• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Techreport's SSD endurance trial is over

mikeymikec said:
it's interesting to see which of their drives have lasted the longest.
Click to expand...
from the Anandtech review of 850 Pro
In fact, Samsung said that they have a 128GB 850 Pro in their internal testing with over eight petabytes (that is 8,000TB) of writes and the drive still keeps going, so I tip my hat to the person who is able to wear out an 850 Pro in a client environment during my lifetime.
Click to expand...
 
coercitiv said:
from the Anandtech review of 850 Pro
Click to expand...

However, Samsung saying that their own products can do something impressive is not the same as a (hopefully) independent source testing for it.

Furthermore, doesn't Samsung say something like that their warranty only covers something like 100TB of writes? I believe I read that in a review.
 
mikeymikec said:
However, Samsung saying that their own products can do something impressive is not the same as a (hopefully) independent source testing for it.

Furthermore, doesn't Samsung say something like that their warranty only covers something like 100TB of writes? I believe I read that in a review.
Click to expand...
The number of TB written before the SSD dies is not the same as the number of TB written at which data retention is still at least 1 year at room temperature.
 
mikeymikec said:
However, Samsung saying that their own products can do something impressive is not the same as a (hopefully) independent source testing for it.

Furthermore, doesn't Samsung say something like that their warranty only covers something like 100TB of writes? I believe I read that in a review.
Click to expand...

Yeah, but based on techreport's results from the 840 pro, I don't think Samsung's claim is flatly absurd.

The warranty exists to 1) It guarantees retention for 1 year at that write point 2) prevent enterprise users from grabbing consumer drives and exploiting a time-based (X-year) warranty with extreme workloads.

Edit: Also I already posted a thread about this here 🙂
http://forums.anandtech.com/showthread.php?t=2424358
 
I didn't say it was absurd either, just that it was a "manufacturer's claim", which I normally file under the same category as printer manufacturers' claims regarding how many pages per minute their printer can print 🙂

Yeah, you posted another thread. AFTER MINE. Then yours got all the attention.

!!!

🙂
 
What I don't understand about these results is, why is the data unrecoverable at the end? If the issue is just that the flash memory has worn out such that new writes are not possible, why can't the drive operate in read only mode?
 
GreenMeters said:
What I don't understand about these results is, why is the data unrecoverable at the end? If the issue is just that the flash memory has worn out such that new writes are not possible, why can't the drive operate in read only mode?
Click to expand...


When the flash memory gets worn out they lost the ability to retain data.
 
mikeymikec said:
I didn't say it was absurd either, just that it was a "manufacturer's claim", which I normally file under the same category as printer manufacturers' claims regarding how many pages per minute their printer can print 🙂

Yeah, you posted another thread. AFTER MINE. Then yours got all the attention.

!!!

🙂
Click to expand...



Deleted.
 
Last edited:
Hulk said:
Samsung "brags" about 8PB. Real world testing shows 2.4PB or 30% of that.

I'd say that while 2.4PB is certainly impressive, their claim of 8PB is absurd. 30% of the claimed "internal lab" testing. Yes absurd. Sure, there is sample variation but not that much.
Click to expand...

Uhm, the SSD that TR tested was the 840 Pro. The one that Samsung claimed survived 8PB was an 850 Pro. Different NAND.
 
Hulk said:
Samsung "brags" about 8PB. Real world testing shows 2.4PB or 30% of that.

I'd say that while 2.4PB is certainly impressive, their claim of 8PB is absurd. 30% of the claimed "internal lab" testing. Yes absurd. Sure, there is sample variation but not that much.
Click to expand...

840 PRO (in endurance trial) versus 850 PRO (Samsung's claim). The latter has a 10-year warranty, the former has 5 years. I wouldn't be surprised if the 850 PRO can do quite a bit more than the 840 PRO.

- edit - aaaah VirtualLarry said most of what I wanted to say already, dang 🙂
 
WhoBeDaPlaya said:
Doesn't matter how lasting the flash cells are.
One controller oopsie-daisy, and the SSD is SNAFU.
Click to expand...

Yes, but it's good to know that in reality they're not likely to have a low cell lifetime. Even the SSDs in that test with the shortest cell lifetime would last over 30 years according to my current usage, and I haven't configured my system with any write-avoidance tactics in mind.

If a similar test was done with HDDs, I would be more likely to go with the longer-lasting ones even with my experiences completely in mind of seeing disks die within a few months of installation.
 
Dumb question..

If I write big files e.g. movies to the drive, fill it, then delete entirely, fill it again and so on, is it the same thing they did? Is the drive going to fail after the said number of writes? I don't think that would take 18 months to do. That way a 128GB Samsung drive will take only 781 write cycles (for 100 TB) if I'm right. Anybody clarify my doubt?
 
rituraj said:
Dumb question..

If I write big files e.g. movies to the drive, fill it, then delete entirely, fill it again and so on, is it the same thing they did?
Click to expand...
Large sequential writes have a generally low write amplification, which means that the drive should last more; in theory at least, since they don't mention how many P/E cycles have been performed or what was the overall write amplification, so it's up to anybody's guess.

This is why I've often written that these kind of endurance tests are flawed and possibly misleading: the amount of data one can write for a certain number of program/erase cycles strongly depends on the workload type and testing conditions (random, sequential, continuous, is trim enabled, how much free space or overprovisioning space is there, etc) and can vary significantly. Luck counts too for SSD lifetime beyond the rated specification: the NAND in your SSD could be from a bad batch and last less than the average, or you might have some from a good batch, which could last more.

Then there's the data retention issue. When these drives fail after so many writes it's because essentially data retention has become so short that data cannot be reliably stored anymore, not even during active usage.

EDIT to clarify:

Is the drive going to fail after the said number of writes? I don't think that would take 18 months to do. That way a 128GB Samsung drive will take only 781 write cycles (for 100 TB) if I'm right. Anybody clarify my doubt?
Click to expand...

I don't know where you took this data point from, but generally speaking the drive won't fail at 100TB if the P/E rating hasn't been exceeded, unless the SSD is defective. The max P/E rating itself isn't a failure threshold, but rather a safe limit telling that by exceeding it, the NAND might not be able to meet the 1 year of power-off data retention JEDEC specification anymore (among other parameters), which means that in practice, as these endurance tests show, you will be able to use it longer than that (how much longer exactly, depending on NAND quality, how good the SSD's ECC algorithms are, if the manufacturer has set fixed overusage limits, etc).
 
Last edited:
I think your supposition is basically correct, however where did you get 100TB from? That was the lowest figure for the lowest drive to start exhibiting bad sectors, not for it to fail.

"Is the drive going to fail after x writes" - assuming it doesn't fail due to some other factor, and that the NAND in your SSD is of the same quality as the ones in theirs.
 
rituraj said:
Dumb question..

If I write big files e.g. movies to the drive, fill it, then delete entirely, fill it again and so on, is it the same thing they did? Is the drive going to fail after the said number of writes? I don't think that would take 18 months to do. That way a 128GB Samsung drive will take only 781 write cycles (for 100 TB) if I'm right. Anybody clarify my doubt?
Click to expand...


But if you were doing this, you would buy a bigger SSD, no?
 
earlyfailures.gif


Clear evidence of flash wear appeared after 200TB of writes, when the Samsung 840 Series started logging reallocated sectors
Click to expand...

The graph in the original article shows the Samsung drive starting to show re-allocations from 100 TB. Anyways, my calculations were wrong considering 128GB, as they were all 250 GB drives. Also I probably should have said "started to fail". May be you are overestimating my knowledge of the subject 🙂

Thanks for the reply though..
 
I don't think sector reallocation is necessarily a bad thing, unless it's due to uncorrectable errors. There have been suggestions that to keep the write amplification as low as possible, static wear leveling on TLC-equipped Samsung 840 SSDs is performed very conservatively. Since those sectors kept increasing more or less linearly for quite a long time (it doesn't really look like they're the result of random failures occurring from time to time), it could be they're the result of this wear leveling strategy, which would result in some blocks getting significantly more worn than others, earlier. This would probably also have the advantage of making errors appear slowly rather than potentially all at once and possibly catastrophically. Other SSDs with errors suddenly quickly increasing might have a less conservative static wear leveling which would result in a more even wear among all blocks.

On certain SSDs like the SanDisk Extreme II/Pro among SMART parameters there's also the number of P/E cycles performed on the most worn block. On mine (which I have moved to a different PC than the one I'm using right now) last time I checked it was at 150+ even though the average number of P/E cycles performed on all blocks was about 45. The write amplification for that SSD was generally much higher than that of the Samsung 840s I also have, given the same usage.

TechReport's Samsung 840 250GB endured for more than 900 TB of writes, which assuming a write amplification of 1.05x, would imply almost 3800 P/E cycles performed on average on all NAND blocks of that SSD. This seems about in-line with results from other endurance tests performed by others.
 
Last edited:
As far as I understand, after a more in-depth look and performing quick calculations, their endurance test was tailored to cause a low write amplification. TRIM was presumably also enabled. More OP wouldn't have helped significantly, if even at all.

Besides, the number of P/E cycles one can perform in total is more or less independent from the overall write amplification factor (and thus the OP), and that's what I was mainly referring about. In other words, more OP can allow you to write more data for a given amount of P/E cycles (depending on the workload and if trim is enabled), but won't actually increase the maximum P/E cycle amount the SSD can endure.
 
SSBrain said:
I don't think sector reallocation is necessarily a bad thing, unless it's due to uncorrectable errors. There have been suggestions that to keep the write amplification as low as possible, static wear leveling on TLC-equipped Samsung 840 SSDs is performed very conservatively. Since those sectors kept increasing more or less linearly for quite a long time (it doesn't really look like they're the result of random failures occurring from time to time), it could be they're the result of this wear leveling strategy, which would result in some blocks getting significantly more worn than others, earlier. This would probably also have the advantage of making errors appear slowly rather than potentially all at once and possibly catastrophically. Other SSDs with errors suddenly quickly increasing might have a less conservative static wear leveling which would result in a more even wear among all blocks.

......

TechReport's Samsung 840 250GB endured for more than 900 TB of writes, which assuming a write amplification of 1.05x, would imply almost 3800 P/E cycles performed on average on all NAND blocks of that SSD. This seems about in-line with results from other endurance tests performed by others.
Click to expand...

There article said the Samsung 840 had uncorrectable errors at 300Tb, so its pretty much done well before the other models.

"The 840 Series didn't encounter actual problems until 300TB, when it failed a hash check during the setup for an unpowered data retention test. The drive went on to pass that test and continue writing, but it recorded a rash of uncorrectable errors around the same time"
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top