New 840 EVO fix adds ”periodic refresh” to firmware

Last year, some Samsung 840 EVO SSDs started exhibiting slower read speeds with old data. The problem was patched in October, but the fix didn't stick, with slowdowns returning months later. Samsung pledged to release another fix in March, and now, that update is scheduled for "later this month."

Since the EVO's slowdowns manifest over time, we won't be able to verify the effectiveness of the new patch right away. However, we can share some details about what the incoming fix entails. Here's what Samsung told us when we asked for specifics about how the new firmware addresses the cell voltage drift that seems largely responsible for the problem:

• Samsung revised the firmware algorithm to maintain consistency in performance for old data under exceptional circumstances. Therefore, read performance was restored without the need for Magician. This algorithm is based on a periodic refresh feature that can maintain the read performance of this older data. The algorithm does not affect normal user scenarios (i.e. occasional PC performance degradation due to background work of SSD) or the lifespan of an SSD and can actively maintain its performance without the help of Magician. However, this algorithm does not operate when the power is off.

• The read performance has been improved by the revised firmware algorithm. If performance recovery is slow in instances where the SSD did not have enough run-time for the firmware algorithm to reach normal performance levels, or similarly, had been powered off for an extended amount of time, the performance can be recovered by using the Advanced Performance Optimization feature in Magician 4.6. This is a supplementary feature to maintain normal performance for a few exceptional circumstances.

• Users can upgrade to the new firmware through Magician 4.6, without using the performance restoration tool.

Interesting. When users first encountered slowdowns with the EVO, they found that rewriting the old data brought reads back up to speed. It sounds like the new firmware's "periodic refresh feature" does something similar.

The refresh routine appears to run in the background, when the drive is idle, so it shouldn't affect performance in normal scenarios. Refreshing old data may consume some of the NAND's limited endurance, though. We've asked Samsung to clarify how frequently data needs to be refreshed and how this affects write amplification.

Since old data can only be refreshed while the drive is on, those who leave the EVO unpowered for extended periods will have to rely on the Magician software's optimization mechanism if they want to restore full performance quickly. It's unclear how long the EVO will take to optimize itself after extended downtime.

Comments closed
    • Razor512
    • 5 years ago

    Seems like a band aid fix for the issue. If cell leakage is so bad that it needs to be refreshed in order to maintain performance after just a few weeks – months, then the drive will likely not meet data retention expectations.

    Remember, they tried relaxing the threshold for when the ECC kicks in, and it still did not fix the issue.

    TLC flash is getting read issues in weeks that MLC would likely not get for 10+ years of stagnation.

    Also, while tech report got a lot of writes out of the flash cells, the true limit of the drive was about 100 TB of writes.
    After that, it started to rapidly burn through reserve flash, and even got a few uncorrectabe errors, in addition to retention errors (after just a few weeks of being powered off).

    When an SSD begins to constantly replace cells, then it is no longer reliable (This applies to TLC, MLC, and SLC).
    Uncorrectable errors are worst than disk failures, as it means data can be corrupted silently. E.g., an old important document has flash cells degrade or drift to the point where it becomes unable to read back the correct info. Your automated backups eventually loops and overwrites old data (unless you have enough storage to never have to overwrite), and you eventually end up with all backups containing the corrupt file, but you will not know until the one day when you need to access that file again.

    (How many people here individually open each and every individual file on their system in order to confirm that they can read properly before each and every backup?)

    • Krogoth
    • 5 years ago

    Protip: This a problem that affects flash memory in general. TLCs just have a much shorter shelf life then their MLC/SLC counterparts.

    Flash memory is ill-suited for archival data storage.

    • DarkMikaru
    • 5 years ago

    So it makes me wonder what percentage of 840’s have this problem. I’ve got 2 and neither exhibit these issues. One in my home file server (120gb) & another in my laptop (250gb) and probably about 2yrs old by now. But i guess even if both are inflicted with the issue both boot Windows quickly and perform daily computing tasks just fine.

    Hope this helps. As I’d hate to see the EVO series take a hit.

    • ozzuneoj
    • 5 years ago

    And still nothing about the 840 non-EVO? I put one of those in a friend’s computer a couple of years ago, after reading all of the reviews talking about how a 128GB TLC drive would still last 10+ years (for normal desktop use) before wearing out.

    I haven’t yet heard any complaints from my friend, but I’m sure its going to be a major problem before too long.

    I have an 840 EVO 256GB in my laptop, a second hand SM841 (840 Pro) in my desktop and an 830 in my wife’s laptop, but I highly doubt that I’ll ever install a Samsung SSD in someone else’s computer from here on out, and there are plenty of other less expensive drives that meet my needs now, so I’ll probably go with something other than Samsung for the next one I use in any of my own systems.

    • thorz
    • 5 years ago

    They speak about their darn Magician software and that it has this function to fix they mess, but what about us that have installed this crap on a Mac? Are they basically saying that we that have Macs must have then on for extended periods of time from now on so that they firmware algorithm can fix their crap because they don’t care to give OSX the same support they give to Windows and give us a Mac version of they Magician peace of dunk? Why the f*ck do they say that their product was adequate for OSX in the first place if they were not going to support it in the same line they do on Windows?

    I hate these f*ckers that sell their products to both Windows and Mac users but after the sell deny and discriminate their customers with the appropriate support. Samsung, you were happy to take my money so give me the same tools on OSX that you give to Windows users!

    • Chrispy_
    • 5 years ago

    Total kludge, and basically a very carefully-worded piece of damage control that is doing everything possible to avoid stating the literal translation:

    [b<]"These drives are faulty; We dun goofed."[/b<] I'm not sure what disappoints me more; The fact that they can't fix their 840EVO or the complete denial that the problem even occurs on the vanilla 840.....

    • rechicero
    • 5 years ago

    As a not so proud owner of a slowed 840 (not EVO) all I understand is: “Samsung is the new OCZ, avoid those SSD like a plague”.

    • ronch
    • 5 years ago

    1. So they claim this fix won’t degrade the cells faster, but wouldn’t it, really? Every time cells are written to they slightly degrade, that’s something you can’t get around of.

    2. Does this make the drive consume more power? Would it generate more heat?

    3. Will Samsung Magician 4.6 be able to flash the drive even in an AMD machine? I hate it how it can’t work with AMD’s SATA/AHCI drivers. I had to wait until I had the time to reformat my system drive (which is the EVO, of course) before I could flash in the first fix. I had to do it before installing AMD’s chipset drivers.

    4. [quote<]However, this algorithm does not operate when the power is off.[/quote<] Um, yeah, I would think so.

    • Arvald
    • 5 years ago

    So they promised March… then the last we heard they said today (April 14) now later this month… so when in May should we expect the next delay?

    Sorry but not liking to have them clamp down on the coverage before allowing this out today.

    Official today my performance on mine at reads nearly 60% of what it was in February, and writes at 50%. Freshly tested using Magician.

      • Mark_GB
      • 5 years ago

      If you think this is bad, you obviously are not an engineer or programmer. Or perhaps you would prefer a quickly tossed together fix that was not tested properly…

      Yes, I have an 840 EVO too… Yes its slow too… But I am being patient because this “slow” is still hundreds of times faster than a hard drive, and the SSD is working just fine, even if it is a bit slower than it should be.

      They will get the fix out. And it will work. Just learn a simple word… PATIENCE!

        • Arvald
        • 5 years ago

        It is not the the delay nor the time for a fix that bugged me it was the clamp down on releasing details that seemed to happen. Acknowledge there is an issue and make your work on fixing it public.
        As to comparing to a hard drive according to my numbers vs a WD Black 2 (as per the spec sheet) my EVO is less than double that speed.
        As everyone who bought one we paid good money to have the speed and expected reliability and integrity of the name behind it.
        I am a patient man as well as a programmer, do keep your yelling to your self.

    • adisor19
    • 5 years ago

    So basically, Samsung TLC nand is defective by design.

    We need to know if other TLC nand manufacturers have experienced the same issue. Pretty sure Apple used TLC nand in some of the 128GB iPhone models and were swiftly and quietly exchanged as soon as users complained.

    Refreshing the data periodically will automatically increase write amplification and wear out the flash faster on an already fragile TLC nand… *sigh* they should just do the right thing and recall all those SSDs. Many users will experience this and will have no idea what’s causing. Not everyone is aware of firmware updates.

    Adi

      • just brew it!
      • 5 years ago

      As has been stated multiple times on this thread by multiple people, the write amplification effects should be relatively minor.

      The implications for long-term data retention and performance of infrequently used drives is a much bigger concern IMO.

      • ronch
      • 5 years ago

      I still applaud Samsung for paving the way for TLC. As these things get cheaper and cheaper manufacturers need to find ways to drive costs down. I am not defending TLC here and wish manufacturers would stick to MLC, but if someone doesn’t get on the TLC bandwagon and try to perfect the technology, who will? Yeah, I know this reasoning is kinda too forgiving to Samsung but it’s not like consumers don’t have MLC options anymore. It’s just that for me, I like how Samsung is exploring and perfecting alternatives.

        • derFunkenstein
        • 5 years ago

        Yes, thank you, Samsung, for bringing out a defective product before it was ready.

        • Topinio
        • 5 years ago

        [quote=”ronch”<]I like how Samsung is exploring and perfecting alternatives.[/quote<] You like "inadequately" ?

          • just brew it!
          • 5 years ago

          “Perfecting”? More like lobbing it over the fence and declaring victory.

    • puppetworx
    • 5 years ago

    We really need the press to, uh, press… Samsung as to why they aren’t issuing a 840 (non-EVO) fix also. Have they even acknowledged the problem?

    • TopHatKiller
    • 5 years ago

    I’m sorry here, but this is pretty much why I decided to avoid Samsung [and Intel] SSD’s.
    They prattle on about extreme validation, testing up the-kazoo [I think that’s a technical term]
    and so on… but in reality they seem to constantly have more bugs problems then other providers. I’m glad I went off and brought Crucial/Micron. Think I should of brought a bigger one though.

      • divide_by_zero
      • 5 years ago

      Fair enough about Samsung’s bugs, but I think you’re way off base with Intel. I mean, it’s you’re money and all; do whatcha want. But Intel is pretty much the gold standard as far as SSD reliability goes.

      The 320 series had an early firmware bug that hit a very small percentage of users, but IIRC their SSDs have been bug-free otherwise and widely regarded as the best in the biz.

        • Chrispy_
        • 5 years ago

        [quote<]The 320 series had an early firmware bug that hit a very small percentage of users[/quote<] A bug that Intel publicly admitted to immediately, investigated, fixed within a timely manner and quashed for good, restoring the affected drives to 100% effectiveness again. IIRC that was the last firmware bug in ANY consumer SSD from Intel.

        • Ninjitsu
        • 5 years ago

        Yup, I’ve got 3 Intel drives, no problems with any of them, and it’s been almost 2 years. They are indeed the gold standard for reliability.

          • TopHatKiller
          • 5 years ago

          Whoa there! Yeah, the points are valid – but, second ssd to go down in TR’s fantastic endurance test? Yep it was Intel. There are good arguments you can maker in defence of Intel, I wouldn’t deny that, but TR’s great test does not inspire confidence, does it? And, unless I imagined it, there has been more then one failure from them – Intel has been obsessed in maintaining a market reputation for unimpeachable reliability, across their products, but in reality the truth undermines them. [cpu/chipset failures etc over the years]
          In any case I can’t stand them. Uhm. That might not be a great argument, but, erh, I don’t like them.

          Also, Mr.Ninjutsu: I’m very glad your drives are fine, may they remain fine for many years to come, but if they broke in two years that would be pretty bloody awful.
          Uhm, also, I humbly suggest YOU DO NOT SAY x and y are fine; that’s just begging for something horrible to go wrong with anything. Cheers!

            • sweatshopking
            • 5 years ago

            Much better posts than previously. I assume these aren’t alcohol fuelled?

            • TopHatKiller
            • 5 years ago

            Au contraire! Of course they are.

            • fade2blac
            • 5 years ago

            Perhaps you should re-visit the TR endurance test results. The Intel firmware forces the drive to brick itself once it has exceeded a pre-determined limit. The drive worked essentially flawlessly up to this point and could probably continue to work if not for the self-imposed expiration date. Your comments misinterpret voluntary retirement for lesser reliability. Rather, I would suggest this is just Intel being overly cautious. They consider your data too important to “let it ride” on chance.

            [quote<]Intel's 335 Series failed much earlier, though to be fair, it pulled the trigger itself. The drive's media wear indicator ran out shortly after 700TB, signaling that the NAND's write tolerance had been exceeded. Intel doesn't have confidence in the drive at that point, so the 335 Series is designed to shift into read-only mode and then to brick itself when the power is cycled. Despite suffering just one reallocated sector, our sample dutifully followed the script. Data was accessible until a reboot prompted the drive to swallow its virtual cyanide pill.[/quote<] This is an entirely predictable terminal condition. The drive is intentionally designed to stop working as opposed to pushing until something catastrophically fails without recourse. I think Intel is overly conservative, but the end of life behavior of each SSD vendor seems to be unique. Intel decides when your drive is done long before things have a chance to degrade to the ragged edge. This behavior is well understood and one at least has a chance to retrieve their data before powering down the drive one last time. Other vendor's drives have a mode of failure that can be wildly unpredictable and while the drive may operate significantly longer beyond it's official rating, it is at the expense of the ever increasing probability of sudden and unpredictable failure.

            • ronch
            • 5 years ago

            So deliberately bricking the drive is better than just letting the drive’s cells work until they simply can’t work anymore? If another company such as Kingston or OCZ did that people would be crying for blood like those crazy people the world over crying for Jihad. IMHO, deliberate bricking is just as bad, if not worse, than increasingly getting read/write errors.

            • TopHatKiller
            • 5 years ago

            Yep yep; it [i<]still[/i<] failed. And, as I said - can't stand the company.

    • Deanjo
    • 5 years ago

    Seeing this:

    [quote<]However, this algorithm does not operate when the power is off.[/quote<] But supposedly "fixes" the issue screams to me that the real actual issue is the TLC NAND's data retention is short lived thus requiring a lot of error correction leading to the slow down of speeds.

      • fade2blac
      • 5 years ago

      It does indicate that this is a compromise to offset shortcomings of a product that is fundamentally flawed in such a way that it cannot meet the specifications under which it was marketed and sold. The only saving grace may be that the fairly conservative endurance ratings for most SSD’s should mean the flash memory can reasonably be expected to soldier on even under the additional burden.

      However, even though TR’s own SSD endurance test showed some impressive and unanimous (but anecdotal) evidence of just how much an average SSD can handle, for practical reasons long term retention (ie. up to 1 year) and temperature variability were not part of the test. According to JEDEC, consumer class SSD’s should be able to retain data for up to 1 year @ 30 deg C (based on an operating temp of 40 deg C). Furthermore, operating temperatures as well as power off temperatures appear to have a direct effect on the length of said data retention. This leaves me a little less willing to accept a reduction in available P/E cycles even if there “should” be plenty to go around.

      [url<]http://www.jedec.org/sites/default/files/Alvin_Cox%20%5BCompatibility%20Mode%5D_0.pdf[/url<]

        • meerkt
        • 5 years ago

        But supposedly that retention period is the minimum, and once a cell has used up all its P/E cycles. Stuff I read from some years ago suggested new cells could be good for 10 years of retention. But that was with much larger manufacturing processes.

        An interesting thing about that table from Intel on page 27 of that PDF: higher active temperatures increase the retention time when off. I wonder how that could be, and if one should consider adding heater devices to SSDs.

        Another thing to note: -5C of storage supposedly doubles the retention time. Assuming it doesn’t bottom out, maybe… long term storage should be in a freezer?

          • fade2blac
          • 5 years ago

          The explanation for the temperature effect is basically that it is easier to charge the NAND cells at higher temperatures and, conversely, the trapped electrical charge leaks away more slowly at lower temperatures. Electrons are more mobile in a semiconductor at higher temperatures because the extra thermal energy reduces the amount of additional voltage required to move charge across the insulating layers (ie. band-gap). As NAND cells keep getting smaller with each process node shrink, effects such as leakage and parasitic capacitance have a proportionately larger effect since each cell holds fewer and fewer electrons and they are more tightly packed.

          The ice cube tray example used by an Intel engineer is a pretty convenient one. Water is often a familiar way to explain electricity so that is easier to visualize and understand. Charging the cells is like filling the tray with water, where each individual cube section of the tray is a NAND cell. Higher temperatures essentially increase the flow rate for both filling the tray and, in the case of this example, leakage (these trays also leak slowly). In the case of TLC NAND, the drive must be able to accurately distinguish between 3-bits of resolution, ie. 8 discreet levels, of water in each cube section (12.5% increments if evenly divided, however it is also possible to overcharge NAND cells so 100% is not the absolute limit). For example, if a cell gets charged to 75% of capacity (6/8ths) and 10% of the water/charge is being leaked away over time, then this makes it significantly harder to detect the correct level. At this point, the cell would measure 67.5% (75%-7.5%) which is exactly half way between 62.5% (5/8ths) and the original 75% (6/8ths) so the drive controller needs to decide whether this cell was supposed to be at 75% or if it was just overcharged from 62.5%.

          For MLC drives, the drive only needs to be able to discern 2-bits of resolution, ie. 4 discreet levels, of charge in each cell. This makes the above problem much easier to tolerate with MLC since there is relatively more wiggle room for leakage and overshoot before things start getting murky. Similarly, if one increases the size of the cell by using a significantly larger process nodes such as in the current 850 EVO (40nm TLC V-NAND vs 19nm TLC NAND), then the above problems, which are exacerbated by smaller geometries, become less significant and much more manageable. This is why the 850 line is probably safe from noticeable voltage drift/charge leakage effects.

            • chuckula
            • 5 years ago

            Excellent explanation! This needs bookmarked. Also includes some interesting parallels to thermal runaway conditions in regular logic transistors.

            • meerkt
            • 5 years ago

            If a higher temperature helps when writing, why would it help with retention time? I assume a successful write leads the the same result, regardless of the time it took to write. If it indeed stores more charge, then perhaps a way to improve retention is to slow down writes.

            I wouldn’t mind having a drive with user-configurable operation settings, perhaps even exposed and used by the filesystem directly. From slow write/long retention to fast write/short retention.

    • DPete27
    • 5 years ago

    Geoff, any way you can contact Sandisk about their Ultra II series SSD’s that use Toshiba TLC nand and Marvell controller? [url=https://techreport.com/forums/viewtopic.php?f=5&t=96590&start=90<]I tested my drive[/url<] and it didn't exhibit any of the performance degradation seen in the 840 EVO even after 3 months of inactivity. It would be interesting to know how their firmware deals with this issue (if it even is an issue with Toshiba TLC?). Figured you have more clout than me.

    • Welch
    • 5 years ago

    Just spoke to a support agent at Samsung with the intent to demand an exchange or refund. The Samsung employee tried to tell me that the 4.6 Magician software update (firmware that is installed by it) would not cause additional wear on the NAND and would in fact be a firmware algorithm update. From his description the firmware would (not all at once) slowly go over cells and as it sees them drifting out of spec correct them, but not cause NAND wear………???

    Anyhow, he suggested I read this in depth article regarding the issue and assured me in was spot on as far as Samsung is concerned. It also includes an official response from Samsung.

    FWIW – [url<]http://www.pcper.com/reviews/Storage/Samsung-Magician-46-and-840-EVO-EXT0DB6Q-Firmware-Review-Finally-Fixed[/url<]

      • DPete27
      • 5 years ago

      Wow, I think you’re overreacting. I’m not sure what the refresh frequency is, but as a TR reader, you should know from TR’s SSD endurance test, that even with a few extra write cycles here and there, your SSD is likely to far outlast it’s warranty regardless.

      • cobalt
      • 5 years ago

      Okay, not sure what to make of that.

      The pcper article you linked does have the official response from Samsung. In that official response from Samsung, it says “This issue had been reported for the 840 EVO SSD only.”

      But the article itself, which Samsung pointed you to, contradicts Samsung’s official statement — pcper says: “… disappointed to see Samsung still ignoring their other TLC SSDs, on which many have reported seeing the same type of slow down (us included)”.

      So one one hand, we have Samsung denying the standard 840’s are exhibiting the issue, and on the other hand, we have Samsung officials claiming the article, which darn near calls Samsung liars about which drives are affected, is spot-on correct.

        • Welch
        • 5 years ago

        Yeah, I found it funny that he was allowed to point me to the article. While I found it also interesting that he would not email me the link (smart move) for deniability. It sounded to me like a single employee in a call center with his own opinion about the issue. Figuring it has official responses from Samsung, he probably though it kosher to suggest reading.

        No doubt what-so-ever that Samsung is purposefully ignoring the 840 series for all sorts of reasons. End of warranty in sight, age of the drives, time required to fix the issue, a lot less users with the 840 vs the 840 EVO. All around its one that their legal guys most likely told them would be best brushed under the rug. I don’t think they will ever acknowledge the issue and so long as the 840 EVO is “fixed” most people will forget about it save for a few poor enthusiasts who probably won’t buy Samsung again.

        I know that I’m buying Crucial drives as is for their performance, reliability record thus far and they are cheaper (plus MLC!). BX100 and MX200’s all of the way for now.

          • cobalt
          • 5 years ago

          I was pleased enough with my 840 that I bought an 840 EVO for a second system. I’m generally one to be pretty tolerant of companies taking time to fix things, but it sure seems like they’re sticking their fingers in their ears and going “LA LA LA I CANT HEAR YOU”. If they put up any evidence, or even explanation, of their assertion that vanilla 840s are unaffected, I’d be inclined to believe them, but all we have is:
          “Only EVOs are affected.”
          “Vanilla 840s are too, here’s proof!”
          “SHUT UP. Here’s another fix for the EVO.”

      • just brew it!
      • 5 years ago

      Given that it takes weeks to months for the drift to become an issue, the amount of additional wear will indeed be small, assuming they only refresh on an “as needed” basis.

    • Palorim12
    • 5 years ago

    PCPER posted a more detailed review of the new Magician and FW fix for the EVO:
    [url<]http://www.pcper.com/reviews/Storage/Samsung-Magician-46-and-840-EVO-EXT0DB6Q-Firmware-Review-Finally-Fixed[/url<]

    • cygnus1
    • 5 years ago

    So, does anyone want to join me in contemplating a real class action lawsuit and recall of the entire product line?

    I’m honestly fairly serious about this. If the drive has to constantly refresh (re-write) cells and the cells have a specific endurance, this drive can’t possibly end up lasting as long as it’s rated for.

      • alloyD
      • 5 years ago

      Although I’m not opposed to the idea in principle, TR’s own SSD endurance articles showed that actual SSD endurance can far outstrip “rated” endurance, so I’m not sure if you’ll be able to show that the firmware fix reduces life beyond the actual spec.

        • Srsly_Bro
        • 5 years ago

        And a conclusion based off of one or even 10 samples is useless in providing usable data about the population.

          • sweatshopking
          • 5 years ago

          Not true. I use my fatter friends as my base population to make me feel better about my gut. Seems to work.

      • divide_by_zero
      • 5 years ago

      Eh, seems like an overreaction to me. I’ve been very critical of how Samsung is handling this in the past, but I don’t think it merits a class action suit.

      I get your point, but it’s not like these things are dying in droves, nor are they corrupting data. And things fail to survive what they’re “rated for” all the time. Check out the MTBF numbers on mechanical drives sometime for a good laugh.

      As far as I see it, if the drive fails in warranty, get it replaced. If it outlives the warranty period (which seems extremely likely based on the TR endurance tests), great. At that point, either replace it on principle, or, ya know, just keep running it as you would any other piece of tech…?

        • cygnus1
        • 5 years ago

        First off, it’s the principle of the matter. But even without that, I really can’t trust the drive now. I can’t trust Samsungs firmware to report accurate anything. And I have to expect these 3 drives will die prematurely. I’d bet this effect is worse on larger drives too, as much more data will be fairly static. The resale value of the drives is also going to be lower than if the drive didn’t have the issue. This firmware update “fix” has also materially altered the specifications of the drive. Idle power consumption is going to increase now too. Battery life of mobile devices with “fixed” 840 Evo’s will be less than what they were, and less than what they should be.

        This would be like Intel “fixing” a CPU by having the turbo boost randomly boosting your CPU to 10 Ghz because some instruction ran erroneously slow. It’s doing something it wasn’t originally sold to do, wasn’t in the spec, and prematurely wears the product out. Would you expect people to just accept that too?

        And god I hate car analogies, but it’d be like a car manufacturer “fixing” a recall by changing your idle RPM to double the RPM. sure it’ll still go down the road and you’ll never notice much if anything, but wtf is happening to that engine in the process? And there goes your fuel economy too. No, it won’t be ridiculously different, but you’ll still notice.

          • divide_by_zero
          • 5 years ago

          So, before I respond to your points, a quick story. My wife and I recently needed to replace the failing washing machine that came with the house we bought last year. While researching what to buy, it quickly became clear that Samsung was the manufacturer to beat. They had the highest-rated model in our target price range, as well as being rated number one in washing machine customer satisfaction by J.D. Power for like 5+ straight years. As I was getting close to decided to purchase the Samsung, I remembered how I told myself that I was so annoyed with how they’ve handled this firmware mess that I would take my business elsewhere when I had the opportunity to do so.

          Just telling this story to reenforce that I am *not* a defender of how Samsung has handled this. It’s been botched through and through, and this kludge of a fix is long overdue.

          That being said, I still don’t have any idea why you think this merits a class action suit. “The principle”? Because the resale value is lower? Because you think the fix has altered the specs in some unspecified way? Some of these points are debatable while others are ridiculous. Worth getting frustrated about? Absolutely – it’s a crappy situation. Worth a lawsuit though? Not in my eyes.

          And you’re right – boo on car analogies. Especially bad ones. If refreshing the data once every ~4 weeks (which seems like a plausible time frame given what users have reported prior to this fix) is really how this works, then I’d analogize this to a car periodically idling a bit faster, ya know, a couple days a month. Great for the engine? Nope. Anything that’s likely to dramatically reduce the lifespan by a significant margin? Seems unlikely.

          Semi-related: I’ve personally replaced or overseen the replacement of thousands of bad 7200.11 and ES2 drives from Seagate. Forget the firmware issue, these things were just horrifically prone to thousands of bad sectors right around year 2 of operation. I managed a mid-size PC repair center at the time, and from talking to others in the industry and looking at the Backblaze data, our results were not atypical. After getting through that mess, I will *never* buy another Seagate drive.

          So I vote with my dollars. Seagate and Samsung won’t be getting any of my money, nor will they be getting any from friends, family, or colleagues who come to me for advice on what to buy.

          You want to involve the legal system – well, it’s your right to try. Feels like unnecessary litigation to me, and the kind that’s only going to make the lawyers rich. (Unless you feel like getting ~$10 back per drive is going to make this all better?)

            • cygnus1
            • 5 years ago

            Obviously we’re not going to see eye to eye on this. That’s ok. Happens a lot on the internet. But I’m not giving them the benefit of the doubt. They burned that bridge in the way they’ve bungled this issue. No one except Samsung employees have any idea how often cells will need to be refreshed. We don’t know how aggressive the firmware has to be. We don’t even know if it’s the same for all drives. Unless Samsung goes into more detail about the fix and possibly even gives you a way to see the amount of refreshing your drive requires, I have to assume the worst possible case in that it will essentially be refreshing cells constantly. Is that a likely situation? Probably not. But it is possible because they haven’t ruled it out by being transparent about the fix. Ideally I would simply like the drive replaced with a non TLC model.

            Also, why this line? Do you disagree the specs have been altered?

            [quote<] Because you think the fix has altered the specs in some unspecified way? [/quote<] Not some unspecified way. They have stated the drives normal function has been drastically altered to maintain normal performance. The implications of the issues mean the specs have been thrown completely out the window. None of the performance or power consumption specs are even remotely trustworthy now.

      • just brew it!
      • 5 years ago

      The small amount of additional write activity required to refresh the data isn’t the real issue here; in the grand scheme of things it shouldn’t have a meaningful effect on drive lifetime. The bigger issues/questions are: A) If the drive is left unpowered for a few weeks, performance will suck until it has had a chance to refresh itself; B) if the drift is bad enough to affect read performance this badly, how long of a powered down period can the drive tolerate before actual data loss occurs; and finally C) will this make the drive more susceptible to data corruption in the event of unexpected loss of power, since this refresh activity will be happening at a low rate in the background even when the drive is “idle”.

      And yes, this fix sounds like a kludge.

        • meerkt
        • 5 years ago

        Regarding (C), unless it’s buggy I don’t think it should lead to corruption on power loss. And I suspect the drive already did some refreshes to help with retention. They may have just devised a more aggressive schedule/threshold.

        • cygnus1
        • 5 years ago

        You don’t know how much write activity it’s going to take. On a 1TB SSD it might have to rewrite 800GB or more every day it’s powered on and idle. You just don’t know what their “fix” is actually doing.

          • just brew it!
          • 5 years ago

          Well, we can hope that the refresh algorithm isn’t that stupid. Since it seems to take several weeks (or months) for the cell voltage to drift enough to be a problem, they should only be rewriting every few weeks at most.

            • cygnus1
            • 5 years ago

            Right, I agree, you would hope it’s not that stupid. But you really can’t know for sure because they’re not being transparent about the details. The only way we can hope to know is to hook a drive up and measure its power usage. That’s really going to be the only way to determine how busy this, now critical to performance, algorithm is. And that is terrible.

      • MadManOriginal
      • 5 years ago

      It will still last far longer than it’s rated for.

      • meerkt
      • 5 years ago

      Let’s say each cell is refreshed once a month. After 10 years you will have used 8.3% of the P/E cycles. And that excludes unused cells and spares.

        • cygnus1
        • 5 years ago

        You don’t know how often it will have to refresh. Its likely different for every drive, and probably different for each nand die in a drive. You might have one chip that needs to be refreshed every day. Or their algorithm goes haywire and just constantly refreshes cells for all idle time. Power usage goes through the roof and your drive dies quick. You just don’t know, they’re not giving enough detail.

          • meerkt
          • 5 years ago

          I don’t know the specifics of any SSD drive or firmware version, from any manufacturer.

          I’m not saying I like this drive, never liked the idea of TLC, but seeing that people restored their performance level with a single rewrite once every few weeks or so, assuming one refresh per month seems like a good guesstimate.

            • cygnus1
            • 5 years ago

            I don’t think that’s a good guesstimate at all. It took people a month to *notice* performance degradation. I would imagine the SSD controller can detect the voltage drift much sooner than that and perform a refresh of a cell. But we don’t know, they’re not giving us any details on the operation or the affects of this new algorithm that drastically changes the background operation of the drive.

    • barich
    • 5 years ago

    This isn’t a fix, this is a workaround. They’re defectve, full stop. They ought to be exchanging these for an unaffected model for anyone who complains, and they ought to be doing the same for the 840.

    I would also wonder if the drive is left unpowered for long enough if the controller will be unable to read the data at all. That’s even less excusable if it’s possible.

      • derFunkenstein
      • 5 years ago

      But are there any unaffected models? What about the 850 family? Is it affected, too?

        • smilingcrow
        • 5 years ago

        I haven’t seen any reports of the 850 Evo having an issue but it’s relatively early days for that model. Even so I imagine enough owners are tracking it for it to be reported if an issue arises.

          • dragosmp
          • 5 years ago

          Apparently the 850 can’t really have this problem, or better said it can’t be diagnosed. Here’s why, hope I got it right. The 840 EVO’s NAND chips are of a certain speed which happens to be fast enough to saturate SATA. Once the cells’ voltage drifts, overall drive performance drops below the SATA bottleneck and the drift effect becomes visible.

          I’ve heard, perhaps on pcper, that 850 EVO’s 3D NAND chips are quite a bit faster than their 2D predecessors, so the 850 EVO new is badly bottlenecked by SATA. Assuming the cells drift and lose some performance, they will probably still be fast enough so that the bottleneck is still the SATA bus. Then the 850 EVO from the outside will work full speed even if the controller works to compensate the drift. Add to this that 3D NAND is more durable, Samsung can aford to refresh the drive in the background and burn some cell life without it being visible for the end user. Prior the 850, they had a few months of the TLC scandal to include something in the firmware of the 3DEVO from the start.

          The 2 cents I gathered is the 850 EVO won’t slow down, but will it last past the warranty? Should you buy one? Probably not when there are reasonable MLC alternatives cheaper, as fast, with good customer service and more reliable to boot.

            • just brew it!
            • 5 years ago

            The 840 EVOs were dropping *way* below SATA speeds. The effect was quite extreme in some cases.

    • derFunkenstein
    • 5 years ago

    So don’t put your SSD on the shelf or you can expect slow performance until it’s been powered on for the firmware to refresh everything, and don’t put it on the shelf too long lest ye lose all yon data.

    • anotherengineer
    • 5 years ago

    Econo drive, econo price. Be interesting to see how much of an affect this has on future sales of TLC Nand drives though?
    I also wonder if the drive is not powered for 2 years how the data integrity would be?

      • DPete27
      • 5 years ago

      [quote<]I also wonder if the drive is not powered for 2 years how the data integrity would be?[/quote<] How do you go about measuring integrity if all the data has disappeared?

        • anotherengineer
        • 5 years ago

        Easy, all the data is gone, I would call that a fail! 😉

          • DPete27
          • 5 years ago

          Well, there you have it.

      • jihadjoe
      • 5 years ago

      The sad thing is the Samsung Evo drives weren’t even priced accordingly. MLC drives from other manufacturers could have been had for less.

        • Wren
        • 5 years ago

        Yeah, they didn’t pass the savings down to the consumer, that is what annoys me :I (I have 6 builds with several 840 EVO drives each and I’m regretting choosing them).

      • Deanjo
      • 5 years ago

      [quote<]I also wonder if the drive is not powered for 2 years how the data integrity would be?[/quote<] No NAND manufacturers recommend a unpowered cycle of more than a year. Most recommend the maximum length of unpowered NAND be limited to 30-90 days.

        • just brew it!
        • 5 years ago

        Interesting, I wasn’t aware of that. I wonder how many read errors you’d get if you just took a pile of thumbdrives, wrote some data to them, and chucked them in a drawer for 2 years before reading them back.

          • Deanjo
          • 5 years ago

          USB sticks may have a longer retention time due to being manufactured with a larger process. The smaller the lithography is the more the issue of unpowered data retention is a concern. Micron is one of the few that say a years unpowered retention but that is only on their larger process 34nm flash.

            • meerkt
            • 5 years ago

            On the other hand, UFDs have less processing power and a less elaborate firmware. And as cheap devices that aren’t expected to see a massive amount of writes, they probaby use 2nd-grade NAND, with the good stuff going to SSDs, cellphones, and tablets.

            • Deanjo
            • 5 years ago

            Actually cell phones and tablets are where they can get away with using cheaper NAND due to those being in a “nearly always powered state”.

            • meerkt
            • 5 years ago

            Maybe in cheap ones? But even there I’m not sure that would be a good strategy for the manufacturer. UFDs are easily replaceable. Cellphones and tablet with unreliable internal storage are practically dead.

            • Deanjo
            • 5 years ago

            Cells and tablets are also easily replaceable. They are also devices that manufacturers want you to replace often. Do you really think those manufacturers who stop even sending out firmware updates even care about long term unpowered storage? Hell no. They know that most people replace their phone at about a two year cadence and it is highly unlikely for those devices to be sitting for a long term, unpowered, during that period.

            • meerkt
            • 5 years ago

            UFDs are cheaper. And phones, I would assume, are something people would be far less happy having to replace forcibly or unexpectedly.

            Using reject NAND in cheapo noname phones? Maybe. Brand name phones for several hundred dollars, I would expect, hope, use first grade flash.

            • Deanjo
            • 5 years ago

            I never said about using reject NAND, they just don’t need to use NAND with a long “no power retention”. You have way to much faith in because you are buying a premium phone that they will use NAND with long retention (and in essence rely on a larger process for that). They are going to use NAND that is more efficient while powered to maintain battery life.

            • meerkt
            • 5 years ago

            I didn’t mean different flash geometries/models, but chips that were binned based on various quality factors.

            For example, some of the stuff mentioned here about SD cards:
            [url<]http://www.bunniestudios.com/blog/?p=918[/url<] [url<]http://www.bunniestudios.com/blog/?p=3554[/url<]

          • anotherengineer
          • 5 years ago

          Interesting indeed.

          I think it’s time for TR to round up a bunch more SSD’s, throw on an OS on some, and just data on some others and pull the plug and put them in the closet for a year and fire them up and check the results.

          • Razor512
          • 5 years ago

          I have a few from around 4 years old, and read them today, and speeds were fine. Though the flash I the drive is likely in the 40+nm range, and MLC.

            • Deanjo
            • 5 years ago

            I’ve got a couple of Crucial V4 drives that get used sporadically. Like clockwork, after 3 months of an unpowered state, you plug them in and they are seen as unformatted requiring a format to bring them back to a usable state. I also have a Mushkin SSD that does the same thing after about 4 months.

    • cobalt
    • 5 years ago

    Can someone clarify the situation for the normal 840 (non-EVO)? I thought I’d heard reports that it’s affected, but haven’t heard whether these fixes apply.

      • Ninjitsu
      • 5 years ago

      Yup, seems affected (mine does) and no one’s talking about it so far officially.

        • ronch
        • 5 years ago

        I think Samsung’s just trying to get away with it.

          • Ninjitsu
          • 5 years ago

          Lol obviously.

      • fade2blac
      • 5 years ago

      I can say that the 500GB 840 (not EVO) in the girlfriend’s laptop has shown very similar symptoms. Like, Ninjitsu and many others, the 840 product line appears to be similarly affected, but frustratingly just not similarly acknowledged by Samsung.

      As I understand things, both the 840 and 840 EVO products use the same type of Samsung TLC NAND. The issue is widely accepted as being due to charge leakage/drift over time for Samsung’s TLC NAND cells. The flash controller has trouble differentiating the relative charge levels and so it requires substantially more time to successfully read older cells.

      If this is really the fundamental issue, then I don’t see why Samsung has chosen to say or do nothing for regular 840 drives. As a customer who paid for one such 840 drive, I don’t feel that I should continue to consider Samsung products when there are so many other alternatives from companies who stand behind their products or at least don’t take occasion to treat customers with indifference. I would accept a firmware solution, but perhaps I should just try and RMA the drive altogether. /rant

        • cobalt
        • 5 years ago

        I haven’t noticed any poor performance with mine, but I haven’t tried to benchmark the thing. I also leave my drive powered on 24/7, but I’m now confused whether or not being unpowered is partly to blame for the problem, or if it only inhibits the cure.

      • Pholostan
      • 5 years ago

      We have several ordinary 840’s (non evo) at work, I have one in my workstation. None of them show any signs of slowdowns. Most were bought a couple of years ago. All machines run linux, but not all are running 24/7. My workstation usually is powered down on weekends.

      • cwj717
      • 5 years ago

      My 840 non-evo drive has this problem, I posted about it in the “Samsung 840 & 840 EVO Bug” thread on the TR forums. Here are the benchmark results for my drive: [url<]http://s22.postimg.org/ukryoony9/840.png[/url<]

      • Chrispy_
      • 5 years ago

      Yes, Vanilla 840’s are definitely affected.

      No, Samsung is definitely not intending to admit that it’s affected.

      No, these fixes for the EVO are incompatible with the vanilla 840. but if you install diskfresh it will do exactly what Samsung are doing with their “kludge” fix by just making sure that your data never gets old enough to show symptoms of problematic TLC NAND with old data.

    • Welch
    • 5 years ago

    Sounds like its time to demand a refund/exchange to something not affected by this gross oversight from Samsung’s engineers.

      • thorz
      • 5 years ago

      I am really disappointed with the problem on these drives.
      If the 850 is not affected I would also like an exchange.

    • smilingcrow
    • 5 years ago

    TLC = (handle with) [b<]T[/b<]ender [b<]L[/b<]oving [b<]C[/b<]are

    • FubbHead
    • 5 years ago

    So crappy hardware then. Won’t touch a TLC drive for some time, and definitely not a Samsung drive, I’ve had enough.

      • xeridea
      • 5 years ago

      Not necessarily crappy hardware, just a shortcoming of TLC. The TLC flash is a lot more sensitive to voltage drift over time. The other issue is that the TLC drives aren’t much cheaper than MLC. These factors combined with falling SSD prices make it not worth the inconvenience.

        • smilingcrow
        • 5 years ago

        The NAND is a very big part of the hardware though and most of the cost.
        So far only Samsung’s TLC has issues so will need to keep an eye on SanDisk’s TLC as well as the 850 EVO.

          • RazrLeaf
          • 5 years ago

          Agreed. Though I wonder if the TLC in the 850 EVO will experience this issue, as it’s manufactured on a much larger process node. Perhaps this issue is only present on smaller node TLC?

          • thorz
          • 5 years ago

          Isn’t Samsung TLC also present on the faulty iPhone 6? The 128GB ones ?
          Mine is one of those and I can tell you that I have never experienced these app crashes and reboots on an iOS device before. I am waiting that Apple starts to change to MLC on these high capacity iPhones to go and change mine but until now there are no findings on these iPhones with MLC.

      • Srsly_Bro
      • 5 years ago

      My 830 256 GB drive is fine. Idk how the newer drives are tho……

    • smilingcrow
    • 5 years ago

    This sounds like a kludge to me.

Pin It on Pinterest

Share This