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

[Jon555]

The disk doesnt show on Crystal Disk. windows disk management picture is attached

If you right-mouse-click the left column what does it say for the General, Volume and Driver Tabs?

Also what CiTay said... (UGreen make a nice SATA-USB3 Caddy for example...)

It's possible the Boot Sector for the drive has been nuked, and that can often be repaired with TestDisk, but it's a little techy and I'd try other things first.

 

[PowerDuck]

I just bought a used 870 Evo 500GB with original packaging from marketplace locally (quite a good price, had it been issue free), before I realized it is a defective model. Should have done more research before buying. But to my defense, it's not widely reported unlike the 980 Pro.

Came erased. Hadn't put my own files on it yet. 56 TBW, magician is already showing red circles around Uncorrected Error Count and ECC Error Rate. Reallocated error count is also non-0. Manufacturing date is Dec 2020, serial starts with S62ANJ. Firmware is SVT01B6Q (latest). Pretty certain it is affected by the problem.

Freshly downloaded Magician doesn't let me run diagnostic scans. Drive directly connected to SATA on the 3rd gen Intel motherboard. In ACHI mode. Genuine according to Magician. I can't think of any other reason why diagnostic scan is disabled. Maybe Samsung doesn't want to damage the drive further?

The seller didn't say a blip about a a failing drive. I don't think he's hiding anything. I even got him to send me a screenshot of CrystalDiskInfo before the sale (I take the blame for focusing on the "health percent" and not thinking about the other metrics)

 

[Chrispy_]

I just bought a used 870 Evo 500GB with original packaging from marketplace locally (quite a good price, had it been issue free), before I realized it is a defective model. Should have done more research before buying. But to my defense, it's not widely reported unlike the 980 Pro.

Came erased. Hadn't put my own files on it yet. 56 TBW, magician is already showing red circles around Uncorrected Error Count and ECC Error Rate. Reallocated error count is also non-0. Manufacturing date is Dec 2020, serial starts with S62ANJ. Firmware is SVT01B6Q (latest). Pretty certain it is affected by the problem.

Freshly downloaded Magician doesn't let me run diagnostic scans. Drive directly connected to SATA on the 3rd gen Intel motherboard. In ACHI mode. Genuine according to Magician. I can't think of any other reason why diagnostic scan is disabled. Maybe Samsung doesn't want to damage the drive further?

The seller didn't say a blip about a a failing drive. I don't think he's hiding anything. I even got him to send me a screenshot of CrystalDiskInfo before the sale (I take the blame for focusing on the "health percent" and not thinking about the other metrics)

Try a different SATA port on your board. All the boards for 2nd and 3rd gen Intel Core-i series were defective at launch with an uncorrectable hardware fault baked into the chipset silicon by intel. "B3 stepping" boards were something I remember hunting for in that era since boards marked B3 had the fixed chipset that didn't haemorrhage SATA ports over time.

Original Cougar Point Intel boards suffered from degrading SATA ports that got worse and worse with more and more errors/corruption until the entire port eventually died - but ports died one at a time so others should still work fine.

Intel P67 - Wikipedia

en.wikipedia.org

and

The Source of Intel's Cougar Point SATA Bug

www.anandtech.com

It's also possible that your motherboard is fine and the seller just fobbed you off with a dud and is now feigning innocence.

 

[RJARRRPCGP]

Freshly downloaded Magician doesn't let me run diagnostic scans.

The diagnostic screen shot, is a red flag for me! I only heard of such limitation, if it's a 970 Pro. You should never get that error with an Evo.
I think it's because the 970 Pros have baked in error correction that the Evos don't have.

Except for the later Pros, where the 990 Pro has a diagnostic scan supported by Samsung.
Wouldn't be surprised if the same for 980 Pro as well.

The 970 Pro was a special breed, where people seemed disappointed when the 980 Pro was launched.

 

[PowerDuck]

I swapped the SATA port from SATA 1 to SATA 5, still the same issue.

But I'm pretty sure it is not a counterfeit. I have the packaging with the correct serial. I also tested the drive with a fake flash tester. It has all its capacities and the read/write speed is up to spec. If it were a counterfeit, they usually have fake capacity or use very low grade flash in order to make money. Another good thing is that after the full capacity write test, the reallocated sector count or ECC was not increased. The read test was also 100% successful.

I'm going to try RMA the drive to Samsung. I noticed the following language in the warranty booklet: "if no proof of the date of purchase is provided, the warranty period will be considered to have started three months after the date that the product was manufactured". Since the manufacturing date is Dec 2020, I have until Feb 2026 if I calculated correctly.

 

[chrcoluk]

Well we will find out from the RMA if its a legit drive.

 

[Jon555]

I swapped the SATA port from SATA 1 to SATA 5, still the same issue.

But I'm pretty sure it is not a counterfeit. I have the packaging with the correct serial. I also tested the drive with a fake flash tester. It has all its capacities and the read/write speed is up to spec. If it were a counterfeit, they usually have fake capacity or use very low grade flash in order to make money. Another good thing is that after the full capacity write test, the reallocated sector count or ECC was not increased. The read test was also 100% successful.

I'm going to try RMA the drive to Samsung. I noticed the following language in the warranty booklet: "if no proof of the date of purchase is provided, the warranty period will be considered to have started three months after the date that the product was manufactured". Since the manufacturing date is Dec 2020, I have until Feb 2026 if I calculated correctly.

Good luck - I hope it goes well!

 

[amiens80]

since i have issues on my 870, i take extreme mesure :
once a file is bad, i don't delete it, i just rename it as "bad_sector##"

Ok i lose space, but at least,i will not put new file of the bad cell !

 

[SPDIF]

since i have issues on my 870, i take extreme mesure :
once a file is bad, i don't delete it, i just rename it as "bad_sector##"

Ok i lose space, but at least,i will not put new file of the bad cell !

That wont work, a SSD is not the same as a HDD. An SSD swaps the files around the whole space to use each cells evenly...

Furthermore, bad cells are marked as bad and replaced by spare cells. Now your files only take places doing nothing, and the SSD has to swap them around the SSD the whole time.

 

[tpousr]

I just bought a used 870 Evo 500GB with original packaging from marketplace locally (quite a good price, had it been issue free), before I realized it is a defective model. Should have done more research before buying.

The 870 Evo is not a defective model. It's just that some batches shipped with a buggy firmware, which can be fixed by updating to a newer version.

The errors in your screenshot do not correlate with the known issue caused by the bad firmware. The errors you are getting are typically issues caused by a bad SATA cable or SATA port. Try a new SATA cable and a different port. I am pretty sure these numbers won't increase anymore (in fact, the error rate might even decrease) so they are nothing to worry about if they don't keep increasing.

 

[iameatingjam]

And here I was thinking the 870 evo was the last decent sata ssd since it hasn't gone to qlc no dram. And it has reliability problems.... great. Or is this only certain older batches?

 

[Jon555]

I have a couple of 870s, both are working fine, one had been running constantly for over 3 years... I bought them after the bad firmware thing...

 

[tpousr]

Or is this only certain older batches?

The 870 ist a good SSD. Only drives with a certain firmware had issues causing data loss due to the firmware marking sectors as bad/unrecoverable.

This could be fixed by a FW update (however the lost data could not be recovered).

I had a drive which was affected by the bad FW which also marked a few sectors as bad. Did a FW upgrade and secure erase afterwards and have not experienced any issues ever since despite writing almost 25TB to it again. So I'd say as long as your drive is running the latest firmware it's a good and reliable drive.

 

[RJARRRPCGP]

The 870 Evo is not a defective model. It's just that some batches shipped with a buggy firmware, which can be fixed by updating to a newer version.

The errors in your screenshot do not correlate with the known issue caused by the bad firmware. The errors you are getting are typically issues caused by a bad SATA cable or SATA port. Try a new SATA cable and a different port. I am pretty sure these numbers won't increase anymore (in fact, the error rate might even decrease) so they are nothing to worry about if they don't keep increasing.

A bad SATA connection causes ATA port resets/retries, usually. For people with SATA, check the event log. IIRC, it tends to cause CRC errors as well. (which the "CRC Error Count" entry, IIRC, most likely means the same thing as ATA CRC Error Rate)

 

[PowerBuilder]

TLC SSDs are not reliable. Samsung is no different.

 

[Shrek]

?
I thought TLC was better that QLC

 

[RJARRRPCGP]

?
I thought TLC was better that QLC

Yeah, QLC is the one I'm worried about the most. Now, even TLC isn't bad like it was feared to be years ago. TLC is now sort of like MLC that came before, reliability-wise. Even when the smaller capacity versions can't handle as much writes.
I'm only seeing problems with mostly SATA SSDs now. Usually, they are questionable to begin with.

The WD Black SN850 / SN850X seems solid. I wouldn't be surprised if similar with Samsung 990 Pro. (Their NVMe SSDs tend to be top-notch)

 

[iameatingjam]

Usually, they are questionable to begin with.

I know right. There's no good sata drives anymore. The sata interface is enough for decent performance but none of the drives are made to take full advantage of it anymore. Sure they can for a short period.... but the days of high end sata ssds with dram and tlc with reliable high tbw and good sustained write seem to be a thing of the past, with the 870 evo being the only exception I was aware of.

Even the mx500 is discontinued now, but even before that its corners were cut badly. My new 4TB models are... very poor compared my old 1TB models when it comes to performance. I mean they still are fine for storage... treated more like hdds. But only 1000tbw for 4TB, and only 512mb dram for the entire 4TB, just pathetic. I'm glad they discontinued the brand, if crucial wants to sell crappy sata ssds I'd rather them not be under the mx500 name. And my bad for not doing my research on how much they changed.

Are you aware of any other good sata brands? Not that I'm in the market.... but it would be good to know.

Yeah, QLC is the one I'm worried about the most. Now, even TLC isn't bad like it was feared to be years ago. TLC is now sort of like MLC that came before, reliability-wise.

Yeah, I wonder how long it will be until people refuse to buy drives unless they are qlc, several years after whenever the next cell is invented that holds 5 or 6 bits or something.

 

[amiens80]

That wont work, a SSD is not the same as a HDD. An SSD swaps the files around the whole space to use each cells evenly...

Furthermore, bad cells are marked as bad and replaced by spare cells. Now your files only take places doing nothing, and the SSD has to swap them around the SSD the whole time.

i do that (renaming) unreadable files since a couple of months (maybe more than one year), and each time i test to copy the renamed files, they still unreadable/uncopyable. So, i think you're wrong. i never saw a SSD feature that "move" the filled cell, even when they are not read since a long time. I migh be wrong, but TRIM operation only move empty cells, not the filled one.

 

[remixedcat]

I just bought a used 870 Evo 500GB with original packaging from marketplace locally (quite a good price, had it been issue free), before I realized it is a defective model. Should have done more research before buying. But to my defense, it's not widely reported unlike the 980 Pro.

Came erased. Hadn't put my own files on it yet. 56 TBW, magician is already showing red circles around Uncorrected Error Count and ECC Error Rate. Reallocated error count is also non-0. Manufacturing date is Dec 2020, serial starts with S62ANJ. Firmware is SVT01B6Q (latest). Pretty certain it is affected by the problem.

Freshly downloaded Magician doesn't let me run diagnostic scans. Drive directly connected to SATA on the 3rd gen Intel motherboard. In ACHI mode. Genuine according to Magician. I can't think of any other reason why diagnostic scan is disabled. Maybe Samsung doesn't want to damage the drive further?

The seller didn't say a blip about a a failing drive. I don't think he's hiding anything. I even got him to send me a screenshot of CrystalDiskInfo before the sale (I take the blame for focusing on the "health percent" and not thinking about the other metrics)

Many sellers clear the SMART info so you think it's a fresh drive but it's not and even worse a knockoff.


I'm dealing with a drive issue on my 860 evo as well but I tried to get this drive cuz many server admins love tf out of it but it's so pricey for just 1tb and is always backordered!!

I have to go w sata since my precision t3600 didn't support pcie boot at all even hacks are dodgy at best...

 

[chrcoluk]

since i have issues on my 870, i take extreme mesure :
once a file is bad, i don't delete it, i just rename it as "bad_sector##"

Ok i lose space, but at least,i will not put new file of the bad cell !

SSD's do housekeeping which moves data around internally, so that file wont be just left in the same part of nand.

I know right. There's no good sata drives anymore. The sata interface is enough for decent performance but none of the drives are made to take full advantage of it anymore. Sure they can for a short period.... but the days of high end sata ssds with dram and tlc with reliable high tbw and good sustained write seem to be a thing of the past, with the 870 evo being the only exception I was aware of.

Even the mx500 is discontinued now, but even before that its corners were cut badly. My new 4TB models are... very poor compared my old 1TB models when it comes to performance. I mean they still are fine for storage... treated more like hdds. But only 1000tbw for 4TB, and only 512mb dram for the entire 4TB, just pathetic. I'm glad they discontinued the brand, if crucial wants to sell crappy sata ssds I'd rather them not be under the mx500 name. And my bad for not doing my research on how much they changed.

Are you aware of any other good sata brands? Not that I'm in the market.... but it would be good to know.



Yeah, I wonder how long it will be until people refuse to buy drives unless they are qlc, several years after whenever the next cell is invented that holds 5 or 6 bits or something.

Would be great if sites like TPU brought a new retail every year or so to verify any bait and switch shenanigans.

I have a 4tb 870 evo purchased for my linux STB, which feels like a disaster waiting to happen but at least I stuck a USB stick in there for the auto recording timeseek feature, so the writes will just be for scheduled/manual recordings. Failed usb stick will be far more convenient and cheaper to replace.

Also stuck one in my AMD rig, not sure if in my sig yet, to get some extra NAND in there. Although not put it to use yet.

I brought two WD SA510's, put one in my PC to replace a failed spindle, but then realised there is no DRAM on these and the reputation doesnt look great, although mine is a new 4TB model with no known history, so sent the other back to amazon which got replaced with the 4TB 870 evo. Luckily it will be doing very little writes, so it should be fine. These SA510 show factory defects in the SMART, mine shipped with 179, usually that info is hidden. New bad blocks are labelled as grown bad blocks which of course is 0. They do at least have 5 year warranty though.

 

[amiens80]

SSD's do housekeeping which moves data around internally, so that file wont be just left in the same part of nand.


fine. These SA510 show factory defects in the SMART, mine shipped with 179, usually that info is hidden. New bad blocks are labelled as grown bad blocks which of course is 0. They do at least have 5 year warranty though.

housekeeping can't be formed on unreadableable cell
so, i think my way is good.

To perform a housekeeping of a cell, you need to have a non-failure signal after reading the source cell.

 

[tpousr]

I'd like to correct or clarify some statements from the past posts.

An SSD swaps the files around the whole space to use each cells evenly...

What you've probably meant is that SSDs try to distribute writes evenly across cells. This is what the "wear levelling" algorithm of SSDs does. If the operating system writes a file to let's say something like block address 42, deletes this file again and writes it to block 42 again, the same chunk of data, written to the same block address, will end up in a different flash cell. On the contrary, when using an HDD, the chunk of data written to block address 42 will always end up at the exact same physical location (same platter, same track, ...).

SSD's do housekeeping which moves data around internally, so that file wont be just left in the same part of nand.

This sounds like an SSD is always shovelling all data around internally when being left alone. This is not the case and would dramatically shorten the lifespan of an SSD due to limited write cycles per flash cell.

The one time a SSD might move data around is when data is actively being read and the SSD controller notices an above-average voltage drop in the cell being read. Whenever this situation occurs a number of things may happen, depending on the SSD controller and it's firmware:

• The data might actually be unreadable if the voltage dropped below a certain readability threshold. In this case the controller has to report a read error to the operating system and the data is lost.
  • The controller might actually try a refresh of this cell before finally marking it as bad. The cell might still be okay, because a certain voltage drop is normal when a cell is being left alone for a long time.
• The data may still be readable. In this case the controller returns the data to the operating system and will normally try to write the data to the same cell again (recharge). After that it reads the data again and checks the voltage level of the cell. If it's still below a certain threshold the cell is marked as bad and the data will be written to a spare cell.

And the firmware update, which has fixed the problems with 870 Evo drives, is very likely changing that part of the controller operation. My guess is that the voltage threshold for cells has simply been adjusted to match the real-life behavior of the TLC flash used in these drives. This should have nothing to do with bad flash or anything like that. Another explanation for the early 870 failures could be a bug in the wear levelling algorithm that pushed some cells harder than others, causing them to die prematurely. Although this is very unlikely because Samsung has a lot of experience in these algorithms and they sureley would not develop this from scratch for this specific kind of drive.

And I have personally never heared of SSD which scan their cells regularily completely on their own. This would dramatically increase power consumption. I am not saying there are none. But normally they only do that either if a read is actually requested from the host or when performing a full/extended S.M.A.R.T. test.

Another occasion when data is moved around is when a single bit of a TLC cell is erased. The remaining 2 bits may be moved by the garbage collector later.

I migh be wrong, but TRIM operation only move empty cells, not the filled one.

TRIM does not really move cells. The only thing it does is writing zeroes to the sectors/cells being written. Why do we do this? Flash cells need to be erased (with a zero voltage) before they can be re-written. This may take some time. That's why writing to an empty cell will be faster than writing to a cell which is already holding a charge. And since we want fast writes we zero-out the cells when deleting data so writing to them again will be fast.

BTW: This is not the same as writing an actual "zero" to a specific sector. TRIM is more than that. It's telling the SSD controller: "erase this cell". Depending on the controller this cell may also be used as a spare cell in case other cells are failing.

EDIT: To be more precise, TRIM is telling the garbage collector of the SSD that a certain block/cell may be cleared. Normally this clear operation is not synchronous, because it is expensive. And the garbage collector might actually move filled cells. E.g. when 1 bit of a TLC cell was erased the garbage collector might move the other two bits of this cell to a new cell so the original cell can be completely cleared (for improved performance).

To perform a housekeeping of a cell, you need to have a non-failure signal after reading the source cell.

This is correct.

 

[Shrek]

Many sellers clear the SMART info

Is that even possible?

 

[remixedcat]

Is that even possible?

yes it is and it happens A LOT!