Samsung 850EVO owners... read speed okay?

[Hulk]

I'm just wondering if 850 EVO owners are experiencing any read speed issues with their drives?

 

[Hulk]

I was thinking there should be a few 850 EVO owners around here?

If so, how long have you had your drive? Any read speed problems noted on HD Tach?

I'm looking for an SSD for the Lenovo T450s I'm getting ready to purchase and while the MX100 had good performance and price, the 850 EVO is faster, and more importantly has lower power usage.

A couple of good reports would definitely help solidify my confidence in the new 3D TLC NAND from Samsung...

 

[hojnikb]

Given that 850evo uses 40nm class 3D nand, its highly unlikely, that it suffers from read degredation the same way 840 series does.

 

[Quad5Ny]

I have a 1TB 850 EVO but I've had it for less than a month.

 

[Hulk]

Given that 850evo uses 40nm class 3D nand, its highly unlikely, that it suffers from read degredation the same way 840 series does.

I agree. Also, after thinking about the fact that the 840 EVO seems to have really good endurance I am thinking the problem probably is with the controller programming.

http://techreport.com/review/26523/...eriment-casualties-on-the-way-to-a-petabyte/2

But it would be nice for a couple of people that have purchased the 850 EVO to let us know that thus far they are experiencing no read slowdown. If we just know how old the drive is and that the HD Tach report is okay then we have even more evidence that the 850 EVO will hold up.

If we get some positive reports then I have a feeling that people on the edge of going for one of these might jump in. I know I would.

 

[coercitiv]

I agree. Also, after thinking about the fact that the 840 EVO seems to have really good endurance I am thinking the problem probably is with the controller programming.

Actually we're talking about two different characteristics: one is maximum P\E cycles, the other one is data retention. The TLC cell might survive (almost) the same number of P/E cycles as a MLC cell on the same process geometry (say 19nm) but it may have drastically lowered data retention. However, seeing how the 250GB 840 drive died close to 1PB of writes, I would say the 300TB limit signaled by the wear leveling counter indicates a rather resilient drive.

Meanwhile, my PM851 drive patiently awaits a fix, which will come even later than the one for 840 EVO, if ever

 

[Hulk]

Actually we're talking about two different characteristics: one is maximum P\E cycles, the other one is data retention. The TLC cell might survive (almost) the same number of P/E cycles as a MLC cell on the same process geometry (say 19nm) but it may have drastically lowered data retention. However, seeing how the 250GB 840 drive died close to 1PB of writes, I would say the 300TB limit signaled by the wear leveling counter indicates a rather resilient drive.

Meanwhile, my PM851 drive patiently awaits a fix, which will come even later than the one for 840 EVO, if ever

Honestly I don't know what the relationship is, if any, between endurance and data retention. I was assuming that perhaps the fact that the 840 EVO did do a rather remarkable 1PB of writes this could be an indication of it's ability to retain data? Is there a relationship between the two? I'm thinking about it now and I can see how they may not be related. During endurance testing the drive is continually written to, and read, I suppose? Over and over. The written data is read rather quickly, and by that I mean months don't go by between writes and reads. So I suppose a drive with 1PB endurance could have rather poor data retention? Down the rabbit hole we go...

 

[coercitiv]

Honestly I don't know what the relationship is, if any, between endurance and data retention. I was assuming that perhaps the fact that the 840 EVO did do a rather remarkable 1PB of writes this could be an indication of it's ability to retain data? Is there a relationship between the two? I'm thinking about it now and I can see how they may not be related.

All we know for sure is if the manufacturer gives a consumer drive a TBW rating of say 70TB, that drive will have a power off data retention of at least 1 year at 30C after 70TB of host writes. Fun fact is simply storing the drive at 25C might double power off data retention. (see this doc, page 27)

Being able to withstand 1PB of host writes will surely indicate a rather conservative 70 TBW rating, but won't give any real info on data retention at say... 200TB.

During endurance testing the drive is continually written to, and read, I suppose? Over and over. The written data is read rather quickly, and by that I mean months don't go by between writes and reads.

They did test data retention at some intervals, but they only left the drives on the shelf for 1-2 weeks: enough to spot a really weakened drive, but not much more.

 

[Glaring_Mistake]

All we know for sure is if the manufacturer gives a consumer drive a TBW rating of say 70TB, that drive will have a power off data retention of at least 1 year at 30C after 70TB of host writes. Fun fact is simply storing the drive at 25C might double power off data retention. (see this doc, page 27)

Being able to withstand 1PB of host writes will surely indicate a rather conservative 70 TBW rating, but won't give any real info on data retention at say... 200TB.


They did test data retention at some intervals, but they only left the drives on the shelf for 1-2 weeks: enough to spot a really weakened drive, but not much more.

Actually the TBW rating is not related to that but is instead the limit for your warranty so if TBW is 70tb and you've written more than 70tb then it is your warranty that is expiring and not your SSD.

The data retention that you're talking about is related to when all write cycles have been consumed,which for the Samsung 840 and 840 EVO would be about one thousand write cycles.

So for an 120gb SSD of one of those it would run out of write cycles at about 120tb if write amplification is around 1.

And the 250gb at about 250tb and so on.

So TBW is usually a lot lower than an SSD can take before running out of write cycles.

 

[coercitiv]

So TBW is usually a lot lower than an SSD can take before running out of write cycles.

And where did I state otherwise?

 

[Glaring_Mistake]

And where did I state otherwise?

"All we know for sure is if the manufacturer gives a consumer drive a TBW rating of say 70TB, that drive will have a power off data retention of at least 1 year at 30C after 70TB of host writes."


The power off data retention after a year is something that it should be able to do according to JEDEC standards after all the write cycles it is rated for is consumed.

And not when it has reached it's TBW rating.

And in that sentence it looks like you are equating the two.

 

[SSBrain]

The confusing thing (for most people) is that on consumer SSDs the TBW rating is most often not related with actual NAND endurance. For that one should check out internal wear parameters (ie the amount of depleted P/E cycles). On those where the manufacturer's TBW rating actually has something to do with NAND endurance, it usually (although not necessarily so) refers to a workload of continuous random writes with the drive filled up with data, which increases significantly the write amplification. So, under normal (consumer, even) usage patterns it should be much higher than that.

The P/E cycle limit on consumer drives is specified according to the JEDEC requirement that the NAND's "end-life" is when the power-off data retention time at 30°C drops below 1 year. For Samsung 840/840 EVO it's 1000 P/E cycles. JEDEC also states that at 1/10 of the max P/E limit data retention should be 10 times longer, meaning >10 years.

People have been having stale data performance issues with Samsung 840/840 EVO with barely a few P/E cycles performed out of 1000. With so few write cycles performed data retention isn't even supposed to be an issue.

 

[coercitiv]

The power off data retention after a year is something that it should be able to do according to JEDEC standards after all the write cycles it is rated for is consumed.

And not when it has reached it's TBW rating.

And in that sentence it looks like you are equating the two.

From the JEDEC SSD Specifications doc I linked above, page 25:

The SSD manufacturer shall establish an endurance rating for an SSD that represents the maximum number of terabytes that may be written by a host to the SSD, using the workload specified for the application class, such that the following conditions are satisfied:
1) the SSD maintains its capacity;
2) the SSD maintains the required UBER for its application class;
3) the SSD meets the required functional failure requirement (FFR) for its application class; and
4) the SSD retains data with power off for the required time for its application class.
This rating is referred to as TBW. Requirements for UBER, FFR, and retention are defined for each application class.

Following on page 26:

SSD endurance classes and requirements
Application Class: Client
Active Use (power on): 40°C, 8hrs/day
Retention Use: 30°C, 1 year

 

[Hulk]

From the JEDEC SSD Specifications doc I linked above, page 25:
Following on page 26:

My initial post was misleading.

The JEDEC spec states 1 year data retention. But that doesn't really mean much if the read speed decreases by a significant amount rendering the drive virtually useless.

 

[Hulk]

I bought a 1Tb 850 PRO SSD on 2 February.

For 3 weeks it worked really well, then Samsung's Magician software kept nagging me that a firmware update was available. I did nothing for a couple of days, but on 23 February, I gave in and applied the firmware. I turned off my laptop, but I was unable to boot to this SSD again. My laptop's BIOS/UEFI was unable to "see" the drive.

I have since discovered a large number of other 850 PROs have been "destroyed" by this firmware update including Xeon_Addict of this forum.

We have 250 members of a FB page created last Tuesday to collate worldwide victims of the deadly firmware update hoping to pressurise Samsung into responding appropriately. We are Samsung SSD 850 PRO firmware death if anyone else wants to share their experiences.

When we had no response from Samsung after nearly a week, we wrote this open letter to Samsung but again we have not had a response.

My recommendation is to avoid Samsung like the plague!

Samsung had better do right to you guys and the 840 EVO owners. Because it's not just the owners of these devices who are going to avoid ALL Samsung products but it will be everybody who has ever heard of how these issues were handled by Samsung. Samsung should keep in mind that fortunes can change quite quickly in the tech industry and there is always another player ready to step up.

 

[SSBrain]

Why are people assuming it's a data retention problem, that of Samsung 840/840 EVO SSDs? Has anybody lost data yet because of this issue when their drive was turned off?

I'm more inclined to think that since JEDEC doesn't specify a minimum performance level that the drive has to maintain to comply with the requirements (as far as I know, at least), Samsung took advantage of that, meaning that the data will still be there in the [very] long term, but it will be slow to read for the same reasons other manufacturers have had for a long time with TLC NAND, likely overcame with modern advancements in ECC techniques.

I wonder if Samsung 840 EVOs which had this issue permanently corrected with the latest firmware bundled with the Performance Restoration Software were the most recently produced ones, in fact. If yes, it could mean that something was changed hardware-side.

 

[coercitiv]

Why are people assuming it's a data retention problem, that of Samsung 840/840 EVO SSDs?

I did not say it's a data retention problem, but I did say TLC drives may have lower data retention than MLC drives. I don't now how Hulk arrived to the conclusion that Samsung TLC drives are not JEDEC compliant, and I'm sorry if it was somehow my doing.

Anyway, trying to steer this thread back on topic, the main reason 850 Evo owners may not experience slowdowns on old data has to do with how much easier is to read&write data from TLC V-NAND.

Anandtech review

The key aspect of 3D V-NAND is the process node. By going back to 40nm lithography, the number of electrons increase exponentially, which makes TLC a much more viable technology than it was with modern planar NAND. Obviously, V-NAND doesn't change the basics of TLC NAND because it still takes eight voltage states to differentiate all the possible 3-bit outputs, but thanks to the increased number of electrons there is more breathing room between the states and thus the cells are more error tolerant.

​

Samsung claims 10x reduction in voltage state overlaps, which is a massive change for the better. You can see how crammed the planar TLC voltage states are, so it's no wonder that the endurance is low because the states are practically overlapping at each point in the voltage distribution and hence even tiny changes in the cell voltage can alter the cell's voltage state.

The larger cell structure also enables higher performance because it takes less iterations to program a cell. With planar TLC NAND it took multiple very high voltage pulses as well as numerous verification process to reach the right charge, but with looser voltage distribution the programming process has less steps and thus takes less time.

PS: please, let's not turn this thread into another Samsung is bad / firmware bug / horrible RMA thread. If we were to judge all manufacturers by their mistakes, I don't think any of them would emerge unscathed. Well, maybe Toshiba would.

 

[coercitiv]

The confusing thing (for most people) is that on consumer SSDs the TBW rating is most often not related with actual NAND endurance.

I think the real confusion came from not having a proper term for the number of P\E cycles that leads to NAND cell death. We have "rated P\E cycles" to express the threshold for data retention, but when I wrote "maximum P\E cycles" earlier in the thread I meant the number of cell erases that leads to cell failure.

 

[Hulk]

I did not say it's a data retention problem, but I did say TLC drives may have lower data retention than MLC drives. I don't now how Hulk arrived to the conclusion that Samsung TLC drives are not JEDEC compliant, and I'm sorry if it was somehow my doing.

Sorry for any additional confusion I brought to this issue. I edited that post.

My point was that the JEDEC spec doesn't have much teeth if the drive in question does in fact retain data after a year but the read rate of the data is so slow as to make the drive for all practical purposes unusable.

 

[Batmeat]

No issues here.

 

[Glaring_Mistake]

From the JEDEC SSD Specifications doc I linked above, page 25:
Following on page 26:

However the TBW rating given by the manufacturer is often a number that is a bit arbitrarily set.

I mean if we look at the Crucial M500,it has a TBW rating of 72tb,regardless of the size of the M500 you're looking at.
If they have determined that the 120gb version can endure that under specific conditions,then the ones that are twice,four and eight times larger should have correspondingly higher TBW ratings yet they do not.

And if we look at the Transcend SSD370 which uses the same NAND,their 128gb version has a TBW rating of 150tb which is more than twice than that of any M500.
And the 1tb version has a TBW rating of 1180tb or 16 times that of any of the M500s.
They accomplish this despite having less overprovisioning.

So the TBW rating is often inconsistent both within their own line of SSDs and when compared to the rating given to other SSDs using the same NAND.

 

[Glaring_Mistake]

I did not say it's a data retention problem, but I did say TLC drives may have lower data retention than MLC drives. I don't now how Hulk arrived to the conclusion that Samsung TLC drives are not JEDEC compliant, and I'm sorry if it was somehow my doing.

Anyway, trying to steer this thread back on topic, the main reason 850 Evo owners may not experience slowdowns on old data has to do with how much easier is to read&write data from TLC V-NAND.

Anandtech review
PS: please, let's not turn this thread into another Samsung is bad / firmware bug / horrible RMA thread. If we were to judge all manufacturers by their mistakes, I don't think any of them would emerge unscathed. Well, maybe Toshiba would.

I like to think that an SSD-manufacturer that hasn't made a mistake simply hasn't been making SSDs long enough.

And Toshiba?

Well it didn't take me long to find this: http://arstechnica.com/apple/2013/1...-fix-for-toshiba-drives-in-2012-macbook-airs/

 

[coercitiv]

And Toshiba?

Well it didn't take me long to find this: http://arstechnica.com/apple/2013/1...-fix-for-toshiba-drives-in-2012-macbook-airs/

maybe not even

 

[readymix]

I'm just wondering if 850 EVO owners are experiencing any read speed issues with their drives?

canceled