Assistance - Calling all Samsung EVO owners

There are several theories, all feasible and not related to a physical fault. Only samsung can tell us exactly why this occurs and what the fix is though. Hopefully they will shed some light on it when they are ready, I'd be interested in the details. If the rumours are true we will have at least a firmware update middle of Oct.

In the mean time I'll be running a defreg pass whenever it gets noticably slower because it's a drop in the ocean compared to total available write cycles. I think a firmware update will come before it's necessary again though.

One positive to note, this could be a lot worse - there has been no data loss reported or dead drives.
 
wazza300 said:
Why does a secure erase cure it then? If it was a tlc fault it wouldnt?
Click to expand...

If the fault is the TLC NAND itself (I hope it's not, as I`ve always preffered Samsung SSDs over others), then it is quite logical that a secure erase will temporarily solve the problem.

We know that data that's written to the SSD will suffer from a drop in read speed after a few weeks/months (my Evo is now starting to show this). When you perform a secure erase/restore, all the data that's on the SSD will be "new". Leave it there for a few weeks and it starts to slow.

If it is a NAND fault, and it can't be effectively fixed with firmware, Samsung could have a bit of a problem !
 
pc-guy said:
hang on guys...the problem lies in the TLC . how is a firmware update going to fix it? it is a logic gate level fault no software can correct that. you may be able to correlate the degradation over time for compensation, but it will get to a point where ECC cannot fix the problem and therefore you will be left with a completely useless drive...unless the controller programming is changed to periodically move the files...but then you are effectively reducing the life span of the drive...as well as needing the drive to be plugged in and powered up all the time for that to happen. so what if i am using it as a backup or storage...then i am still going to be screwed.

if it is controller level error i.e. garbage control or whatever the term is...then i guess there is a fix. but that shouldn't affect the dramatic speed drop off seen on wider community especially there are tests done to the drives which effectively rewrites the entire drives over and over to test longevity, and it is proven that the speed do not drop off that quickly. or rather nominal drop offs.

this symptom seems to be related to age of file and therefore can only be attributed to a design flaw in the nand gates...

I would be very curious as to what the update will do. personally i was looking forward to getting one of these as the price is very attractive, but I certainly won't bother now.
Click to expand...

I assume you have no source for your assumption that the problem lies with the TLC NAND and is not fixable?
 
Dervious said:
If the fault is the TLC NAND itself (I hope it's not, as I`ve always preffered Samsung SSDs over others), then it is quite logical that a secure erase will temporarily solve the problem.

We know that data that's written to the SSD will suffer from a drop in read speed after a few weeks/months (my Evo is now starting to show this). When you perform a secure erase/restore, all the data that's on the SSD will be "new". Leave it there for a few weeks and it starts to slow.

If it is a NAND fault, and it can't be effectively fixed with firmware, Samsung could have a bit of a problem !
Click to expand...
Has anyone tried a regular ole format instead of a secure erase?

If the theory is that the TLC is degrading over time, then a format and re-copy of the data should yield exactly the same result as a secure erase, i.e. every single bit of data on the drive has to be re-written, hence refreshed, hence immune from slow performance for x weeks.

Has anyone actually tried it?
 
pc-guy said:
hang on guys...the problem lies in the TLC . how is a firmware update going to fix it? it is a logic gate level fault no software can correct that. you may be able to correlate the degradation over time for compensation, but it will get to a point where ECC cannot fix the problem and therefore you will be left with a completely useless drive...unless the controller programming is changed to periodically move the files...but then you are effectively reducing the life span of the drive...as well as needing the drive to be plugged in and powered up all the time for that to happen. so what if i am using it as a backup or storage...then i am still going to be screwed.

if it is controller level error i.e. garbage control or whatever the term is...then i guess there is a fix. but that shouldn't affect the dramatic speed drop off seen on wider community especially there are tests done to the drives which effectively rewrites the entire drives over and over to test longevity, and it is proven that the speed do not drop off that quickly. or rather nominal drop offs.

this symptom seems to be related to age of file and therefore can only be attributed to a design flaw in the nand gates...

I would be very curious as to what the update will do. personally i was looking forward to getting one of these as the price is very attractive, but I certainly won't bother now.
Click to expand...
I tend to be with you on this. I keep trying to think of scenarios where performance could drop over a period of time on untouched data, and everything points to cell degradation, hence causing the drive to recalculate ECC on the fly, which is CPU intensive, resulting in slow performance.
 
wazza300 said:
Why does a secure erase cure it then? If it was a tlc fault it wouldnt?
Click to expand...
Au contraire... the secure erase simply means that you have to re-copy all the data, which refreshes all the cells.

My theory is that a simple format (without the secure erase) will produce the same result.
 
Dave2150 said:
I assume you have no source for your assumption that the problem lies with the TLC NAND and is not fixable?
Click to expand...

well for the sake of millions who have these drive ,hope it is not TLC gate issue. But that is not only my own opinion, many on the original overlcok.net as well as anandtech believe it is the fundamental design flaw in the TLC gates.

the evidence and the test people have carried out stacks against any controller errors. controller or software level error can be easily tested out by torture tests which a lot of benchmarks have done when these drives came out as well as home users. the fact that the drive has been stable and still is stable and no dead drives proves the controller and software is fine. when you have a drive thats 9month old and read data is slowing down to 50mbs and it hasn't been used or even plugged in for 9months then that would suggest it is hardware level issue.

Simply put, plenty data out there if you want to put a logical set of reasoning behind it. Not trying to scaremongering just trying to point out the white elephant in the room here!

i also recall - maybe anandtech, there was discussion of the these TLC gate's stability and longevity over the MLC...when samsung first produced the 840s
 
Last edited:
rotor said:
Au contraire... the secure erase simply means that you have to re-copy all the data, which refreshes all the cells.

My theory is that a simple format (without the secure erase) will produce the same result.
Click to expand...

I'll test this theory.

Shall I do a quick or full format ?
 
It there is a hardware issue with the TLC memory, Samsung may well "fix" the problem by adding an extra process that re-writes data that has been on the SSD for a while. How they will manage this, I'm not sure. Maybe the controller can be programmed to detect when memory cells are starting to slow, then the data can be "refreshed".

There are a couple of downsides to this solution. Obviously it will cause more wear to the memory cells, so might shorten the life of the SSD. The other issue might be that it could potentially affect the performance of the SSD while this process is being carried out (unless it can be configured to only run when the SSD is "idle").
 
Dervious said:
It there is a hardware issue with the TLC memory, Samsung may well "fix" the problem by adding an extra process that re-writes data that has been on the SSD for a while. How they will manage this, I'm not sure. Maybe the controller can be programmed to detect when memory cells are starting to slow, then the data can be "refreshed".

There are a couple of downsides to this solution. Obviously it will cause more wear to the memory cells, so might shorten the life of the SSD. The other issue might be that it could potentially affect the performance of the SSD while this process is being carried out (unless it can be configured to only run when the SSD is "idle").
Click to expand...
And if a drive is powered off for longer than the life of the cells...? You lose all your data?
 
If this really is a problem with the durability of the TLC cells (just conjecture at this point), this is going to end up costing Samsung an enormous amount of money to fix. They are essentially going to have to replace every drive, and if data loss starts occurring on powered-off drives (e.g. go away on a long holiday, come back and your data is gone) there will be such an outcry... It's going to be interesting to watch.

At this point, my advice would be wait for the promised firmware update, and make sure your data is backed up!
 
Dervious said:
It there is a hardware issue with the TLC memory, Samsung may well "fix" the problem by adding an extra process that re-writes data that has been on the SSD for a while. How they will manage this, I'm not sure. Maybe the controller can be programmed to detect when memory cells are starting to slow, then the data can be "refreshed".

There are a couple of downsides to this solution. Obviously it will cause more wear to the memory cells, so might shorten the life of the SSD. The other issue might be that it could potentially affect the performance of the SSD while this process is being carried out (unless it can be configured to only run when the SSD is "idle").
Click to expand...

i would hate to have this process running on my storage drive as system resources will be constantly utilised as well as significantly reducing the drive's life span...anyone know if the 850 is using the TLC gates?

edit: the 850 is also using the TLC >.<
 
Last edited:
rotor said:
Quick would achieve the same result, which is to re-copy all the data back on to the drive. Does the drive currently have slow performance?
Click to expand...

I secure erased nearly a month ago, and the benchmarks are showing signs of reduced performance.



Shortly after secure erase/restore......



I'll image the drive, quick format, restore, then post the results.
 
pc-guy said:
i would hate to have this process running on my storage drive as system resources will be constantly utilised as well as significantly reducing the drive's life span...anyone know if the 850 is using the TLC gates?

edit: the 850 is also using the TLC >.<
Click to expand...


It wouldn't be ideal, but all we can do is wait to see what Samsung come up with. They designed these SSDs, so they should be able to identify and correct the issue. Hopefully they'll also shed some light on the cause and the solution. Only then can we decide if the 840 Evos are, or have been a good product.
 
It's obvious a secure erase isn't needed because a defrag acheives the same thing. And of course manually copying the slow files off and back onto the drive also works.

The firmware already has built in routines to detect cell degredation, if that routine is over cautious it will result in excessive error correction to kick in and result in slow reads. Tweaking the cell degredation detection parameters would fix it.

It could be rapid cell degredation causing this but there are so many other firmware related possibilities the odds are against it being that. Also there has been no data loss, even on drives with data several months old that hasn't been powered on. If it was cell degredation you'd expect to see data loss by then considering speeds are dropping of after just a few weeks.
 
You must log in or register to reply here.
Back
Top Bottom