Samsung Promises Another Fix for 840 EVO Slow Down Issue

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
Apparently Samsung is working on another fix for the slow down issue with its 840 EVO SSDs. Yes, this is a repeat of last October when the company originally "fixed" this problem. It's like déjà vu all over again.

So it appears that Samsung is still looking into the issue, but will update their Magician software to periodically refresh stale data until they can work out a more permanent fix that would correct all affected 840 EVOs. We have not heard anything about the other TLC models which have been reported to see this same sort of slow down, but we will keep you posted as this situation develops further.
Click to expand...
 
Hopefully they'll actually fix it this time, mine's starting to slow down again. Averages ~200Meg/sec overall when I did a read pass a week or so ago.

May have to refresh the surface manually though I forget what tool I used last time (before the first fix)
 
Interesting, I have three of these drives. Was thinking that maybe I need to bite the bullet and reformat. Will have to try the diskfresh and see if it makes a difference. WAs not aware the Evo had issues till now.
 
Chas said:
Okay, your solution to a slowdown was to run a tool that prematurely ages your drive?

Okay!
Click to expand...

Yeah , I just took a look at magician and it updated from 4.3 to 4.5. had to reboot a couple times. According to that everything is running well, at least on this machine. After rereading the OP it appears Magician fixes the issue, I just haven't run it in some time. Will steer clear of the other program for now as it does appear to be oriented towards mechanical drives. I might test it out in my mechanical drives though.
 
Chas said:
Okay, your solution to a slowdown was to run a tool that prematurely ages your drive?

Okay!
Click to expand...

In that vein I recommend setting your 840 EVO on fire, then you'll never have any problems with it ever again.
 
First of all the mentioned software explicitly says to use read only mode on flash drives and ssds. Secondly do the original 840 non evos have any issues? I have a 120 and 480 never had any problems on those. I also have never seen any slowdown on my msata 500 or 250 Evos. I there chance of data corruption or loss?
 
Hmm.

I have several Samsung SSD's and have been very happy with them (much more so than with the OCZ units I typically used before). For my own use I have picked up the Pro models though, as I thought I might write too much for TLC.

I did stick a 128GB 840 Evo in my mother-in-law to be's laptop when I refreshed it in late 2013. Hadn't heard about the slowdown issue, and she hasn't mentioned anything, but I'm not sure she does anything heavy enough to notice it slowing down.

Will have to take a look at it next time she is over.
 
I ordered a 250GB 840 Evo mSATA for a laptop build for a client a few weeks ago, the thing was dogshit slow right out of the box with the latest firmware. I would have bet money that it had a 5400RPM drive in it if I didn't know better.
 
Hopefully they'll get it right with this next go-around. If not, then the fix will have been confirmed to be the 850 Evo...
 
Maybe they can fix it, maybe not. Could be that TLC can't be used reliably in the long run. This tears it for me no more TLC ssds. Anyway most MLC ssds are cheaper than Samsung. I'm going back to Intel.
 
So, if I understand this problem correctly, it has to do with NAND state decay, a normal issue with any NAND, which TLC is much more sensitive to due to the larger amount of states per cell.

So, you write data, and the state slowly decays over time, until a certain point when it confuses the drive firmware, causing aggressive read retries, and thus performance slowdowns. Rewriting the data returns the state to its original state, and temporarily fixes it again, until it degrades once more.

The original fix updated the firmware to address how it reads degraded states initially eliminating the slowdowns, but they resurfaced as the states decayed further.

So, it seems as if Samsung either underestimated how much decay there would be in their TLC configurations, or overestimated their firmwares ability to deal with it, or both.

The main question I would have about this if I had one of these drives is as follows:

We are seeing slowdowns due to NAND state decay causing aggressive retries on reads to get data. Will there come a point when the NAND state decay causes data to no longer be recoverable, instead of just slowing down reads due to retries?


From the fix perspective it would seem Samsung has a few options:

1.) Further improve how firmware deals with NAND State decay, to recover decayed states better without slowdowns or data loss.

2.) include a NAND state decay monitor in the idle garbage collection pass which can tell when a cell is close to having decayed far enough to be a problem, rewriting that cell.


#1 may run into limitations of what is physically possible.

#2 may reduce lifespan by using up write cycles.


I would also be interested in knowing if this problem is expected to affect their 850 EVO drives as well, or if their 3D NAND technology is more resistant to decay

It will be interesting to see how this all plays out.

I have been much more trusting of Samsung SSD's than I have of many other brands, but they have been doing some pretty bleeding edge type stuff with TLC and 3D NAND and other things, and when you are bleeding edge, there are always risks.
 
zalazin said:
Maybe they can fix it, maybe not. Could be that TLC can't be used reliably in the long run. This tears it for me no more TLC ssds. Anyway most MLC ssds are cheaper than Samsung. I'm going back to Intel.
Click to expand...

Yeah, I would hold off on buying TLC drives for now until this issue is resolved (or not).

I'm throwing TLC out quite yet, but I'll be looking to see how this goes before I buy one.
 
Honestly, I don't think it is going to happen, not without significantly affecting the life of the drive. After finding out what the source of the slowdown is, It would require rewriting the old data on the drive just before it gets old enough to get slowed down. That's trouble.

The trouble is with the TLC technology itself. It loses charge over time, making it slower and more difficult to retrieve. In order to keep it from slowing down, the data must be rewritten to the cell, and that reduces the life of the drive.

I'm avoiding TLC drives from now on. I have a 500GB EVO and two 120GB versions. I use the 120s for boot drives on servers, so I won't be bothered by those slowdowns. The 500GB is in my backup laptop, which spends a lot of time powered down, at times up to two months. That will be significantly affected by this. No, no more TLC drives here.
 
dgingeri said:
Honestly, I don't think it is going to happen, not without significantly affecting the life of the drive. After finding out what the source of the slowdown is, It would require rewriting the old data on the drive just before it gets old enough to get slowed down. That's trouble.

The trouble is with the TLC technology itself. It loses charge over time, making it slower and more difficult to retrieve. In order to keep it from slowing down, the data must be rewritten to the cell, and that reduces the life of the drive.

I'm avoiding TLC drives from now on. I have a 500GB EVO and two 120GB versions. I use the 120s for boot drives on servers, so I won't be bothered by those slowdowns. The 500GB is in my backup laptop, which spends a lot of time powered down, at times up to two months. That will be significantly affected by this. No, no more TLC drives here.
Click to expand...

I agree that it would degrade the life of the drive, what I don't think has been proven out yet is if that degradation of life is enough to be significant.

It will depend on how often the data needs to be rewritten.

From what I have read on Anandtech, assuming 3x write amplification, the 840 EVO's have enough write endurance to completely rewrite every bit of data on them 333 times over their life.

Now assume that a garbage collection type routine can rewrite the cells in an orderly optimized fashion as opposed to client writes, and this would probably almost completely remove the write amplification. So we are talking more like being able to rewrite every cell on the drive 1000 times.

Now, here's what I don't know. How long does it take for the state to decay to the point where problems start to appear. 2 Months? 3 Months? 6 months? Lets assume that every cell with data on it needs to be rewritten once every 3 months. That's 40 times if we assume a drive has a 10 year life.

In other words, this just used up 4% of the write cycles, and that assumes the drive is full the entire time. (you wouldn't have to rewrite empty cells.

If the drive is usually half full, then that figure goes down to 2%, etc. etc.

So there are some guesses above, but the math seems to suggest that it is a small percentage of the cycles that get used up, IF a routine is properly optimized and executed regularly. Something that just goes through and rewrites the entire drive would be more problematic.

And if this is an issue on the 850 EVO, keep in mind, it has MUCH higher write cycles than the 840 EVO.

So it might be a problem, but I don't think it is a HUGE problem, if they get their acts together.
 
Zarathustra[H];1041444623 said:
I agree that it would degrade the life of the drive, what I don't think has been proven out yet is if that degradation of life is enough to be significant.

It will depend on how often the data needs to be rewritten.

From what I have read on Anandtech, assuming 3x write amplification, the 840 EVO's have enough write endurance to completely rewrite every bit of data on them 333 times over their life.

Now assume that a garbage collection type routine can rewrite the cells in an orderly optimized fashion as opposed to client writes, and this would probably almost completely remove the write amplification. So we are talking more like being able to rewrite every cell on the drive 1000 times.

Now, here's what I don't know. How long does it take for the state to decay to the point where problems start to appear. 2 Months? 3 Months? 6 months? Lets assume that every cell with data on it needs to be rewritten once every 3 months. That's 40 times if we assume a drive has a 10 year life.

In other words, this just used up 4% of the write cycles, and that assumes the drive is full the entire time. (you wouldn't have to rewrite empty cells.

If the drive is usually half full, then that figure goes down to 2%, etc. etc.

So there are some guesses above, but the math seems to suggest that it is a small percentage of the cycles that get used up, IF a routine is properly optimized and executed regularly. Something that just goes through and rewrites the entire drive would be more problematic.

And if this is an issue on the 850 EVO, keep in mind, it has MUCH higher write cycles than the 840 EVO.

So it might be a problem, but I don't think it is a HUGE problem, if they get their acts together.
Click to expand...

Actually, considering the firmware fix, that updated the code and rewrote the data came out in October, and we are now in February talking about how the slowdowns are returning, it would seem that 4 months is approximately the time it takes for written cells to degrade sufficiently for there to be issues.

So, doing the math again (assuming no write amplification, as opposed to regular use scenarios, the drive can schedule the rewrites for optimal times and groupings of cells.

This means three rewrites per cell with data on it per year.

30 rewrites over an assumed 10 year lifespan.

We just used up 3% of the drive's write cycles if it is full that entire time. 1.5% if it is half full, etc.

Also keep in mind that just by using the drive many cells will be wiped and rewritten on their own, so the additional write cycles used over what the drive would already be using to write its normal day to day data (assumed at 10GB average per day for a heavier than average user), so the actual figures for consumed write cycles will likely be lower than the 3% for a full drive.

Is this an issue? Definitely!

Does it mean current 840 EVO drives are doomed? Probably not, unless Samsung REALLY screws up.

Does it mean the end to TLC forever? Definitely not. the 3D NAND of the 850 EVO already has double the write endurance, which halves our percentage of the drive write cycles to 1.5% for a full drive right off the bat. Also judging by how the issue hasn't popped up as massively yet on these drives, it would seem NAND decay is slower on 3D NAND, probably not needing rewrites every 4 months like with the 840 EVO, so this number probably shrinks even further.

This is a bump in the road resulting from Samsung learning from its bleeding edge R&D development.

I feel like the compelling cost argument in favor of TLC will drive it to be used more and more as kinks like these are fixed.

I wouldn't buy a TLC drive right now though, until Samsung has addressed this satisfactorily, but I also wouldn't worry TOO much if I were currently using one.
 
Just checked my drivers performance in Samsung Magician, I can confirm my drive is "slower" like it was before I applied the last fix. The biggest hit is Random Write speeds, everything else is just marginally slower. Though I haven't done a fresh reinstall since Windows 8.1 came out, I'm patiently trying to wait til Windows 10!
 
Chunder said:
Just checked my drivers performance in Samsung Magician, I can confirm my drive is "slower" like it was before I applied the last fix. The biggest hit is Random Write speeds, everything else is just marginally slower. Though I haven't done a fresh reinstall since Windows 8.1 came out, I'm patiently trying to wait til Windows 10!
Click to expand...

This particular problem is a read performance issue only, nothing to do with writes.

Slower writes typically indicate an issue either with TRIM or with garbage collection or both.
 
Good thing I coughed up the extra cash for the Crucial M550.

This issue is why.
 
Hmm, I was thinking about getting one of these but went with the Intel 480GB offering instead. Smells like vindication.
 
No issue seen with my 840 EVO. Reads and writes are both in 450-500 range.
 
Well I just moved an older Intel 180 gb ssd into one of my I7 Evo 500 laptops and bought an Intel 240 530 series for my New Asus to replace the 250 evo. I have made the appropriate sacrifices to the technology Gods. Samsung will now be able to fix the TLC issue!!!!
 
Chas said:
Okay, your solution to a slowdown was to run a tool that prematurely ages your drive?

Okay!
Click to expand...

I cant tell others how to run their computers.

But I would guess, that even if I did this once every month, the drive would still last many many years, and would be replaced due to obsolence before it failed.

I don't plan to do this every month, probably a couple of times a year at most, and I also expect Samsung to fix this, so that this is a temporary fix.
 
You must log in or register to reply here.
Back
Top