Samsung 840 EVO SSDs suffer from performance degradation

[+Warwagon MVC]

Fix on its way! :)

 

http://www.anandtech.com/show/8550/samsung-acknowledges-the-ssd-840-evo-read-performance-bug-fix-is-on-the-way

 

No ETA.

 

how does samsung update the firmware is it destructive or non destructive?... is it updatable through windows or do you have to burn an iso to a disc or make a bootable USB?

[+Nexus18 MVC]

how does samsung update the firmware is it destructive or non destructive?... is it updatable through windows or do you have to burn an iso to a disc or make a bootable USB?

Update-able through Samsung magician software.

[+Tyranade Subscriber²]

I have a Samsung 840 Evo and I am hoping they fix this soon now that they know about the issue. I am just hoping the issue is fixed within the next couple of weeks and there is no permanent damage to current drives.

[epk]

You could just back up the data, perform a ATA secure erase, and restore the data for better performance right?

On the overclock.net thread they were mentioning running a montlhy defrag using MyDefrag which can be booted from a cd or usb. I haven't seen such big dips on my system so I'll just wait for a fix.

[xendrome]

On the overclock.net thread they were mentioning running a montlhy defrag using MyDefrag which can be booted from a cd or usb. I haven't seen such big dips on my system so I'll just wait for a fix.

 

If you want to totally kill your drive, go ahead and do this. Defrags are not necessary on SSDs. Windows takes care of TRIM and other housekeeping on SSDs

 

http://www.mydefrag.com/faqgeneralinformation-willmydefragreducethelifespanofmyflashssdmemorydisk.html

[epk]

If you want to totally kill your drive, go ahead and do this. Defrags are not necessary on SSDs. Windows takes care of TRIM and other housekeeping on SSDs

 

http://www.mydefrag.com/faqgeneralinformation-willmydefragreducethelifespanofmyflashssdmemorydisk.html

No, no, of course not as a defrag. The idea is that doing that will move the data around and therefore making those files be read fast again.

 

Even if it works it's definitely not a long term solution but neither would be performing a secure erase as xWhiplash suggested.

[farmeunit]

If you want to totally kill your drive, go ahead and do this. Defrags are not necessary on SSDs. Windows takes care of TRIM and other housekeeping on SSDs

 

http://www.mydefrag.com/faqgeneralinformation-willmydefragreducethelifespanofmyflashssdmemorydisk.html

I know it's not really necessary and people say it will "kill" a drive, but how does it REALLY do that?  I know the flash only has a certain write life, but it only moves the data once to each block during defrag, correct?  Maybe twice depending on where it's at.  It'll move a block out, then write a block to make it inline, and so on, so each block only receives one extra write.  Is it the outside blocks where the temporary data is stored that get so many writes?

 

I would recommend doing it daily, but hardly see how doing it once a month would hurt.  I'm not an expert by any means.  Just curious.

[+Warwagon MVC]

I've heard running spinrite on level 2 also works. Level 2 is a read only.

[xWhiplash]

No, no, of course not as a defrag. The idea is that doing that will move the data around and therefore making those files be read fast again.

 

Even if it works it's definitely not a long term solution but neither would be performing a secure erase as xWhiplash suggested.

 

 

Defrag with an SSD is 100% pointless.  There is no moving parts, the data doesn't have to be read faster.

 

Just do a SSD secure erase using Parted Magic.  Only takes 15 seconds.  Restores any drive to its initial performance and does NOT wear out the drive.  A few months of use might make the drive slow again (especially if it is a firmware issue), but a SSD secure erase is pretty much known to give any SSD its default performance back.  Defragging does nothing.

 

Plus with wear leveling on SSDs, how will it be guaranteed to be "close to other files" like defragging a HDD does?  SSDs do not need to spin and seek for files at 7200 RPM, so defragging is pointless.

 

 

I know it's not really necessary and people say it will "kill" a drive, but how does it REALLY do that?  I know the flash only has a certain write life, but it only moves the data once to each block during defrag, correct?  Maybe twice depending on where it's at.  It'll move a block out, then write a block to make it inline, and so on, so each block only receives one extra write.  Is it the outside blocks where the temporary data is stored that get so many writes?

 

I would recommend doing it daily, but hardly see how doing it once a month would hurt.  I'm not an expert by any means.  Just curious.

 

Because as I have mentioned, it is POINTLESS write cycles.  SSDs have NO need for this, there is absolutely NO need to defrag an SSD.  Ever.  They are not 7200rpm drives.

[Mayhem]

more sites starting to report this issue, strangely neowin still hasn't

 

in PCPer they have made a very good report of this issue ---- http://www.pcper.com/reviews/Editorial/Samsung-840-840-EVO-susceptible-flash-read-speed-degradation-over-time

 

either way supposedly Samsung responded and promised a firmware fix by October 15, lets see...

[epk]

I've heard running spinrite on level 2 also works. Level 2 is a read only.

If that's true I'm curious as to how does it work. From what I'd heard the data has to be rewritten to go back to normal performance.

[gohpep]

 

 

 

Defrag with an SSD is 100% pointless.  There is no moving parts, the data doesn't have to be read faster.

 

Just do a SSD secure erase using Parted Magic.  Only takes 15 seconds.  Restores any drive to its initial performance and does NOT wear out the drive.  A few months of use might make the drive slow again (especially if it is a firmware issue), but a SSD secure erase is pretty much known to give any SSD its default performance back.  Defragging does nothing.

 

Plus with wear leveling on SSDs, how will it be guaranteed to be "close to other files" like defragging a HDD does?  SSDs do not need to spin and seek for files at 7200 RPM, so defragging is pointless.

 

 

 

Because as I have mentioned, it is POINTLESS write cycles.  SSDs have NO need for this, there is absolutely NO need to defrag an SSD.  Ever.  They are not 7200rpm drives.

 

No.

Yes, it is pointless, but this is a different case. The SSD is reading old files slowly, so defragging will make the firmware think it is new again and will go to full speed. SSDs nowawadays are durable enough anyway.

[TruckWEB]

I'm having a different kind of problem with my EVO 840, the disk seems to corrupt itself.  It happens randomly, Windows can't shutdown and when I try to reboot, Windows runs in loop trying to fix the disk and reboot and reboot...  Did a new partition, format, clean install and not even a week later, same problem again.

 

Using Win 8.1 x64.

[Boo Berry]

I'm having a different kind of problem with my EVO 840, the disk seems to corrupt itself.  It happens randomly, Windows can't shutdown and when I try to reboot, Windows runs in loop trying to fix the disk and reboot and reboot...  Did a new partition, format, clean install and not even a week later, same problem again.

 

Using Win 8.1 x64.

Might be a bad drive you've got there.

[+LogicalApex MVC]

If you want to totally kill your drive, go ahead and do this. Defrags are not necessary on SSDs. Windows takes care of TRIM and other housekeeping on SSDs

 

http://www.mydefrag.com/faqgeneralinformation-willmydefragreducethelifespanofmyflashssdmemorydisk.html

 

 

 

 

 

Defrag with an SSD is 100% pointless.  There is no moving parts, the data doesn't have to be read faster.

 

Just do a SSD secure erase using Parted Magic.  Only takes 15 seconds.  Restores any drive to its initial performance and does NOT wear out the drive.  A few months of use might make the drive slow again (especially if it is a firmware issue), but a SSD secure erase is pretty much known to give any SSD its default performance back.  Defragging does nothing.

 

Plus with wear leveling on SSDs, how will it be guaranteed to be "close to other files" like defragging a HDD does?  SSDs do not need to spin and seek for files at 7200 RPM, so defragging is pointless.

 

 

 

Because as I have mentioned, it is POINTLESS write cycles.  SSDs have NO need for this, there is absolutely NO need to defrag an SSD.  Ever.  They are not 7200rpm drives.

 

You're both severely missing the point here...

 

The reason the problem is occurring is because the TRIM function on the 840 EVO is defective. The drive is failing to actually move the data properly around the drive and instead is packing it all together in a single block. The amazing speed you're used to seeing from an SSD is due in part to the drive being able to read multiple blocks of NAND in parallel. If the data isn't spread across multiple blocks it can't be read in parallel so you'll only get read speeds as fast as a single NAND block; which is what is happening here.

 

In order to "reset" this you have to write the data to the drive again. Since the problem doesn't occur until the drive does GC. Writing new data to the drive is writing new data to the drive... It doesn't matter if you use a defrag utility or if you do a secure erase and write everything back again. You're still utilizing write cycles to get the job done.

 

Defraging and SSD isn't going to magically kill it. It is simply discouraged because under normal circumstances it won't lead to any performance improvement and would lead to wasted write cycles. This is not a normal circumstance and this case would benefit from a defrag...

[+Warwagon MVC]

If that's true I'm curious as to how does it work. From what I'd heard the data has to be rewritten to go back to normal performance.

 

FR. ROBERT: That's right, the man who provides us all the tools that we need to keep everything running, including ShieldsUP! and my personal favorite, SpinRite. Now, Steve, I talked about this last week, that I was having some issues with an SSD. And I still, I want you to explain to me how this actually works. Someone told me that the throughput issue I was having with some of my older Samsung SSDs could be solved by running SpinRite in Level 2. And I was thinking in the back of my head, I'm like, this is ridiculous, it's an SSD, it's not a rotating drive, that's stupid. I ran it, and by golly, it worked. How does this work? What magic did you bake into SpinRite to revive SSDs?

Steve: What we would like to believe is that, because they're solid-state, they're like RAM, where when you write the data, you are sure you're going to get it back. Now, even RAM is not perfect, which is why there's parity in one case, just to find an error when it's read back. We have parity and non-parity RAM. And there's also ECC on RAM. When really, for example, space shuttle RAM will have ECC because you can have, at the quantum level, you can have bits that don't read back correctly.

Well, the situation is far worse with SSD. Essentially, the technology of SSD is like dynamic RAM, where in order to get the bits small enough, dynamic RAM uses capacitors to store the bits as ones and zeroes, just uses electrostatic charge. And that tends to bleed off over time just due to leakage because the cells are so small, the capacitance isn't large enough. And it's a series of tradeoffs. And that's why you have to do so-called "refreshing" of dynamic RAM. You've got to come back and read it before it's bled out, before the data has sort of leaked away to a point where you can no longer differentiate the ones and the zeroes. So you keep coming back and reading it and rewriting it to, like, to recharge the little capacitors with their data.

Now, freaky as it sounds, that's the same technology as in SSDs. It's a much slower leakage, but it's still doing that. And so it actually - an SSD is just a huge plantation of little capacitors where charge is essentially stranded out on a plateau, and a field effect transistor is able to sense the field, the electrostatic field created by that charge. But over time, these cells weaken. And the important thing to understand is that, if engineers only needed to make a 1K SSD, oh, my god, it would be bulletproof, absolutely reliable. We could do that because, with so few bits, the bits could be so large that they could be reliable.

But the world runs on economics and competition. And so we've got multiple vendors who are competing with each other to get the highest density and the lowest cost. What that means is the smallest bits. So just as we have pushed hard disks to the point where they are now using error correction all the time, that is, you can't, you often cannot read a sector correctly on a hard disk because the engineers said, well, you really don't have to. We can correct it, as long as it's not too bad. So although, I mean, it's really cringeworthy because this is our data and we care about it, but the same thing has happened with SSDs. The engineers have pushed them so far that they're operating more in an analog fashion, not just one and zero, but somewhere between one and zero.

So what SpinRite is able to do is it's able to turn off, by talking to the drive, it turns off some of the sort of, oh, don't worry about this, we'll take care of it stuff, in order to show the drive when it has a problem which it would otherwise ignore. That forces the drive to address the fact that this area is no longer stored safely, that is, some of the bits are beginning to wander toward an indeterminate state. And that causes the drive to rewrite them firmly, but that takes a little bit of time.

See, SSDs don't write very quickly, as we know. They only read quickly. The reason is they actually have - there's a layer of insulation between this little floating island, and they have to ramp up a high voltage and push electrons through an insulator using high voltage in order to recharge that island. Well, that's why there's a limited number of writes that an SSD can do because every time you push electrons through that insulation, it weakens the structure of the insulation a little bit. And so that creates a lifetime on the number of times you can do that.

But so essentially what SpinRite does is it allows the SSD to be more picky. And instead of being lazy and using error correction to fix the sector which is becoming weak and taking more time, it says, no, let's, like right now, let's fix this. And so the SSD rewrites that sector which was using error correction so that it no longer needs it, which then speeds up the execution in the future. So, I mean, there is - we just plug these things in and format them and go. But there's an incredible amount of technology under the covers.

FR. ROBERT: Now that you say it, it makes so much sense. Rather than the drive trying to fix the errors on the fly as it's reading and writing, SpinRite just goes in and says, you know what, I'm going to fix everything for you, and you're golden. And that would explain why suddenly I get all my performance back. It wasn't necessarily that the drive was damaged, it's just that the drive was busy. But because the way that SSD drives work, it didn't want to do all that maintenance because in doing that, it's actually wearing itself out. I like that. That's fantastic.

[epk]

That actually makes sense, thanks! :D Should give it a try

[xWhiplash]

 

 

You're both severely missing the point here...

 

The reason the problem is occurring is because the TRIM function on the 840 EVO is defective. The drive is failing to actually move the data properly around the drive and instead is packing it all together in a single block. The amazing speed you're used to seeing from an SSD is due in part to the drive being able to read multiple blocks of NAND in parallel. If the data isn't spread across multiple blocks it can't be read in parallel so you'll only get read speeds as fast as a single NAND block; which is what is happening here.

 

In order to "reset" this you have to write the data to the drive again. Since the problem doesn't occur until the drive does GC. Writing new data to the drive is writing new data to the drive... It doesn't matter if you use a defrag utility or if you do a secure erase and write everything back again. You're still utilizing write cycles to get the job done.

 

Defraging and SSD isn't going to magically kill it. It is simply discouraged because under normal circumstances it won't lead to any performance improvement and would lead to wasted write cycles. This is not a normal circumstance and this case would benefit from a defrag...

 

ATA Secure Erase is not the same as writing zeroes.

 

ATA Secure erase issues a voltage spike marking all cells as writeable and (if the drive has this feature) throws out the current encryption key.  That is why a secure erase on a SSD only take about 10-15 seconds.  It does not use up write cycles.

 

Also, how would a defrag work?  These SSDs have wear leveling.

 

[+LogicalApex MVC]

ATA Secure Erase is not the same as writing zeroes.

 

ATA Secure erase issues a voltage spike marking all cells as writeable and (if the drive has this feature) throws out the current encryption key.  That is why a secure erase on a SSD only take about 10-15 seconds.  It does not use up write cycles.

 

Also, how would a defrag work?  These SSDs have wear leveling.

 

If you're doing a secure erase and then restoring an image backup I believe it is safe to say you'll need to use write cycles. You can't restore your backup image without writing to the disk...

[+Zlip792 MVC]

Firmware Update to Fix the Samsung SSD 840 EVO Read Performance Bug Coming on October 15th

 

 

A week ago Samsung acknowledged the existence of the read performance bug in the SSD 840 EVO and I just received a note that the fixed firmware is in validation process and is expected to be released to the public on October 15th. Unfortunately I don't have any further details about the bug or the fix at this point, or whether the update is coming to the 'vanilla' SSD 840 and OEM models, but I hope to get more details as the public release gets closer, so stay tuned.

 

Link: http://www.anandtech.com/show/8570/firmware-update-to-fix-the-samsung-ssd-840-evo-read-performance-bug-coming-on-october-15th

[JJ_]

Too bad I've returned mine

[Barney T. Administrators]

Seagate used to have these issues and developed a bad reputation because of this. Sorry to see Samsung is suffering from this too.

[aberg]

I have almost the same degradation on my Intel 520 180 GB. After two years it's about half the speed it had when it was new.

[guitmz]

got this ssd yesterday (evo 840). glad to hear that a fix is ready, its still in the box since my new laptop has not arrived yet

[farmeunit]

Because as I have mentioned, it is POINTLESS write cycles.  SSDs have NO need for this, there is absolutely NO need to defrag an SSD.  Ever.  They are not 7200rpm drives.

 

Apparently you didn't didn't actually READ my post.