SSD Write Endurance 25nm Vs 34nm

[Christopher]

The 830 is also twice the size of the Plextor

And the OCZ drives tended to start dieing after a couple of hundred erase block failures, the samsung is only on 40.

Edit: The plextor reached almost 6 million reallocated sectors before failing outright.

http://www.xtremesystems.org/forums/...=1#post5117724

The Plextor had 15 read failures, the 830 now has 16.

[Gilgamesh]

Sorry for OT, but Anandtech tested Intel 335 240GB (IMFT 20nm sync):
http://www.anandtech.com/show/6388/i...240gb-review/2

"In a few days, I managed to write a total of 37.8TB to the NAND and during that time, the MWI had dropped from 92 to 79. In other words, I used up 13% of the drive's available P/E cycles. This is far from being good news. Based on the data I gathered, the MWI would hit 0 after around 250TB of NAND writes, which translates to less than 1,000 P/E cycles."



preliminary endurance test is "shocking"

I hope that is just only a bugged firmware.

[Anvil]

Kingston SSDNow 40GB (X25-V)

1327.31TB Host writes 43493397*32)
Reallocated sectors : 05 768 -
Available Reserved Space : E8 86 (86)
POH 11971
MD5 OK

31.76MiB/s on avg (~333 hours)

--

As for the 335, even if the figures are correct it'll still do ~200++TiB. ("normal" users will have updated their drives long before one hits 50TiB)

I expect there'll be a few ~20nm drives thrown into the test shortly, a 335 might be one of them.

[Kain665]

Intel's not having an easy time with their 20NM parts. I wouldn't be surprised if <1000 P/E cycles was the case, maybe 1500.

[gordo]

Based on the data I gathered, the MWI would hit 0 after around 250TB of NAND writes, which translates to less than 1,000 P/E cycles.

I hope that is just only a bugged firmware.

I dunno, so far as I can tell, manufacturers set MWI completely arbitrarily. You'll notice the drive used up exactly no reserved space in that time. It could still easily hit 3k P/E cycles, but Intel certianly seems to have set MWI very conservatively on those drives.

[Christopher]

Perhaps it's a warranty consideration? I mean, the 830 exhausted its MWI 180 days ago. It's written 4.5 more petabytes since then, so while Anand's statement doesnt fill me with confidence, it wont keep me up at night. But if they set the warranty length by MWI... It makes more sense. You also have to think that keeping 20nm flash within error specs is hard after a little bit of wear, so the reduction in MWI could be to keep uBER in check. The enterprise spec from JEDEC is 10^16, I don't know what the consumer spec is, but maybe it goes out of spec after a few thousand PE cycles (and isn't so much an endurance issue).


---------

I actually read the review, and I think there is merely something wrong with the FW. I've had two drives both ship with MWI less than 100. In one case it was FW, and in another it was just caused by reallocations out of the box.

[Gilgamesh]

Intel ssd warranty don't cover MWI to 1 on OEM version.

"Any Product the usage of which has reached its write endurance limit, as measured by the SMART attribute (E9) “Media
Wear-out Indicator” value of “1”, as set forth above."

On Boxed there isn't warranty limitation.

Perhaps Intel want preserve 335 "cheap" series, but I think on bugged firmware.

OT closed:

[Christopher]

Samsung 830 256GB Day 223

(GiB) 5,559,835
(TiB) 5,429
(PiB) 5.33


(Avg) 298.76 MB/s over the past 2500+ hours

(B1) Wear Leveling Count: 24,037

(B6) Erase Fail Count: 14
(B6) Runtime Bad Block Count: 37
(05) Reallocated Sectors: 151552

(POH) 5435


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

Reallocations have been growing steadily over the past week, one every few hours.

[Zaxx]

Intel ssd warranty don't cover MWI to 1 on OEM version.

On Boxed there isn't warranty limitation.

So that's the difference between boxed and oem...had I known, I'd have paid the extra few bucks for that little extra piece of mind.

[Zaxx]

Intel ssd warranty don't cover MWI to 1 on OEM version.

On Boxed there isn't warranty limitation.

So that's the difference between boxed and oem...had I known, I'd have paid the extra few bucks for that little extra piece of mind.

[Christopher]

So that's the difference between boxed and oem...had I known, I'd have paid the extra few bucks for that little extra piece of mind.

I ended up buying a 120GB Intel 320 OEM, and the warranty on that was 1 year and not five. Totally busted if you ask me, but I suppose I bear most of the blame.

[Christopher]

Samsung 830 256GB Day 225

(GiB) 5,601,463
(TiB) 5,155
(PiB) 5.37


(Avg) 298.77 MB/s over the past 2000+ hours

(B1) Wear Leveling Count: 24,216

(B5) Program Fail Count: 0
(B6) Erase Fail Count: 14
(B7) Runtime Bad Block Count: 41
(05) Reallocated Sectors: 167936

Read Failures: 27

(POH) 5475


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

[B.A.T]

Here are todays update:

m4
1882.8672 TiB
7727 hours
Avg speed 73.79 MiB/s.
AD 63 to 44
P/E 32412
C3 48482335 to 104131897
01 13329 to 30726 (1)
CE 90
MD5 OK.
Reallocated sectors : 00


Intel X25-E 64GB
1.25 PiB
4192-30=4162 hours
Reallocated sectors : 173 to 255 to 43
Available Reserved space: 71 to 67
MWI= 92 to 91
MD5 =OK
79.21 MiB/s on avg


Mtron Pro 7025 32GB
Paused until new firmware is updated.

OCZ Vertex 1 120GB
Will start after an update to the latest firmware.

[Christopher]

Samsung 830 256GB Day 228

(GiB) 5,687,048
(TiB) 5,553
(PiB) 5.45


(Avg) 298.86 MB/s over the past 2000+ hours

(B1) Wear Leveling Count: 25,586

(B5) Program Fail Count: 0
(B6) Erase Fail Count: 14
(B7) Runtime Bad Block Count: 59
(05) Reallocated Sectors: 167936

Read Failures: 45

(POH) 5557


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

[B.A.T]

Here are todays update:

m4
1897.9654 TiB
7789 hours
Avg speed 73.67 MiB/s.
AD 44 TO 35
P/E 32674
C3 104131897 to 136440144
01 30726 to 41196 (1)
CE 90
MD5 OK.
Reallocated sectors : 00


Intel X25-E 64GB
1.26 PiB
4253-30=4223 hours
Reallocated sectors : 43 to 85
Available Reserved space: 67 to 66
MWI= 91
MD5 =OK
79.03 MiB/s on avg


Mtron Pro 7025 32GB
Paused until new firmware is updated.

OCZ Vertex 1 120GB
Will start after an update to the latest firmware.

[Christopher]

Samsung 830 256GB Day 230

(GiB) 5,732,148
(TiB) 5,597
(PiB) 5.50


(Avg) 298.86 MB/s over the past 2000+ hours

(B1) Wear Leveling Count: 24,781

(B5) Program Fail Count: 0
(B6) Erase Fail Count: 15
(B7) Runtime Bad Block Count: 68
(05) Reallocated Sectors: 278528

Read Failures: 53

(POH) 5600

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

The 830 is continuing to fail blocks predictably and steadily, which is reassuring to see.

[gordo]

Is the 830 going to hit 6PB at this block failure rate?

Incidentally, looks like a typo on the WLC number...

[vpr5703]

When write amplification is taken into account, 24K Write Cycles sounds right. At 5.5PiB write, that is 22528 WLC. But that assumes a perfect 1.0 write amp. rate, which only a few SSDs can accomplish using compression (SandForce chips, I think). If we assume that 24781 WLC is the entire drive, then the drive has has 6.05 PiB written. 6.05 / 5.5 is a write amplification rate of ~ 1.1 Not too bad!

[Christopher]

Is the 830 going to hit 6PB at this block failure rate?

Incidentally, looks like a typo on the WLC number...

I believe so, but maybe not by much. There was a typo in the last update's WLC, this one is right. Incidentally, because the read and write failures are so steady and predictable, I think it bodes well for 6PB.

When write amplification is taken into account, 24K Write Cycles sounds right. At 5.5PiB write, that is 22528 WLC. But that assumes a perfect 1.0 write amp. rate, which only a few SSDs can accomplish using compression (SandForce chips, I think). If we assume that 24781 WLC is the entire drive, then the drive has has 6.05 PiB written. 6.05 / 5.5 is a write amplification rate of ~ 1.1 Not too bad!

The WLC is probably accurate for the whole drive, but most WLCs are usually for just a single device/channel/etc. I figured WA at 1.1 -- if you take the LBA written and divide it by WLC you get about 231GB written per PE cycle. lBA writes are ahead of ASU by about 15GB (I "lost" some writes for about 14 hours one day).

[Christopher]

And Since it came up, it looks like SF drives using non-2281 controllers (ones with better SMART data) have another attribute that shows GB erased in addition to host and raw writes. I'll have to perform some more in depth analysis, but it looks like the actual WA for a mostly incompressible workload is significantly higher than the host/raw number would indicate.

[B.A.T]

Here are todays update:

m4
1909.4236 TiB
7789 hours
Avg speed 73.58 MiB/s.
AD 35 to 29
P/E 32873
C3 136440144 to 172698757
01 41196 to 54019 (1)
CE 90
MD5 OK.
Reallocated sectors : 00


Intel X25-E 64GB
1.27 PiB
4299-30=4269 hours
Reallocated sectors : 85 to 130
Available Reserved space: 66 to 65
MWI= 91
MD5 =OK
78.89 MiB/s on avg


Mtron Pro 7025 32GB
Paused until new firmware is updated.

OCZ Vertex 1 120GB
Will start after an update to the latest firmware.

[B.A.T]

Here are todays update:

m4
1921.2473 TiB
7884 hours
Avg speed 73.46 MiB/s.
AD 29 to 23
P/E 33079
C3 172698757 to 221068763
01 54019 72375 (1)
CE 90
MD5 OK.
Reallocated sectors : 00


Intel X25-E 64GB
1.29 PiB
4348-30=4318 hours
Reallocated sectors : 130 to 183
Available Reserved space: 65 to 63
MWI= 91
MD5 =OK
78.72 MiB/s on avg


Mtron Pro 7025 32GB
Paused until new firmware is updated.

OCZ Vertex 1 120GB
Will start after an update to the latest firmware.

[vpr5703]

I believe so, but maybe not by much. There was a typo in the last update's WLC, this one is right. Incidentally, because the read and write failures are so steady and predictable, I think it bodes well for 6PB.



The WLC is probably accurate for the whole drive, but most WLCs are usually for just a single device/channel/etc. I figured WA at 1.1 -- if you take the LBA written and divide it by WLC you get about 231GB written per PE cycle. lBA writes are ahead of ASU by about 15GB (I "lost" some writes for about 14 hours one day).

That is still pretty good, considering. I like where this drive is going. Gives me lots of hope (providing that my device doesn't bite the dust due to flash failure) that my drive will outlast my laptop.

[Anvil]

Kingston SSDNow 40GB (X25-V)

1348.84TB Host writes 44198881*32)
Reallocated sectors : 05 768 -
Available Reserved Space : E8 86 (86)
POH 12257
MD5 OK

32.02MiB/s on avg (~194 hours)

--

[B.A.T]

Here are todays update:

m4
1927.2346 TiB
7908 hours
Avg speed 73.42 MiB/s.
AD 23 to 18.
P/E 33183
C3 221068763 to 247252828
01 72375 to 82343 (1)
CE 90
MD5 OK.
Reallocated sectors : 00


Intel X25-E 64GB
1.29 PiB
4372-30=4342 hours
Reallocated sectors : 183 to 205
Available Reserved space: 63 to 62
MWI= 91
MD5 =OK
78.72 MiB/s on avg


Mtron Pro 7025 32GB
Paused until new firmware is updated.

OCZ Vertex 1 120GB
Will start after an update to the latest firmware.