SSD Write Endurance 25nm Vs 34nm

[B.A.T]

It looks like it. I think I'll do an retention test at 700 TiB to see whats going on.

It's the brother of the last Mtron I tried. GullLars had two i r0. It's a MTRON Pro SSD 32GB

[Christopher]

--------------------------------------------

Plextor M3P 128GB Day 4

96,000.12 GiB
93.75 TiB

274.34 MB/s Average

96 Hours

Rellocated Event Count 0


--------------------------------------------

Samsung 830 256GB Day 74

1,757,421.07 GiB
1,716.38 TiB

303.08 MB/s Average
1741 Hours

7632 Wear Leveling Count
MWI 1

6/0 Erase/Program Fail
Used Reserved Block Count: 12/24572 sectors

--------------------------------------------

The combined average speed for both drives is 577.42MB/s. That's a not inconsiderable amount of data going through the board every second.

[Christopher]

Wow, I was just thinking to myself that the 830 had gone awhile without another block fail, and lo! 45 seconds later, another went bad.

7 Erase Fails, 7 runtime total blocks used, 28672 Reallocated Sector Count

That's 14MiB if you're keeping score at home (out of 262,144 MiB on the drive).

[canthearu]

Intel 520 60GB - Day 98

Drive hours: 2,298
ASU GiB written: 729,146.54 GiB (712.06 TiB)
Avg MB/s: 93.07 MB/s
MD5: OK

Host GB written (F1): 734,281.25 GiB (717.07 TiB, 23497000 raw)
NAND writes (F9): 519,818 GiB (507.63 TiB)

Reallocated sectors (05): 0
Failure count (AB, AC): 0 program, 0 erase
Raw Error Rate (8B): 100 normalised
Avaliable Reserved Space (AA): 100 normalised
Media Wearout Indicator (E9): 1

On the topic of data retention .... it is time for a retention test on my Intel 520 ... It has exceeded 500TiB in NAND writes (approx 8k NAND cycles). Will be off and disconnected for 3 days.

[Anvil]

canthearu
Thanks for the update!
Could be an exciting 3 days

B.A.T
A few reboots per month might catch such issues earlier, I reboot once a month and often as a result of the Tuesday windows updates.
The G1 is really getting some beating, no TRIM no nothing, wouldn't surprise me at all if it start showing signs of weakness.
(or maybe it's the curse of the MTRON )

I'll post an update later tonight, the X25-V is very close to 1PB, 8-9 hours or so.

Endurance_cr_20120531.png

[kaktus1907]

Plextor released new firmware for M3s, 1.04.. not much info what it changes

http://www.goplextor.com/asia/index....gory&catid=147

EDIT: ok i think i found it..

[Kain665]

Guys, I know this would be a pain, but would it be possible to compile a chart of all the drives tested since the inception of this thread?

[Christopher]

Guys, I know this would be a pain, but would it be possible to compile a chart of all the drives tested since the inception of this thread?

There is a chart of the past drives in the first thread, right above a chart of the current drives.

[gordo]

Christopher, I have a Samsung 830 256GB and have been following this thread closely in recent days... Would it be safe to say that the MWIs for most of these drives are 1. arbitrarily chosen by the manufacturers and 2. extremely conservative based on these endurance tests? Also, how far do you think the 830 will go given the current low error count? 3 PBs? 10 or more PBs?

I had been using a 160GB Intel X-25M G2 since 2009 and calculated that on average I was writing roughly 6-7GB per day over the period of a couple years (I mostly use it for documents and web surfing and some light gaming, hibernation/hybrid sleep is off, it's on for about 8 hours out of every 24, with maybe 4 days off per month average.)

I expect usage for my Samsung 830 to be about the same. Assuming that I'll write 7GB per day, and that I can get roughly 2PB of writes out of this drive (which currently appears extremely likely), it would work out to ~820(!) years of usage.

Now, I get that for enterprise workloads like database work you might be doing a whole lot more writing, but even if you turned over the entire drive (238GB) every day, you'd still get around 24 years of use out of this drive, (again, assuming it craps out at 2PB.) I'd have to be writing at least a terabyte every single day to ever get close to thinking about this wearing out in a reasonable technology upgrade timeframe. My guess is that 99.9% of these drives will inevitably end up in a landfill due to obsolescence with only a small fraction of their writes ever being used.

[B.A.T]

Todays update:

m4
974.8412 TiB
4105 hours
Avg speed 74.58 MiB/s.
AD gone from 55 to 52
P/E 16810.
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
669.92 TiB
23416 hours
Reallocated sectors : 103 to 104
Available Reserved space: 78
MWI= 116 to 114
MD5 =OK
43.07 MiB/s on avg


Intel X25-E 64GB
165.15 TiB
504-30=474 hours
Reallocated sectors : 0
Available Reserved space: 100
MWI= 98
MD5 =OK
100.12 MiB/s on avg


I'll update you on the Mtron in 30 min. Need to check something first.

@Anvil
The last reboot was when I entered the E into the test-rig. Maybe the Mtron jinxed it Everything looks normal no so we'll just have to wait and see.

[B.A.T]

Mtron Pro 32GB SLC

Here are the smartvalues


The nand


The controller


And the spec
Link Link

I need some help to translate the raw value in total erase count.

[Anvil]

B.A.T

Record the raw value before and after writing a 1GB file.
(ASUs sequential [4MB seq] test will do that)

Two SLCs
I hope the MTron behaves this time

--

It's getting close.
2012-05-31-2324.PNG

(a bit busy downloading the new stuff from MS but will post an update when it gets past 1PB)

[Christopher]

Gordo,

The MWI for most drives is extremely conservative, but that helps take into account the fact that some drives will not make it as far as other drives (if you had two identical drives, under an identical workload, they probably wouldn't die at exactly the same time). The way PE cycle ratings are generated is extremely pessimistic, and a little arbitrary, but there are reasons they do it that way.

For a 830 only using 1 PE cycle a day (based on ~3600 PE cycles) it would take 9.8 years give or take to get to MWI 1. It's unrealistic to expect the drive to last that long though, as generally speaking, it's going to be something else that kills the drive.

I would hate to venture how far my 256GB 830 will go, I've pretty much been wrong on every prediction in the past. The 830 has been slowly marking blocks as bad, and that's a good thing. That means its doing what it's supposed to. What you don't want to see is many reallocations occurring within a short period of time - you want them to be spread out. The 256GB 830 used 107 PE cycles yesterday. The 128GB Plextor M3P I'm also testing writes almost as many GB per day, but it uses almost 2x the PE cycles. So all other things being equal, the M3P 128GB should die ~2x as fast. There are many other variables, making it hard to tell for certain -- that's why I'm testing them.

You never know.



I would say that the G2 Intel and the Samsung 830 aren't going to be obsolete any time soon. And if people are going to be dumping them in landfills in a couple years mostly unused, I wish someone would tell me first

[B.A.T]

Thanks Anvil. I'll try that in the weekend. First I need to do an SE and some maintenace on my rig. I run the benchmark and got good read numbers but write was far off.

Almost there with the V. I'll pass 1 PiB in late sunday or early monday.

[Christopher]

B.A.T. --

I'm excited -- I love those damn irritating MTRONs.


The erase count is literally the number of erases (I think it's in 1MB chunks). My 16GB MTRONS would decrease MWI around every 15 000 000 RAW. Unfortunately, because the drive is always doing GC and wear leveling the number is always increasing even if you're not writing to it. You could fill the drive to 100% look at before and after SMART shots, then erase the drive. The Samsung 51nm SLC is rated for 100K PE cycles, so 1 MWI should be 1000 PE cycles. 1000 * 32 768 MB = 32 768 000. Something like that, anyway.

The maximum raw value is 3 200 000 000 for the 32GB, should be 1 600 000 000 for the 16GB (but I have no way of verifying what the hell happens when you get that far).

Your drive is probably ~98.4 MWI if you want to think about it that way.

Writing a 1GB file to the PRO isn't going to tell you all that much. If you do it a couple times, it will be different each time.

I also think I have the .18R FW for your drive, but I would avoid flashing it like the plague. It's not like 4K RWs are going to skyrocket from 0.45MB/s to 0.50MB/s. But I get 46MB/s QD1 Random Reads with my PROs though. So that's nice, and you should be able to get a high avg speed.

[Anvil]

Intel 330 120GB

82.64TB Host writes
Reallocated sectors : 05 0
Available Reserved Space : E8 100
MWI 92
[F1] Total LBAs Written 2707960
[F2] Total LBAs Read 22623 // no movement
[F9] Total NAND Writes 59652GB
POH 228
MD5 OK

122.70MiB/s on avg (~132 hours)

--

Kingston SSDNow 40GB (X25-V)

1000.01TB Host writes
Reallocated sectors : 05 46
Available Reserved Space : E8 99
POH 8953
MD5 OK

33.49MiB/s on avg (~132 hours)

[Christopher]

Kingston SSDNow 40GB (X25-V)

1000.01TB Host writes
Reallocated sectors : 05 46
Available Reserved Space : E8 99
POH 8953
MD5 OK

33.49MiB/s on avg (~132 hours)

Ha Ha!

Excellent

[Hopalong X]

[B]
Kingston SSDNow 40GB (X25-V)

1000.01TB Host writes
Reallocated sectors : 05 46
Available Reserved Space : E8 99
POH 8953
MD5 OK

33.49MiB/s on avg (~132 hours)

Ha Ha!

Excellent

Break out the champagne it is party time!!!!!

[gordo]

For a 830 only using 1 PE cycle a day (based on ~3600 PE cycles) it would take 9.8 years give or take to get to MWI 1. It's unrealistic to expect the drive to last that long though, as generally speaking, it's going to be something else that kills the drive.

Really? What else would kill it? An electrical surge? A blown capacitor? The drive controller chip wearing out?

As far as PE cycles go, would you say that your drive is indicative of what one should expect with Samsung 830s? With your 830 at ~7600 WLC, doesn't that mean that most cells have more or less lasted to about double the manufacturer expected lifetime so far? I assume a core assumption in all this endurance testing is that there's not much variation between drives of the same type at least as far as PE cycles go, which is why you feel comfortable making basic conclusions using only a single drive of each type. If there was a lot of PE cycle lifetime variation between two drives of the same model (say your 830 is a real outlier, and mine is much closer to manufacturer speced PE counts), I would think this whole exercise would not be particularly useful as far as forming generalized conclusions about what a particular drive should be expected to do.

Also, for a given write load, doesn't increasing the size of the drive quickly make up for say the difference between 3k and 5k PE cycle NAND? In other words, if you have a massive write load application, just pick a bigger drive. Alternately, can't manufacturers just move to larger and larger SSDs to account for lower PE Counts associated with smaller and smaller smaller process NAND chips (all else being equal)?

What you don't want to see is many reallocations occurring within a short period of time - you want them to be spread out

Right, but once the NAND is beginning to reach its actual physical limits, assuming you have an excellent wear leveling algorithm, wouldn't one expect that the reallocations actually do start to happen almost all around the same time as each NAND cell reaches its limit?

[Christopher]

Really? What else would kill it? An electrical surge? A blown capacitor? The drive controller chip wearing out?

As far as PE cycles go, would you say that your drive is indicative of what one should expect with Samsung 830s? With your 830 at ~7600 WLC, doesn't that mean that most cells have more or less lasted to about double the manufacturer expected lifetime so far? I assume a core assumption in all this endurance testing is that there's not much variation between drives of the same type at least as far as PE cycles go, which is why you feel comfortable making basic conclusions using only a single drive of each type. If there was a lot of PE cycle lifetime variation between two drives of the same model (say your 830 is a real outlier, and mine is much closer to manufacturer speced PE counts), I would think this whole exercise would not be particularly useful as far as forming generalized conclusions about what a particular drive should be expected to do.

The number one problem is firmware rendering the drive unusable, but you can short them out if you do something stupid (I did that to an MTRON I was testing). The 830 is probably pretty indicative of most -- most of the flash is pretty much identical, at least in so far as it will far exceed what it's rated for. However, maybe through the luck of the draw you get one drive that has one part of one NAND die go bad, or die early. The whole drive would be unable to function unless it has some kind of redundancy, like SF or the Intel 320 series.

Because it would take so long to get to 7700 PE cycles in a modern Win7/w TRIM environment under normal useage scenarios it would be more likely that something else happens along the way (20 years at 1PE cycle a day, 10 years at 2/day).

Also, for a given write load, doesn't increasing the size of the drive quickly make up for say the difference between 3k and 5k PE cycle NAND? In other words, if you have a massive write load application, just pick a bigger drive. Alternately, can't manufacturers just move to larger and larger SSDs to account for lower PE Counts associated with smaller and smaller smaller process NAND chips (all else being equal)?

Absolutely -- there is more flash to write to, and unlike endurance testing, the workload isn't infinite. So if your workload doesn't change, but you upgrade the current drive to a unit 2x the size, it will last much, much longer. It might take you 12 days to go through one PE cycle instead of 7.

Right, but once the NAND is beginning to reach its actual physical limits, assuming you have an excellent wear leveling algorithm, wouldn't one expect that the reallocations actually do start to happen almost all around the same time as each NAND cell reaches its limit?

You would think there is a brick wall, beyond which the NAND is just pushed too far. That's not the case though for the most part... all of the flash is different. Even if it does wear at the same rate, it will still not all die at the same time. If you were just putting the flash chips in a machine that did nothing but write and erase the flash, you might see most flash chips wear out at or around the same time.

[B.A.T]

Kingston SSDNow 40GB (X25-V)

1000.01TB Host writes
Reallocated sectors : 05 46
Available Reserved Space : E8 99
POH 8953
MD5 OK

33.49MiB/s on avg (~132 hours)

Congratz

That took only a year

[B.A.T]

B.A.T. --

I'm excited -- I love those damn irritating MTRONs.


The erase count is literally the number of erases (I think it's in 1MB chunks). My 16GB MTRONS would decrease MWI around every 15 000 000 RAW. Unfortunately, because the drive is always doing GC and wear leveling the number is always increasing even if you're not writing to it. You could fill the drive to 100% look at before and after SMART shots, then erase the drive. The Samsung 51nm SLC is rated for 100K PE cycles, so 1 MWI should be 1000 PE cycles. 1000 * 32 768 MB = 32 768 000. Something like that, anyway.

The maximum raw value is 3 200 000 000 for the 32GB, should be 1 600 000 000 for the 16GB (but I have no way of verifying what the hell happens when you get that far).

Your drive is probably ~98.4 MWI if you want to think about it that way.

Writing a 1GB file to the PRO isn't going to tell you all that much. If you do it a couple times, it will be different each time.

I also think I have the .18R FW for your drive, but I would avoid flashing it like the plague. It's not like 4K RWs are going to skyrocket from 0.45MB/s to 0.50MB/s. But I get 46MB/s QD1 Random Reads with my PROs though. So that's nice, and you should be able to get a high avg speed.

Thanks for the info. I'll try some during the weekend.
btw here are the benchmark before I have done anything to the Mtron.


I'm going to SE it before I'll put it to the test.

[Christopher]

--------------------------------------------

Plextor M3P 128GB Day 5

118,444.08 GiB
115.67 TiB

273.64 MB/s Average

119 Hours

Rellocated Event Count 0

--------------------------------------------

Samsung 830 256GB Day 75

1,782,813.16 GiB
1,741.03 TiB

305.46 MB/s Average
1765 Hours

7742 Wear Leveling Count
MWI 1

6/0 Erase/Program Fail
Used Reserved Block Count: 12/24572 sectors

--------------------------------------------

[neeyuese]

Mtron Pro 32GB SLC

Here are the smartvalues


The nand


The controller


And the spec
Link Link

I need some help to translate the raw value in total erase count.

Hi B.A.T, please find the file attached. it is the special tool for Mtron SSD. (Smartview)

http://www.mediafire.com/?34h0m6qk0nty25n

[B.A.T]

Thank you! I'll try it later today.