Samsung 840 EVO and 840 user suffers serious preformance degradation over time

[pa1983]

The past two months people have slowly noticed that there is a problem with the Liner Read performance of Samsung 840 EVO, The 840 have also shown to have the same problem.

Update:

3dnews has written a article about it. Its in Russian but google translate works well enough.

http://www.3dnews.ru/901747

I have also updated the forum links below with more forum threads about the problem.

End of Update.

To make the story short the SSD will over time lose Liner Read performance in data stored for a long time. So you windows files, games, programs that are pretty static will reduce in read speed after a few months usually and then degrade rapidly. This is not how SSD's usually behave since the technology it self usually allows read speeds to stay unaffected but writes can degrade for a number of reasons.

Data that is new will show full read speeds so benchmark programs that write temporary files for reading will not show the problem.

I know a few forums where users have started to collaborate on the problem and I will give links to them down below.

http://forums.overclockers.co.uk/showthread.php?t=18615995

http://www.overclock.net/t/1507897/samsung-840-evo-read-speed-drops-on-old-written-data-in-the-drive

http://www.reddit.com/r/buildapc/comments/2g1jfi/seems_that_the_samsung_840_evo_is_notorious_for/t3_2g1jfi

http://linustechtips.com/main/topic/214537-samsung-840-evo-ssd-performance-degradation/page-3

If y our a member on any of those forums I and you are effected by the problem I suggest you make a post.

So the long story is that this is a unusually problem with an SSD since read speed even on a badly misaligned SSD should still be very fast but the 840 EVO and 840 can drop to 50MB/sec or less and for some go down to a crawl of 1-2MB/sec on some old files meaning it only gets a fraction of its max speed fresh out of the box. Problem is not just during benchmarks but confirmed by many that real files actually reads slower.

I have been involved with many effected users in IRC and other forums so its a big problem and not just a few effected users.

I my self own an 840 EVO and an 840 but also have access to another 840 500GB that is 1.5 years old that runs at an average 50MB/sec on about 1/5 of its storage space.

Theory's on the problem are many, bad ware leveling that places old static data in a way allowing slower access like only on channel being used from the controller. Others think the TLC is bad some how others thinks its a TLB bug with the look up table.

Samsung seems to be investigating but since there support is supplied by a 3:e party its hard to say what they are actually doing about it.

Over at overlocker.net a user developed a program that will allow on to test any files for read speeds to make it easier to find effected files instead of manually copying files to ram disk or an fast realiable SSD.

I my self have contacted samsung support.

I have also contacted sweclockers.com and nordichardware.se and they are looking in to it since they found the evidence very compelling and will have drives available for testing later.

Dont know yet what will come of that tough but I hope for an article.

Here are some images on how an effected 840 or 840 EVO will look like in HD tune 5.5. This one is a 1.5 year old 840 500Gb running windows 7 premium installed the normal way (no cloning etc) so TRIM is on (Dont effect reads any way) and should be aligned properly since windows 7/8 is SSD aware. It runs on the Intel Z77 Sata 6Gbit port

Some images of the same SSD using this program a user at overclock.net wrote.

I suggest using the trail version of HD tune 5.5 to test out the problem.

To get the speed map you have to wait for the read test to complete once thats done you can click on speed map.

HD Sentinel 4 also works but its speed map is not as visually representative but its graph benchmark is just as good.

Samsung support has apparently got multiple calls about the issue and they recommended using this link if you want to submit a complaint about the problem

https://contactus.samsung.com/customer/contactus/formmail/mail/MailQuestionGeneralNew.jsp?siteId=592&PROD_ID=G303850

For those that want a temporary fix to speed things up can either defrag the SSD. Yes you should not do that to an SSD since it has no purpose on a "proper" working one but since rewriting the data on the 840 and 840 EVO regains some if not all off the lost performance temporarily a defrag will re-orientate the data since the defrag program assumes its a hard drive and realy has no control over what LBA's is addressed to what cells since the SSD will change that all the time to even out any ware on the cells. Also using Secure Erase will work but for some max performance have not been restored but a new some what lower base line. But Defrag rely only works once on static data since the defrag program will think the data is ordered systematically since it dont know how an SSD works.

Secure erase should work every time but doing that every 2 months realy sucks. And I have spoken with users that have done it 2 months or so ago and its not a long term fix even if some users seems to think so but thats simply because the problem takes a few weeks at minimum to manifest again.

So I urge you all to test your drives, you can post dumps of HD tune 5.5, HD Sentinel 4 since they read existing data of the SSD.

There are other programs that do this to but you must know for sure that this is the case when you test. IF it writes data to do its read test performance will not be effected since what ever is wrong is done to the data after a longer period of time.

This issue is not a small one and even people thinking this is a user error have found there 840 and 840 EVO effected once they have tested them and just like me people owning two or more have found most if not all of there 840 and 840 EVO effected by the slow read speed on old data.

I my self have own other SSD's (intel 320, Crucial M4) and non of them have them has this problem. It only seems to be Samsung TLC based drives (840 and 840 EVO), MLC drives for other manufacturers and from Samsung so fare has not shown the problem so no need to worry about other units.

Edited September 17, 2014 by pa1983

[Camacha]

Are you very, very sure you are using a native port and not one on an external controller?

[pa1983]

Read what I wrote and posted. I have used SSD since generation 2 and know very well how they work. Ether read the threads created by other effected users where a lot more effected users have respond whit the same problem so this is FARE beyond user error and control when it comes to a fix.

Any one knowledgeable with how SSD's works can concluded that this is a Hardware/firmware issue. Question is what will samsung do about it.

I did not create this thread because there is one or two users with the problem but hundreds of reports from users that know they suffer from the same problem and there are fare more that are effected that are not aware of the problem as of yet. Peleny of threads where people have the problems to with out knowing it just by looking at the image dumps of HD tune or HD sentinel.

The threads I posted are just the biggest once available for public viewing. I can tell you there are more discussions going on all over the Internet about this.

Samsung have taken in 840 EVO's from users to diagnose the problem so its not some minor issue where talking here that the user can fix. A firmware upgrade is a minimum IF that can fix the problem. If its hardware related then a fix might not be possible on existing 840 and 840 EVO.

I very good at computer hardware and know perfectly well what sata chipset supports trim or not and what there I/O and bandwidth performance they have. So yes it uses native Intel Sata 6Gbit ports.

I have had at least 7 different SSD's and installed more in other systems and none of them behaves like this. Most of my machines run gentoo linux with the discard option and proper partition alignment.

The 840 500Gb I showed image dumps is a machine that runs windows 7 premium 64bit.

[Camacha]

Read what I wrote and posted. I have used SSD since generation 2 and know very well how they work.

I am always working with the assumption that even the most knowledgeable people mess up sometimes, because they do. The most obvious things often get overlooked. Since you claim to have a lot of knowledge I will leave you to your own devices and help somewhat more receptive folks. Good luck with your problem.

[pa1983]

So this tech site Article is laying then with the hundreds of use reporting the problem?

http://www.3dnews.ru/901747

Or another well known forum and tech site discusses the problem.

http://linustechtips.com/main/topic/214537-samsung-840-evo-ssd-performance-degradation/

Edited September 16, 2014 by KasperVld

[Nibb31]

I'm not sure why you are posting this here and not on a PC hardware forum.

[pa1983]

If I read the forum description properly this topic should be allowed I think.

I would assume there are plenty of people here that might own an Samsung 840 or 840 EVO that might find it useful.

Im also posting in hardware forums, more then one to be exact. This is the first on this topic I created just to make people aware of the issue not to cause trouble.

And I would like post to be on the topic, perfectly fine to ask questions but the problem exist and is not the fault of the users.

Edited September 16, 2014 by pa1983

[Nibb31]

If I read the forum description properly this topic should be allowed I think.

I would assume there are plenty of people here that might own an Samsung 840 or 840 EVO that might find it useful.

It's not that it's not allowed. It's just that it's not very relevant and just looks a bit random next to other posts about career counseling, Kerbals vs Grox, or Finnish folklore. It's just not the sort of place where you should expect people to take your post seriously.

I didn't even know what a Samsung 840 was before I read that you were talking about a model of SSD. I'm pretty sure there must be forums specifically for people who are looking for information about SSDs. There might be a couple of people here who understand (and care) about your topic, whereas on a PC hardware forum, you will probably get more support and more on-topic discussion than on a forum that is specifically about "off-duty chatter".

[ZedNova]

Well, i guess i'm glad i got a MX100.

[pa1983]

It's not that it's not allowed. It's just that it's not very relevant and just looks a bit random next to other posts about career counseling, Kerbals vs Grox, or Finnish folklore. It's just not the sort of place where you should expect people to take your post seriously.

I didn't even know what a Samsung 840 was before I read that you were talking about a model of SSD. I'm pretty sure there must be forums specifically for people who are looking for information about SSDs. There might be a couple of people here who understand (and care) about your topic, whereas on a PC hardware forum, you will probably get more support and more on-topic discussion than on a forum that is specifically about "off-duty chatter".

Yes but the thread is mainly to inform people that are unaware of the problem and might be effected. If they are member in any of the forums mentioned I assume they will go there. And even if they are not a member the information migth help them in a RMA case if thats the route they want to take.

Well, i guess i'm glad i got a MX100.

Yes seems to be a popular alternative and cheaper even.

[rtxoff]

I own even 2 840EVO, one with 250GB and the other 1TB. I have not experienced any slowdowns yet. However i did not benchmark yet.

[pa1983]

Well due to the nature of how SSD's works you might not notice it at first. An SSD even at 50MB/sec and access times in the low 0.2ms will still feel faster then a hard drive especially at random access.

But people have noticed slow downs in daily use. Depends very much on how heavy your use scenario is.

People that do daily backups have noticed a big increase in time it takes for the scheduled backup to complete.

Any way I hop the information was helpful to some one.

[Camacha]

Yes seems to be a popular alternative and cheaper even.

In a technological sense it also seems superior It has a different type of memory which, in theory, should last longer. Maybe the aforementioned problems are the practical fallout of this. The MX100 also has a form of power protection, which the EVO lacks. This makes the MX100 the superior choice in my eyes.

So this tech site Article is laying then with the hundreds of use reporting the problem?

[...]

I wasn't protesting the information, but the manner in which you responded to someone who tried to help you.

[rtxoff]

I am sure if there is a firmware problem with Samsungs 840 series they'll fix it as soon as they can but if it's a hardware problem, well that would make thing complicated. They would'nt risk admiting that something with the hardware is wrong. It would could cost plenty of money replacing the SSD's for the customers.

[Camacha]

I am sure if there is a firmware problem with Samsungs 840 series they'll fix it as soon as they can but if it's a hardware problem, well that would make thing complicated. They would'nt risk admiting that something with the hardware is wrong. It would could cost plenty of money replacing the SSD's for the customers.

Well, the big question is whether they would have to. The EVO's are cheaper because they used a different form of cheaper memory, but that means the consumer made that choice too. As long as the drives are under warranty and break down it is the manufacturer's problem, but after that or as long as they remain functioning, they are probably not obliged to do anything.

How that pans out for Samsung's name and PR, that is another matter.

[rtxoff]

The different form of cheaper memory doesn't justify the performance loss of some part's of the SSD which are under the level of conventional HDD's. It's clearly something wrong. People buy SSD's because the companies advertise them to be faster then HDD's and the customers expect that to be true. Samsung would loose every lawsuit if it comes to one.

[Camacha]

The different form of cheaper memory doesn't justify the performance loss of some part's of the SSD which are under the level of conventional HDD's.

I think we misunderstand each other. I meant that the choice of memory has consequences for the longevity of it. The user buys cheaper memory, so he can hardly expect the same endurance and reliability as the more expensive option.

Whether the current rate of degradation is reasonable is another discussion.

[pa1983]

I know the differens between SLC, MLC and TLC. NAND. In practice its just a question of the number of voltage states used during programing.

The 840 EVI uses15-18V to reprisent 8 states, 3bits.

But the 840 EVO also uses the TLC NAND in SLC mode for faster writes called torbowrite. Just uses some dedicated TLC NAND for that.

The MX100 is considered better since it has MLC NAND. But NAND is NAND, not realy that much of a difference.

Problem is that the more bits you reprecent the smaller the margines on the voltage used for programing. NAND cant keep data for ever ither.

What samsung saves is not cost per chip but they can use fewer chips for the same amount of storage.

The 840 500GB has less then 300h on it after 1.5 years. Used for surfing and storing photos. At the time 500GB MLC was way to much and even TLC whit less PE cycles would last longer then the machine. So no reason for a expensive MLC SSD at the time.

My linux router uses the EVO, it sees 10-100MB written to it a day I would say. I could never use all the PE cycles even on a TLC SSD.

[Camacha]

But NAND is NAND, not realy that much of a difference.

That's not totally right, but the rest of your post suggests you know that. TLC is more prone to errors, has lower write speeds, less durability (write cycles) and is more power hungry. Samsung has done a wonderful job of compensating for some of those, but the durability (combined with firmware) might be the issue we see here. The situation is a little more complex than just plain write cycles, as firmware has to deal with error correction and things like write amplification. We have seen in a number of other cases that getting everything right with the controller and firmware is still hard, as the technology is still relatively new and manufacturers are still figuring out how to deal with various things.

I really do hope it is not a hardware problem - or a hardware problem caused by firmware - as that would mean potential problems for a lot of people. Those EVO's are pretty popular (for good reason, I would say).

[pa1983]

I also know its less durable etc. Its not realy the users concern if Samsung messed up the reliability of TLC NAND if the user celected a SSD that has a life span that fits the users needs. On all our systems we will never hit the advertised write limit. Also only one system can fully exploit the performance due to sata 6Gbit, the rest are stuck on 3Gbit and are basically boot drives for Linux and some basic services.

Back when Gen1 and 2 SSD's where out static life span of data was some times in the specs. if I recall 10 years where not uncommon for SLC and MLC drives back then.

The basics of NAND is that SLC uses two voltage states to reprecent 0 an 1. MLC represents it with 4 states to encode 00, 01, 10, 11.

TLC needs 8 states to encode 000 to 111. The controller basically programs the cell with a voltage it knows represents a series of bits. Problem is if you try to store 3bits with 8 states in a 3V range the margin is half compared to MLC with 4 states and a 3V range. If I recall teh 840 EVO programs it cells with 15-18V so the 8 states needs to fit in there.

So that means the controller need to write very accurate voltage level but also be able to read it back later and determent what bits the voltage represent on the fly.

To me the later seems to be the problem. The TLC might just be degrading so fast that ECC has to kick in.

If I recall anandtech wrote that TLC used 24bit ECC and MLC 15bit ECC. Couple that with MLC being more proved and should take a lot longer to degrade where it would cause problems for the controller this might explain the slow read performance on 840 EVO afte a few months.

Also even if the TLC is at fault a firmware should be able to refresh cells making the Tech more viable. For light use, typical desktop or boot drive even if the TLC drive shuffled the static data around every so often it might cost 6-12 PE cycles on a 100% full drive etch year. Not a big cost I say with 1000 PE cycles.

From my small sample size and from what I see the 840 seems to degrade slower so I suspect the 21nm TLC NAND dont degrade as fast as the 19nm NAND in the EVO but thats just speculations. Need more drives to be sure of that. My 840 only degrades to EVO speeds after 11 months instead of 3 months so. That theory would to some part explain why we only see 840 owners reporting the fault at a limited number unlike 840 EVO owners and also why 840 owners reports it now and not before to a bigger extent.

I my self is no fan of TLC BUT Samsung good reputation, Sandforce 2 bad reputation, OCZ being junk the 840 and 840 EVO seemed like good enough for many people and the reviews where good to.

In my primary rig I run a MLC Crucial M4 SSD. Also in my HTPC i use an Intel 320. But if it was not for this degradation of the read speed on the 840 and 840 EVO I cant see what use MLC or SLC would do in a system that makes very few writes a year and hardly requires the performance of high end SSD's. SSD for me in my router and NAS is just to keep noise down, power use and save space instead of having an old HDD hosting the OS. The cost between a low end SSD and a low end HDD is not that big.

Edited September 17, 2014 by pa1983

[rtxoff]

Well, finally i was benchmarking my SSD's, the results are simply sobering. Both of my EVO drives suffer under severe performance loss on some parts of it.

I got both of them for aproximately a year in operation, the 250GB one has mainly the OS on it and most of the apps. The 1TB holds all of my games and other data like pics and videos.

Not sure what to do about it because i am not experiencing any performance issues.

Edited September 17, 2014 by gpisic

[Camacha]

Well, finally i was benchmarking my SSD's, the results are simply sobering. Both of my EVO drives suffer under severe performance loss on some parts of it.

Did you test this on drives with data on them?

[rtxoff]

Yes, as i wrote both are for one year in operation, the 250GB one using 70GB of it and the 1TB one using up 700GB.

Edit: I just hope the data won't degrade into an unrecoverable state. Maybe i should backup all of it onto an conventional HDD. Samsung you are giving me the headache.

Edited September 17, 2014 by gpisic

[pa1983]

As fare as I have seen the degradation ONLY happens in cells containing data so yea you wont see this on a new or unused 840 or 840 EVO.

So where the SSD is fast is where it has not stored data or stored data recently.

Windows users can use this program on there own risk. It will read actuall files and tell you how fast they read and how old they are. A user wrote it over at overclockers.net so us it on your own risk.

I used it on the 840 500GB with no problems tough.

https://dl.dropboxusercontent.com/u/27196441/840/SSDReadSpeedTester1.00.zip

Results will look something like this if your files reads slower the older they are.

Linux users can use this script.

https://dl.dropboxusercontent.com/u/27196441/840/SSD840bench.sh

Just add the dir after the script you wish to test. Root is not required.

If your drive is effected the best forum so fare seems to be over at overclockers.net. Link to the thread is in my first post. Registering is easy there. The UK forum denied my registration. Probably because its connected to a store and they want UK users only.

The sad thing is that pretty much every one that has tested there drives after being informed of the issue have had the fault. I have seen users that tested multiple 840 EVO being so sure the next on they test will have it they are willing to bet there lives on it more or less. So the issue seems to effect every TLC samsung model so fare with very few exceptions. New drives usually dont showt he problem until 2-3 months with static data on the.

Edited September 17, 2014 by pa1983

[Camacha]

As fare as I have seen the degradation ONLY happens in cells containing data so yea you wont see this on a new or unused 840 or 840 EVO.

So where the SSD is fast is where it has not stored data or stored data recently.

So, implicitly that means that when you erase your SSD (and/or refill it with the same data) performance is restored? Did anyone test that?