SSD Write Endurance 25nm Vs 34nm

[Anvil]

Intel 330 120GB

641.41TB Host writes
2.32TiB Host reads
Reallocated sectors : 05 43
Available Reserved Space : E8 100
MWI 25
[B5] 21
[B6] 22
[F1] Total LBAs Written 21017815
[F2] Total LBAs Read 75912
[F9] Total NAND Writes 463017GB // ~452TiB
POH 1546
MD5 OK

125.50MiB/s on avg (~11 hours)

--

Kingston SSDNow 40GB (X25-V)

1152.56TB Host writes (37767241*32)
Reallocated sectors : 05 67
Available Reserved Space : E8 98
POH 10282
MD5 OK

36.89MiB/s on avg (~11 hours)

--

I needed the SATA ports last night so both drives had a 10 minute break. (gave them a cleaning as well by triggering TRIM)

--

Endurance_cr_20120728_1500.png

I've changed the filling for the drives that are paused.

[Anvil]

Based on current figures this might be the outcome at MWI 0

I330_act_est.png

[kensiko]

We used a Vertex 3 MaxIOPS 240GB at work for a critical server.

We had write errors from Oracle, so we switched the SSD. We discovered that the SSD, even with a almost perfect SMART data (2TB write, 200TB read, maximum difference between most worn cell and best cell = 1), it was impossible to copy a 8GB file entirely, it blocks when there is 760MB left. All the other data was intact.

How can this be possible? Isn't the controller supposed to mark bad blocks and not use them? Or is it because the SSD doesn't have idle time?
Yesterday I upgraded the firmware to 2.22. It was 2.15 since installed.

[canthearu]

We are really not a support service for OCZ, especially since I'm here to kill OCZ drives, but there seems to be something ether wrong with your SSDs or server. I've never seen such a situation on SSDs here, and I've copied some whopper 20gig files to and from them.

[Zaxx]

Kensiko...this thread is dedicated to endurance testing, maybe you should post in the hardware-storage section.

[B.A.T]

Here are todays update:

m4
1295.2153 TiB
5392 hours
Avg speed 74.43 MiB/s.
AD 133 to 126
P/E 22267.
C3 4457
CE 58
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
825.25 TiB
24679 hours
Reallocated sectors : 226 to 255 to 17
Available Reserved space: 40 to 37
MWI= 128 to 111
MD5 =OK
30.75 MiB/s on avg


Intel X25-E 64GB
594.85 TiB
1860-30=1830 hours
Reallocated sectors : 0
Available Reserved space: 99
MWI= 96
MD5 =OK
91.32 MiB/s on avg


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

[Christopher]

Samsung 830 256GB Day 129

(GiB) 3,197,663
(TiB) 3,122
(PiB) 3.07


(Avg) 298.46 MB/s

(B1) Wear Leveling Count: 13,849

(B6) Erase Fail Count: 9
(05) Reallocated Sectors: 36864
(POH) 3177


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

4PB, here I come. It's odd that it's been almost 1.5PB without an erase failure, but hey, I'll take it.

[Anvil]

Intel 330 120GB

651.40TB Host writes
2.34TiB Host reads
Reallocated sectors : 05 45 +2
Available Reserved Space : E8 100
MWI 24
[B5] 22
[B6] 23
[F1] Total LBAs Written 21345238
[F2] Total LBAs Read 76706
[F9] Total NAND Writes 470230GB // ~459TiB
POH 1569
MD5 OK

125.28MiB/s on avg (~35 hours)

--

Kingston SSDNow 40GB (X25-V)

1155.21TB Host writes (37854060*32)
Reallocated sectors : 05 67
Available Reserved Space : E8 98
POH 10305
MD5 OK

34.38MiB/s on avg (~35 hours)

--

[canthearu]

Sandisk Extreme G25 - 120gb - Day 54

Drive hours: 1291
ASU GiB written: 558,052.55 GiB (544.97 TiB)
Avg MB/s: 125.86 MB/s (69.64 hours)
MD5: OK

Host GB written (F1): 562,193 GiB (549.02 TiB)
NAND writes (E9): 419,639 GiB (409.80 TiB)

Retired Block Count (05): 4 raw
Failure count (AB, AC): 1 program, 1 erase
Raw Error Rate (01): 100 normalized
Media Wearout Indicator (E7): 10 normalized
Wear Range Delta (B1): 4 raw
Reported Uncorrectable Errors (BB): 0 raw

Vertex 4 - 50% OP - 128gb - PAUSED

Drive hours: 214
ASU GiB written: 96,525.65 GiB (94.26 TiB)
Avg MB/s: 0 MB/s (0 hours)
MD5: OK

Host GB written (E8): 97,836.11 GiB (95.54 TiB, 205177187947 raw)

Raw Read Error Rate (01): 5 raw
Reallocated Block Count (05): 0 raw
Remaining Life (E9): 77 normalised

I decided to let the V4 go for a while, but the loop started degrading to sub 100meg per sec speeds (after the initial burst of performance)

Drive just doesn't want to do what I want it to do.

[B.A.T]

Here are todays update:

m4
1301.0084 TiB
5415 hours
Avg speed 74.71 MiB/s.
AD 126 to 123
P/E 22365.
C3 4457
CE 58
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
827.79 TiB
24702 hours
Reallocated sectors : 17 to 42
Available Reserved space: 37 to 35
MWI= 111 to 104
MD5 =OK
29.87 MiB/s on avg


Intel X25-E 64GB
602.03 TiB
1883-30=1853 hours
Reallocated sectors : 0
Available Reserved space: 99
MWI= 96
MD5 =OK
92.44 MiB/s on avg


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

I'm going for 3 weeks on vacation so my updates can be kind of sporadic. I've installed a ups to prevent a shutdown from powerfailures and will be crossing my fingers
I think that the G1 might soon die on me. It droppes reserve space at 1-2% every day, just hope it stays in there until I'm back again.

[Christopher]

B.A.T. -- I don't know about you, but my drives only like to die when I'm out of town. I'd say it's a guarantee that something will die while you're gone. That's just how it works.

[B.A.T]

I'll just have to wait and see

[Anvil]

Intel 330 120GB

661.85TB Host writes
2.37TiB Host reads
Reallocated sectors : 05 52 +7
Available Reserved Space : E8 100
MWI 22
[B5] 25
[B6] 27
[F1] Total LBAs Written 21687618
[F2] Total LBAs Read 77614
[F9] Total NAND Writes 477773GB // ~466TiB
POH 1593
MD5 OK

125.31MiB/s on avg (~59 hours)

--

Kingston SSDNow 40GB (X25-V)

1157.99TB Host writes (37945037*32)
Reallocated sectors : 05 67
Available Reserved Space : E8 98
POH 10329
MD5 OK

33.89MiB/s on avg (~59 hours)

--

Endurance_cr_20120730_1219.png

[Anvil]

Intel 330 120GB

672.40TB Host writes
2.39TiB Host reads
Reallocated sectors : 05 60 +8
Available Reserved Space : E8 100
MWI 21
[B5] 29
[B6] 31
[F1] Total LBAs Written 22033283
[F2] Total LBAs Read 78465
[F9] Total NAND Writes 485387GB // ~474TiB
POH 1617
MD5 OK

125.41MiB/s on avg (~83 hours)

--

Kingston SSDNow 40GB (X25-V)

1160.78TB Host writes (38036572*32)
Reallocated sectors : 05 68 +1
Available Reserved Space : E8 98
POH 10354
MD5 OK

33.69MiB/s on avg (~83 hours)

--

[canthearu]

Sandisk Extreme G25 - 120gb - Day 56

Drive hours: 1343
ASU GiB written: 580,836.71 GiB (567.227 TiB)
Avg MB/s: 125.13 MB/s (121.85 hours)
MD5: OK

Host GB written (F1): 585,091 GiB (571.38 TiB)
NAND writes (E9): 436,741 GiB (426.50 TiB)

Retired Block Count (05): 4 raw
Failure count (AB, AC): 1 program, 1 erase
Raw Error Rate (01): 116 normalized
Media Wearout Indicator (E7): 10 normalized
Wear Range Delta (B1): 4 raw
Reported Uncorrectable Errors (BB): 0 raw

[Anvil]

I gave both drivers a cleaning (trim) last night, doesn't look like it helped.

Intel 330 120GB

682.46TB Host writes
2.42TiB Host reads
Reallocated sectors : 05 74 14
Available Reserved Space : E8 100
MWI 20
[B5] 36
[B6] 38
[F1] Total LBAs Written 22362756
[F2] Total LBAs Read 79261
[F9] Total NAND Writes 492645GB // ~481TiB
POH 1640
MD5 OK

125.22MiB/s on avg (~12 hours)

--

Kingston SSDNow 40GB (X25-V)

1163.53TB Host writes (38126614*32)
Reallocated sectors : 05 68
Available Reserved Space : E8 98
POH 10377
MD5 OK

35.29MiB/s on avg (~12 hours)

--

[Anvil]

Reallocated sectors is currently at 80 on the 330, it does look controlled but it is a bit early for a 120GB drive.

According to the recent review at Anandtech it is populated with "cheaper, lower endurance MLC NAND".

I330_MWI_vs_RSC_2012_08_01_1700.png


Endurance_cr_20120731.png

[johnw]

According to the recent review at Anandtech it is populated with "cheaper, lower endurance MLC NAND".

The odd thing about Anand's review is his short endurance test. He has 7629 GiB of NAND writes to a 60GB 330, but MWI is still at 100:



If the 330 flash has 3000 P/E cycles, then I would expect each decrement in MWI to be 30 (=300/100) cycles. But if there is 64GiB of flash on board, then 7629 GiB comes to 119.2 (=7629/64) cycles. The MWI should be at 97 or 96. But it is still at 100.

If the MWI is linear (and not buggy), then at least 119.2 cycles per MWI comes to at least 11920 P/E cycles for the flash, which is hard to believe.

I wonder if there is a bug in Intel's MWI computation. Maybe they are using host writes instead of NAND writes. Since Anand's host writes are only 1.55 TiB, that comes to only 24.8 (=1.55x1024/64) cycles if we incorrectly assume write amplification is 1.0. Since 3000 P/E cycle flash would be 30 cycles per MWI decrement, then we would expect the MWI to still be at 100 if Intel is incorrectly using host writes rather than NAND writes to compute MWI.

In contrast, the MWI for Anvil's 120GB 330 looks like it is counting for about 5000 cycles:

682.46TB Host writes
2.42TiB Host reads
Reallocated sectors : 05 74 14
Available Reserved Space : E8 100
MWI 20
[B5] 36
[B6] 38
[F1] Total LBAs Written 22362756
[F2] Total LBAs Read 79261
[F9] Total NAND Writes 492645GB // ~481TiB

492645GiB/128GiB = 3848.8 cycles.

If MWI wears out at 0, then it looks like it is counting to:

3848.8 x 100 / (100 - 20) = 4811 P/E cycles

[Anvil]

There is a grace period just like in the original SF firmware.

It took 14.87 TiB of NAND writes (21.12TiB of Host Writes) before it turned 99.

It averages (from WMI 98 and down) 8.29TiB (host) / 5.84TiB (nand) writes between each MWI decrement.

At the current rate, my drive looks to point to about 5400 (writes / capacity "P/E cycles") and on average 53.53 cycles per MWI.
(figures are changing as reallocated-sectors increase)

Bases on NAND writes

I330_MWI_vs_RSC_NAND_2012_08_01.png

[johnw]

There is a grace period just like in the original SF firmware.

That seems like a bug to me. I can see having a "grace period" for not throttling the drive under heavy writes. But it makes no sense to have a "grace period" for MWI. I have seen no evidence that the flash wears out more slowly during the first few percent of flash writes.

I'm surprised Intel did not fix that bug. Maybe Intel does not really have full access to the Sandforce firmware.

[Anvil]

As SMART is adapted to Intel's way of reporting there could of course be side-effects.

Will have to check how the other (non-throttled) SF based drives worked during the first few MWI decrements.

The Force 3 changed to 99 at about 15TiB and then again to 98 at about 20TiB. (Link)

[Kain665]

Maybe Intel does not really have full access to the Sandforce firmware.

Ahem. *Cough*

[B.A.T]

Here are todays update:

m4
1320.3993 TiB
5491 hours
Avg speed 74.58 MiB/s.
AD 123 to 112
P/E 22695.
C3 4457
CE 58
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
835.05 TiB
24778 hours
Reallocated sectors : 42 to 115
Available Reserved space: 35 to 31
MWI= 104
MD5 =OK
27.79 MiB/s on avg


Intel X25-E 64GB
625.97 TiB
1959-30=1929 hours
Reallocated sectors : 0 to 2
Available Reserved space: 99
MWI= 96 to 95
MD5 =OK
91.68 MiB/s on avg


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

The G1 continues its path to death and the E had its first 2 reallocated sectors. The m4 just steams on

[Anvil]

Intel 330 120GB

692.78TB Host writes
2.44TiB Host reads
Reallocated sectors : 05 100 +26
Available Reserved Space : E8 100
MWI 18
[B5] Program fail count 49
[B6] Erase fail count 51
[F1] Total LBAs Written 22700885
[F2] Total LBAs Read 80112
[F9] Total NAND Writes 500094GB // ~488TiB
POH 1664
MD5 OK

125.36MiB/s on avg (~35 hours)

--

Kingston SSDNow 40GB (X25-V)

1166.27TB Host writes (38216332*32)
Reallocated sectors : 05 68
Available Reserved Space : E8 98
POH 10401
MD5 OK

33.90MiB/s on avg (~35 hours)

--

[canthearu]

Sandisk Extreme G25 - 120gb - Day 58

Drive hours: 1343
ASU GiB written: 600,727.94 GiB (586.65 TiB)
Avg MB/s: 124.72 MB/s (167.61 hours)
MD5: OK

Host GB written (F1): 605,082 GiB (590.90 TiB)
NAND writes (E9): 451,673 GiB (441.08 TiB)

Retired Block Count (05): 4 raw
Failure count (AB, AC): 1 program, 1 erase
Raw Error Rate (01): 118 normalized
Media Wearout Indicator (E7): 10 normalized
Wear Range Delta (B1): 4 raw
Reported Uncorrectable Errors (BB): 0 raw