad Flash Block".

If this value continues to increase, it may indicate an increase in bad blocks.
If so, maybe the problem is not resolved?

 

[CiTay]

I don't know how much credence i give to a value whose meaning is not even 100% certain. If i actually see Bad Blocks increase, that's a different story.

 

[E.S]

S.M.A.R.T 252 is described on wikipedia as follows:
"The Newly Added Bad Flash Block attribute indicates the total number of bad flash blocks the drive detected since it was first initialized in manufacturing."

I don't know if this explanation is correct, but I think it's probably an item about bad blocks. I think you should be careful with this value.

 

[CiTay]

If we look at it logically, every SSD has some bad blocks, because no NAND Flash is perfect. For example on a brand new Sandisk 2TB it will always be well over 1000 bad blocks from factory. But those are counted seperate from the "Grown Bad Blocks", and as long as those bad blocks have been mapped with good ones and the user data will never be affected, that's not a problem. This is the main criteria here: The data on the SSD should never become corrupted or unable to be read. And this seems to happen when the values increase which i mentioned in the initial post.

I will keep an eye on the mystery "Vendor specific" data (for which my affected 870 EVO shows 1349, one unaffected one shows 0 and the other unaffected one shows 1210), and i will certainly accept it as an early warning sign when this should translate into real bad blocks later, but so far it's just anecdotal evidence. So there's no need to panic IMO, just calmy watch how the SMART data develops further.

 

[E.S]

I will keep an eye on the mystery "Vendor specific" data (for which my affected 870 EVO shows 1349, one unaffected one shows 0 and the other unaffected one shows 1210), and i will certainly accept it as an early warning sign when this should translate into real bad blocks later, but so far it's just anecdotal evidence. So there's no need to panic IMO, just calmy watch how the SMART data develops further.

I understand.
Our problem maybe happens thousands of hours after writing the data. This increase in S.M.A.R.T 252 occurred 9 months after recovery.
Maybe isolated the problematic block as "Newly Added Bad Flash Block".
What I fear is that this value may continue to increase in the future.

 

[E.S]

This forum cares about the date of manufacture, but I don't think it matters. I searched the internet for many CDI images of the 870 evo, If you use the old farm for 2000 to 5000 hours, there seems to be a high probability that the problem will occur. I don't understand why samsung doesn't product recall or explain.

 

[Captain Danger]

Since we have a speculation about the "Vendor Specific" value, mine has increased from 80 to 100, so just wanted to share. Everything else still looks just fine.