850 EVO life span question

[moonbogg]

I've only had this drive a few days and already is shows .62TB total writes. That's just from installing windows and transferring a few games. Total lifespan is said to be 150TB. That's like about 1.5 years until dead at this pace.
For some strange reason my Samsung 830 that ive had for years only shows 5.3TB total writes. How is that possible when I've already added up .62TB on the new drive in a few days?

 

[tweakboy]

Couple years , Hard drives can live past 60k hours on it. Im sure SSD goes even longer since theres no moving parts.. gl

 

[Jeff7181]

I don't think the first day of usage, especially considering the installation of Windows and your applications/games is indicative of normal usage.

I've and mine for about 6 months and it shows over 800 power on hours and 1.41TB written.

 

[coercitiv]

• 3D TLC NAND is rated at around 2000 P/E cycles.
• your SSD is a 1TB model
• 1TB x 2000 = 2000TB host writes

Even if we consider you write 0.5TB per day (which you will not, you'll write max 20GB per day in the next years) the SSD will wear out after 4000 days.

4000 days = 10+ years

Have fun wearing that SSD down, if you really want to get the job done I suggest electric shock or a hammer. (or some help from Sammy with a special sauce firmware)

 

[KingFatty]

Did you actually write 0.62 TB or do you think that's an error?

 

[Erithan13]

0.062TB (62GB) would be far more believable for the windows install + updates at least, how large are those games you transferred?

Edit: Loaded up magician on my own 850 installed last week, showing 0.49TB written, that's from windows + some small programs + Witcher 3. 0.62TB probably is reasonable after all.

 

[larryccf]

This might make it easier to relate to - i render video files for a hobby, and typically render 30 to 80 GB a day, 2-3 days a week to my 840 EVO i use as a "worktable" SSD - this hard drive is right at 2 years old, note the total TB written -

I generally upgrade my computer every 6 years - at this rate, i still won't be within sight of the 840's demise when i retire this computer

fwiw

 

[moonbogg]

• 3D TLC NAND is rated at around 2000 P/E cycles.
• your SSD is a 1TB model
• 1TB x 2000 = 2000TB host writes

Even if we consider you write 0.5TB per day (which you will not, you'll write max 20GB per day in the next years) the SSD will wear out after 4000 days.

4000 days = 10+ years

Have fun wearing that SSD down, if you really want to get the job done I suggest electric shock or a hammer. (or some help from Sammy with a special sauce firmware)

Interesting. The drive is said to have "5 Years Limited Warranty or 150 TBW Limited Warranty"

So, this means that at 150TB, the drive will explode...just kidding. If its anything like 2000TB like you say, then yeah, it will last for ever. I guess I am focusing on the warranty limit too much. Most products operate well beyond their warranty limit, so I think that's where my confusion must be.

As far as actually using .62TB, yes, I think I did. I was transferring 75gb folders back and forth trying to get a game to work. It didn't work, so I deleted it and transferred again, then dragged it around some more, closed my eyes and spammed the enter button hoping it would work, dragged it back again etc etc. So it makes sense from all the derping I did and messing with large game files.
So I guess I can record gameplay on it and not have to worry too much about wearing out by doing that.

 

[Puffnstuff]

I've also noticed that the writes reported by magician are exceptionally high on both of my pc's with Samsung boot drives. Right now the 850 pro in my desktop is showing 2.7tb of writes and it's only a few months old.

 

[moonbogg]

I've also noticed that the writes reported by magician are exceptionally high on both of my pc's with Samsung boot drives. Right now the 850 pro in my desktop is showing 2.7tb of writes and it's only a few months old.

Maybe magician falsely reports more writes so they get your drive out of warranty faster. Its always more fun if adding a conspiracy theory.

 

[Erithan13]

Is it definitely reporting terabytes and not say terabits by mistake/design? Even 0.49TB on my system seems absurdly high given I've only installed things, not been copying to and fro from the drive. I guess we'd need another SSD to compare what the expected writes would be given a windows + games/other programs install. Although at this point it's becoming rather academic since none of this seems to really matter to the drive lifespan or performance.

 

[Hellhammer]

I've also noticed that the writes reported by magician are exceptionally high on both of my pc's with Samsung boot drives. Right now the 850 pro in my desktop is showing 2.7tb of writes and it's only a few months old.

2.7TB isn't that much. Not sure what exactly is "a few months" but assuming it's about four months that would yield 22.5GB of writes per day, which isn't out of the ordinary for a heavier user.

Is it definitely reporting terabytes and not say terabits by mistake/design? Even 0.49TB on my system seems absurdly high given I've only installed things, not been copying to and fro from the drive. I guess we'd need another SSD to compare what the expected writes would be given a windows + games/other programs install. Although at this point it's becoming rather academic since none of this seems to really matter to the drive lifespan or performance.

Magician is reporting tebibytes (1024^4 bytes) to be exact. Again, 0.49TB is really not much - it translates to only 500GB.

 

[ShintaiDK]

My MX200 500GB 2260 SSD I got with my new Skylake build now got 1.62TB written. You always use more in the beginning when you essentially flood the drive with all your junk and stuff.

I got over 1000 power on hours for it already. It says 1% lifetime used (tho we know it can got multiple times over it). Even by that measure, it got 16-24 years left with current usage.

SSDs are good at reporting numbers, giving the "SSD hysterica" cases. People would dump HDs left and right in panic if they saw expected numbers there.

 

[JimmiG]

I've written 2.76 TB to my 500 GB 850 Evo since I installed it on October 5th.

By the time it has worn out, it will probably be a secondary SSD, while my main SSD might be an 8 TB Evo 1450.

 

[RecoveryForce]

I just finished a data recovery project from a 750GB EVO and have learned a lot about these drives in doing so.

The TLC NAND starts dropping bits within days or weeks of data being written to the sectors. So, after time, the drive starts to slow down when having to rely heavily on ECC to read those sectors. Given enough time a few months, those sectors become unrecoverable.

But, there is good news, they introduced a firmware fix that appears to force the drive to continuously re-write data back to the drive in order to avoid dropping the bits. However, it is a different story if the drive is powered off for any length of time.

The drive I received has been to several data recovery labs over the past few months who said that recovery was not possible. We, on the other hand, have been able to get by the broken firmware and clone the drive. Sadly, there is about 4% of the drive that is unreadable which caused 93% of the user's files to contain bad sectors. All unused sectors filled with zeros are 100% clean with no read errors.

My conclusions from this are:

1. These drives have virtually no shelf life, so don't archive data on them
2. If you have one that fails, get it to a data recovery lab that is capable of handling their recovery ASAP
3. Don't use these drives in mission critical systems
4. If using these drives, be sure to have the data backed up frequently

 

[coercitiv]

Given enough time a few months, those sectors become unrecoverable.

Are you sure? This piece of information has very serious implications.

 

[RecoveryForce]

Are you sure? This piece of information has very serious implications.

I've been conversing with a technician at Ace Labs (the makers of PC3000) and they have found this too. Apparently, they setup 8 EVOs 4 months in advance for data recovery training and when training day came, 3 of 8 lost the root catalog and files were corrupted. 2 months later, they were up to 5 of the 8 being virtually unrecoverable.

If I understand correctly, these were new drives that all had about 40GB of data dumped to them for training purposes, then set aside for training day. I have one on my shelf that I am going to setup today for a test of my own.

However, the case I just finished seems to have confirmed this theory nicely and it certainly explains why these drives were having so many slow down issues before the firmware "fix" that Samsung released.

If any of you have one of these drives to spare, it might be worth your testing the drive too. Let's all copy data to these drives, I'm thinking of filling mine about 50%, then see what happens. If the theory stands, after sitting a month or two, there will be bad sectors in the used data sectors and none in the unused portion of sectors.

 

[eton975]

Are you sure? This piece of information has very serious implications.

No it doesn't. The drive automatically remaps these sectors. Ever wonder why noone's paniced over usb stick/SD card failures even though they use the shittiest NAND on the market?

It's really nothing to worry about if you have backups.

 

[RecoveryForce]

No it doesn't. The drive automatically remaps these sectors. Ever wonder why noone's paniced over usb stick/SD card failures even though they use the shittiest NAND on the market?

It's really nothing to worry about if you have backups.

I'm not sure you fully understand what I'm seeing. A drive can't remap a sector when it is powered off. Let's say that you get sick and land yourself in the hospital for 2 months and your computer is powered on during that time. Don't you think it reasonable that your data on your computer should be in the exact same condition it was when you powered it down?

People every day copy data to a drive to store for archival reasons. They have full and reasonable expectations that the data will still be there in a year when they need to access that data again.

 

[moonbogg]

This drive is faster than S#!T. My PC restarts in just a few seconds and games load in a panicked ass-flash. I like it.

 

[eton975]

I'm not sure you fully understand what I'm seeing. A drive can't remap a sector when it is powered off. Let's say that you get sick and land yourself in the hospital for 2 months and your computer is powered on during that time. Don't you think it reasonable that your data on your computer should be in the exact same condition it was when you powered it down?

People every day copy data to a drive to store for archival reasons. They have full and reasonable expectations that the data will still be there in a year when they need to access that data again.

Sorry mate. Didn't read your whole post. Damn though...

 

[BSim500]

Are you sure? This piece of information has very serious implications.

It does make sense though:-

SLC: 2x voltage states = 0: (0v), 1: (100%) of whatever voltage it's programmed with

MLC : 4x voltage states = 0: (0v), 1: (33%), 2: (66%), 3: (100%)

TLC : 8x voltage states = 0: (0v), 1: (14%), 2: (28%), 3: (43%), 4: (57%), 5: (71%), 6: (86%), 7: (100%)

With small process nodes there are far fewer electrons holding each charge, so it "trickles out" quicker when unpowered. With TLC, a voltage drop of say 5% from state 6 would mean 81% instead of 86% of the charge is being held. If it's then powered on, the ECC could correct for it as it's obviously above 71%. However, SSD firmware can only correct voltage drift when it's powered on. If it's powered off for too long and the voltage fell by say 15% (86% to 71%), the "drift" will be into the next voltage down (and the data corrupted), at which point how can the SSD tell if "70-72%" is a "good 5" or a "badly drifted 6" state? Meanwhile, the same 15% drop is still readable on an MLC (66% to 51%) with a great deal more headroom to spare.

Everyone keeps talking about this chart whenever the subject comes up, however all that data is based purely on Intel's "accelerated testing". It's also obviously going to be radically different depending on technology (MLC vs TLC) and process node (16nm, 19nm, 40nm, etc). Given that it's Intel drives used in testing, I have a suspicion all that data applies to Intel's MLC drives, so those figures are obviously going to be lower for TLC which has less than half the headroom for unpowered "voltage drift" when used as mostly offline storage.

Despite the "promise" of 40nm 3D-NAND to have fixed things, I'm personally steering well clear of TLC drives from all brands. It's not as if Samsung's TLC EVO's are massively cheaper than Crucial's MLC drives, which is after all the only advantage TLC has (or is supposed to have...)

 

[Coup27]

Nice information RecoveryForce, although you should make clear in your posts that everything you have written refers to the 840 EVO not the 850 EVO which is in the thread title. The 850 EVO uses 3D TLC NAND and I believe because of the design of the NAND it does not suffer the bit rot symptoms you have found.

Also points 3 & 4 are a bit unfair. The drive is consumer class so you would never recommend a consumer class product for mission critical systems and regardless of the drive in question, you should always have backups.

I have used my 840 EVO for a couple of years now in my main machine and I have never ran into any issues relating to corrupt data but the most the drive was ever powered off for was 3 weeks one Christmas. If it does implode after being powered down for a couple of months then surely it doesn't meet the JEDEC requirement of 12 months of unpowered data retention? Not sure how Samsung have avoided a class action, and you may have just opened a rather large can of worms

 

[coercitiv]

If it does implode after being powered down for a couple of months then surely it doesn't meet the JEDEC requirement of 12 months of unpowered data retention?

This is what I meant by serious implications: if these drives cannot hold data for at least 12 months even before their P/E counter is depleted, things can go south rather fast.

If any of you have one of these drives to spare, it might be worth your testing the drive too. Let's all copy data to these drives, I'm thinking of filling mine about 50%, then see what happens. If the theory stands, after sitting a month or two, there will be bad sectors in the used data sectors and none in the unused portion of sectors.

I have a 256GB PM851, the OEM version of the 840 EVO.

These OEM drives were used inside many laptops sold in 2014-2105, some of which are still on store shelves. They come with OS installed, from factory, more than 6 months ago.

Now, there are many variables in play here (SSDs are brand new, temperature at which they are written might affect data longevity etc), but I can tell you the system I bought had read speed issues from the day it was opened. They were not visible to the naked eye, but parts of the SSD were read at speeds as low as 10-20MB/s. Overall disk speed rose at ~500MB/s after data refresh.

So consider this data point: even if there is a short threshold of time at which these drives lose data, that threshold must somehow be linked to other factors as well, otherwise stores would be full of DOA laptops just about... now

Here's a fragment from an older post on this matter, SSBrain had some very interesting insights to share from his tests with a Samsung 840 drive.

- Long term performance of old data on Samsung 840/840 EVO SSDs might be affected by the temperature of the drive at the time of writing.

- Firmware issues might be causing improper handling of old data written when the drive was at a low temperature.

 

[RecoveryForce]

Sorry...you are right. I didn't mean to hijack the 850 thread with 840 issues. However, it seems that it has sparked some good discussion.