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

[Icon Charlie]

Not sure what you can construe from "construction" when everything is down to NAND chip and controller these days... and those all look the same construction wise.

True. The difference was enough to bother me.

 

[chrcoluk]

So I have this in my laptop now.

There is a firmware update available, I googled it because as usual no changelog, and apparently this is only offered to some variants of the 870 evo, the guy who posted it says there is a newer controller chip now that isnt phoenix (are samsung now in the bait and switch game?). Anyone else heard of this?

On my SSD for reference all smart self tests passed including the extended test. Hopefully this wont have crazy erase cycles like the mx500 had.

If people want pics of label for sku, I can post as I took before putting in laptop.

 

[CiTay]

I have three 870 EVOs from different production dates, all have the same firmware, and the oldest one is affected by the problem from this thread.

Samsung will never use someone else's controller, they have a focus on vertical integration with their own controllers and their own NAND, this keeps costs down and gives them a competetive advantage. So unlike ADATA, Kingston etc., they won't just swap parts to cheaper ones without notice.

About passing tests, this is for when you want to do the RMA. First, you keep looking in CrystalDiskInfo for the signs i mentioned in the first post. This is the only indication you need to see if the problem appears or not. And depending on the production date, there is a certain higher (early dates) or lower (later dates) probability for it. So the only thing that would be interesting from the label is the production date.

 

[OneMoar]

you put a QLC drive in a NAS and then wonder why it failed?
sigh I can't I just can't with you people

 

[CiTay]

you put a QLC drive in a NAS and then wonder why it failed?
sigh I can't I just can't with you people

870 EVO is TLC NAND. You're probably thinking of 870 QVO. Plus, who did you see mentioning a NAS?

My three EVOs are normal internal SSDs.

(Yes, i enabled XMP later, this was right after a BIOS update.)

 

[chrcoluk]

I have three 870 EVOs from different production dates, all have the same firmware, and the oldest one is affected by the problem from this thread.

Samsung will never use someone else's controller, they have a focus on vertical integration with their own controllers and their own NAND, this keeps costs down and gives them a competetive advantage. So unlike ADATA, Kingston etc., they won't just swap parts to cheaper ones without notice.

About passing tests, this is for when you want to do the RMA. First, you keep looking in CrystalDiskInfo for the signs i mentioned in the first post. This is the only indication you need to see if the problem appears or not. And depending on the production date, there is a certain higher (early dates) or lower (later dates) probability for it. So the only thing that would be interesting from the label is the production date.

Is it either of these firmware's 'SVT01B6Q' or 'SVT02B6Q'? If its SVT01B6Q can you check if magician offers an upgrade?

 

[CiTay]

All on SVT02B6Q, no updates available.

 

[chrcoluk]

Found a tweet which references this thread and the reddit link, mentions the new firmware.

https://twitter.com/i/web/status/1502185863052152833

Attaching my label, its not quite as old as your drives, but not that much newer either. Like most SSD's I purchase its much older than the date I actually purchased it.

 

[CiTay]

Well yeah, i updated them all to SVT02B6Q at one point, but i'm not certain what the previous firmware version was, and once you get these errors, SVT02B6Q does nothing to fix them.
So if anything, maybe SVT02B6Q is meant to prevent this problem somehow, but i don't know if that is possible for the early production batches.
About "silently" releasing a firmware update, that's their normal modus operandi. Just today i happened to find a new firmware for my 980 PRO because i installed the new Magician 7.1.0.

2021-06 should be good. You don't really have to cover your serial number like that, you can remove the last four digits if you want, but the serial is a good way to pinpoint the more exact production date.
Also, the serial number could still be decoded from the barcode, but i don't know what even a black hat could do with any serial number. I'm sure i posted screenshots with a drive serial before. No big deal

 

[chrcoluk]

Sept 2021 manufactured also bad drives. I got mine from Amazon but obviously well past their no frills return policy now. Going to do a badblocks scan on it and see what happens. The 860 EVO's it seems were better.

If the new firmware prevents it from kicking off (I also have same suspicion) I wonder if it does it the horrible way of forcing data refresh which of course accelerates erase cycles, a hacky way of fixing dud hardware. I wanted to see first how long a erase cycle takes on the shipped firmware, then flash and see if its accelerated, of course one can not downgrade flash, so flashing is a full on commitment in that new firmware version.

What I also found interesting is the reports some people got refunds without even been offered a replacement, is that samsung having no faith in a replacement 870 EVO working for their workload or perhaps just lack of stock? I have never heard of a company offering refunds like that before on manufacturer RMA's.

https://www.reddit.com/r/techsupport/comments/og701s/_/hz002k0
Hmm Interesting. https://blog.yuo.be/2021/12/23/the-ten-month-old-ssd-that-ate-my-data/

Near the bottom, a new SMART attribute gets added on the new firmware, says vendor specific.

(If you’re particularly observant, you’d have noticed there’s a new SMART attribute at the bottom in the previous screenshot. It appeared after applying the recently released SVT02B6Q firmware update for the drive. I don’t know what the attribute is and neither does the Magician, it seems. I also have no idea what the firmware update does as Samsung haven’t provided any release notes. The update certainly didn’t help with my problems, but it was probably too late for it to anyway.)

So March 2021, January 2021 (made in korea), Sept 2021 confirmed bad manufacturing dates,

-- Ran the full scan in magician which is pretty much the same as a read only badblocks scan, and no issues, will carry on using it on shipped firmware and just keep an eye on it.

 

[Tetras]

you put a QLC drive in a NAS and then wonder why it failed?
sigh I can't I just can't with you people

They're not QLC (that's the QVO) and it has nothing to do with NAS usage, mine is a desktop drive with barely 3 TB written. Drives manufactured between the above dates seem to be consistently failing.

Sept 2021 manufactured also bad drives. I got mine from Amazon but obviously well past their no frills return policy now. Going to do a badblocks scan on it and see what happens. The 860 EVO's it seems were better.

Hmm, that's worrying if September drives are failing because it suggests 1: it isn't isolated to early batches, 2: Samsung didn't identify and fix the issue.

About the controller, I think that was about the 970 Evo? If I recall correctly, they swapped the phoenix controller of the 970 Evo to Elpis from the 980 Pro.

 

[Captain Danger]

So I was looking for changelog for firmware version "SVT02B6Q" and it lead me to the reddit post/this thread.

I have 870 EVO 1 TB.
Made in Korea.
Production Date : 2021-01-15
Purchased On : 2022-01-19
Stock/Shipped Firmware Version : SVT01B6Q

I have been using it for over 2.5 months now (on shipped firmware, updated the firmware only yesterday) and I think everything looks fine so far - only health dropped by 1%. Can someone please confirm from the attached CrystalDiskInfo screenshot?

I ran a full Diagnostic Scan through Samsung Magician just to be sure as well - screenshot attached. It said it will take 40 minutes for 1 TB SSD but it took 2 hours, I am not sure if that is of any concern?

I am currently running it in Full Performance mode under Samsung Magician.

Attaching the CrystalDiskMark screenshot as well, are those Read/Write values normal when it is set to Full Performance Mode under Samsung Magician?

From the above reply my SSD falls in the bad manufacturing category so I will keep updating here if anything shows up.

 

[Tetras]

So I was looking for changelog for firmware version "SVT02B6Q" and it lead me to the reddit post/this thread.

I have 870 EVO 1 TB.
Made in Korea.
Production Date : 2021-01-15
Purchased On : 2022-01-19
Stock/Shipped Firmware Version : SVT01B6Q

I have been using it for over 2.5 months now (on shipped firmware, updated the firmware only yesterday) and I think everything looks fine so far - only health dropped by 1%. Can someone please confirm from the attached CrystalDiskInfo screenshot?

I ran a full Diagnostic Scan through Samsung Magician just to be sure as well - screenshot attached. It said it will take 40 minutes for 1 TB SSD but it took 2 hours, I am not sure if that is of any concern?

I am currently running it in Full Performance mode under Samsung Magician.

Attaching the CrystalDiskMark screenshot as well, are those Read/Write values normal when it is set to Full Performance Mode under Samsung Magician?

From the above reply my SSD falls in the bad manufacturing category so I will keep updating here if anything shows up.

Seems like you're good for now, you have zero reserved blocks used or unrecoverable errors

 

[CiTay]

Attaching the CrystalDiskMark screenshot as well, are those Read/Write values normal when it is set to Full Performance Mode under Samsung Magician?

1) Your SSD is completely fine so far. "Health" dropping relatively quickly to 99% is normal. For the drop to 98% it will take considerably longer. You probably have to use this SSD for several decades to see it drop to concerning levels. By then, SATA technology will be something that some old people may remember.

2) There are certainly some snake-oil shenanigans going on with full performance mode. It seems to cache things through your RAM. I would use Standard more, or Custom mode with at least the "Full Power Mode" disabled. This aversion against power-saving modes (or the notion that they hinder performance) is completely unjustified. Whenever there is something to do, the SSD will wake up within nano- or milliseconds and stay awake until the transfers are done. It's unnecessary for it then to stay on high alert, because 99% of the time there is simply nothing to do. It's like grandma leaving her car running day and night, for the one time in the week she wants to go grocery shopping, so it saves her the two seconds where she would have to turn the key in the ignition.

 

[ThrashZone]

Hi,
Issue is mainly with 2 & 4tb models not 1tb models.

 

[Tetras]

Hi,
Issue is mainly with 2 & 4tb models not 1tb models.

That's what I thought, but my 0.5tb is included

 

[ThrashZone]

That's what I thought, but my 0.5tb is included

Hi,
I don't have any 870's fortunately but damn they really screwed the pooch on this series.

 

[iddqd]

I just started RMA the second time for my 4TB 870 Evo.
I bought it in February 2021 and noticed problems after a few months. The SSD was produced 2021-01-16. Last summer I contacted Samsung and after some back and forth because i am using linux (no samsung magician) they accepted the RMA. I had to put the ssd in a windows pc and format it to make the necessary tests with their software. The formatting resolved the read errors and Samsung at first declined the RMA, but the SMART data was still there and it showed the bad values so they eventually gave in.

The replacement SSD was produced on 2021-01-15, so one day before my original one. I noticed errors again after after a few months but I had no motivation to transplant the ssd and perform all the tests. I normally back everything on the SSD up to a spinning disk raid array but my backup script didn't work for a few weeks without me noticing and now I lost some data (luckily nothing I cannot get back).

Samsung updated Magician and now I can perform the necessary tests without formatting. As I am writing the full diagnostic scan is ongoing and there are lots of red squares (errors). So there should be no problem with starting the RMA process this time. Oddly enough, with the new version of magician and without formatting I can only perform the tests which were not possible before and vice versa.

Here is a screenhot of the full scan, after more than 4 hours of the estimated 2,5 hours (160 minutes for 4 TB). It is already 14% finished

CrystalDiskInfo, before the full diagnostic scan started:

CrystalDiskInfo now (during the scan and with decimal raw values).

What will happen if the threshold is reached during the diagnostic scan? Before starting the scan it was at 45 an now it is at 35 with only about 15% done.

 

[CiTay]

CrystalDiskInfo now (during the scan and with decimal raw values).

Ouch. Yeah, Reallocated Sectors will increase during a scan, but Uncorrectable/ECC Errors will even increase way more, because the SSD keeps retrying to read the data, each time failing error checking and stalling the drive for a while. With this sort of SMART data, you will have no problem getting the RMA.

What will happen if the threshold is reached during the diagnostic scan? Before starting the scan it was at 45 an now it is at 35 with only about 15% done.

Nothing specifically will happen, CrystalDiskInfo will perhaps mark that item with a red dot and the health status might even go to "Bad", but it could also stay on "Good". Shows you what a bad metric the Health Status is in there. If you don't know how to interpret the raw values (which should always be in DEC, i have no idea why HEX is default), then CrystalDiskInfo is not really the best tool for cases where you are unsure. But it shows everything conveniently at once, so if you know what to look for, it's quite good.

 

[Captain Danger]

The replacement SSD was produced on 2021-01-15, so one day before my original one.

My 1 TB 870 EVO that I posted about above has the exact same date! Is yours made in Korea as well?

 

[iddqd]

My 1 TB 870 EVO that I posted about above has the exact same date! Is yours made in Korea as well?

Yes, both of them are made in Korea. The S/N of the first one, which is manufactured one day after the replacement, is 51 higher than the replacement.

The full scan is still in progress, 42% after 16 hours and lots of errors. The SMART value for reallocated sectors is getting worse as it is scanning, but slower. It was at 45 (2440) before the scan, dropped quickly to 35 (2908) after about 4 hours and is now at 31 (3902) . So it does not appear to be a linear progression.
The value for uncorrectable errors started at 99 (3168) and dropped down to 65 (350894), just like @CiTay described). The raw values of uncorrectable erros and ECC error rate are the same, but the smart value of ECC error rate stayed at 199.

 

[xorbe]

I bought a 4TB 870 EVO in Apr '21 ... guess I'll have to have a look, sigh. Yup, manuf Jan 2021 ... well, shoot. Mine is still looking good at 1,316H and 2,571GB written. Copied my precious data off, just a bunch of non-precious multimedia left. Though the log says one failed LBA after 800H. Will full scan ...

 

[2BitNandBand]

I have 3 870 EVO's here:

• 1TB, JAN 2021, Korea (1 Reallocated Sector)
• 4TB, NOV 2021, Korea (Fine so far)
• 1TB, FEB 2022, Thailand (Fine so far)

The 1TB from Jan 2021 shows 1 reallocated sector, so I assume it's all downhill from here. I pulled it from my system and replaced it with the other 1TB. The new 1TB came with the later revision firmware. My 4TB still came with the older firmware, but I've since updated it. Really hoping I don't have any issues with that one.

 

[ThrashZone]

Hi,
At what point does sammy start sending out 860 rma replacements ?

 

[CiTay]

The 1TB from Jan 2021 shows 1 reallocated sector, so I assume it's all downhill from here.

Just run a full scan in Samsung Magician, it should uncover any new bad sectors. The main problem is not the fact that there are bad sectors, because they can simply be reallocated from the unallocated blocks that are reserved for that very purpose. For example you can see that on Sandisk SSDs, i've had several where the value "A9" Bad Block Count was >1000 from factory (on a 2TB SSD), so every SSD will have bad blocks from factory. Sandisk also have the "05" Reassigned Block Count value, this should stay at 0 ideally. And Samsung just doesn't show the amount of factory bad blocks in the SMART data, so with Samsung SSDs you will only see newly discovered bad sectors in "05".

Anyway, the real problem is when reallocated sectors start increasing while you already use the SSD. Because if the data can't be read out anymore from those sectors, then the entire file is lost, simply because it can't be read in its entirety anymore. And this seems to happen silently. Meaning, unless you try to copy or read out all your files, you will never know that some of them might be partly saved on bad sectors. A full scan should do the same thing, so it should be able to uncover all those lingering bad sectors.

This is the concerning part with this flaw: A top-notch (technology-wise) SSD silently developing bad sectors, when the NAND flash really wasn't punished much yet with outrageous TBW.