SSD Write Endurance 25nm Vs 34nm

[Christopher]

Maybe what the super-vague Toshiba chart is showing is 0 percent chance of losing data retention up PE cycles = rated, increasing chance up to 2x MWI, and a greater chance (greatly reduced retention time) between 2x and 3x rated PE cycles. Once a drive hits 3x rated PE cycles, the chance of retention failure is very high (and or the amount of time rapidly decreases)...

But I don't think it works like that. And the chart could be referring to SLC which is probably quite different from MLC in that respect.

[Anvil]

Christopher

Could you fill in the details (I've posted a preliminary update on the drives in post #1) on the new drive?

I expect you did not physically verify NAND?

[Anvil]

Kingston SSDNow 40GB (X25-V)

643.21TB Host writes
Reallocated sectors : 05 21
Available Reserved Space : E8 99
POH 5806
MD5 OK

33.23MiB/s on avg (~37 hours)

--

Corsair Force 3 120GB

01 82/50 (Raw read error rate)
05 2 (Retired Block count)
B1 54 (Wear range delta)
B6 1 (Erase Fail Count)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 731957 (Raw writes) ->715TiB
F1 974108 (Host writes) ->951TiB

MD5 OK

106.86MiB/s on avg (~37 hours)

power on hours : 2779

[johnw]

Maybe what the super-vague Toshiba chart is showing is 0 percent chance of losing data retention up PE cycles = rated, increasing chance up to 2x MWI, and a greater chance (greatly reduced retention time) between 2x and 3x rated PE cycles. Once a drive hits 3x rated PE cycles, the chance of retention failure is very high (and or the amount of time rapidly decreases)...

No, your interpretation makes no sense. It is clearly a log10 vs. log10 chart, with the decades clearly visible, which is why we are looking at X, 10X, 100X, and not X, 2X, and 3X as you say.

More importantly, data retention DOES decrease significantly before MWI is exhausted. Very roughly, data retention should be at least 10 years at 10% of MWI, 5 years at 50% of MWI, and 1 year at 100% of MWI.

[Christopher]

I don't think it makes sense at all... That's my interpretation.

[Ao1]

That graph can be found on 6 out of the 6 product datasheets I have aquaired for Toshiba drives. As content of a product datasheet I can’t imagine they made it up as any form of misleading claim would be problematic.

I found a Samsung datasheet that stated 10 years for data retention. I suspect 10 years from new and 1 year after the MWI has run out.

[johnw]

That graph can be found on 6 out of the 6 product datasheets I have aquaired for Toshiba drives. As content of a product datasheet I can’t imagine they made it up as any form of misleading claim would be problematic.

It would seem that you have a poor imagination!

It is clearly made up. Real data does not look like that.

[Ao1]

Maybe

[B.A.T]

Todays update.
Kingston V+100
338.6542 TiB
1939 hours
Avg speed 25.03 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
253,2724 TiB
20347 hours
Reallocated sectors : 00
MWI=132 to 131
MD5 =OK
42.33 MiB/s on avg


m4
262.4960 TiB
969 hours
Avg speed 79.73 MiB/s.
AD gone from 208 to 203.
P/E 4618.
MD5 OK.
Reallocated sectors : 00


The Kingston is back up runnung. Everything looks ok for now. I also did some maintenance in the Intel G1 after the screenshot to regain some write speed.

[Christopher]

Samsung 830 64GB Update, Day 45
FW:CXM02B1Q

TiB Written:
268.6338

GiB written:
275081.03

Avg MB/s
126.54

PE Cycles
14881 <-- not the true count as it stopped realtime updates
Reallocated Sectors

40960
20 Blocks, holding steady

1066Hours




So there are still many questions about the SMART attributes, but at least POR recovery count is unsafe shutdown count, and wear leveling is PE cycles and MWI. But I'm puzzled at the lack of new bad blocks. That, and why PE cycles won't update now -- I think it should start in a couple of days.

[Christopher]

OCZ Vertex Turbo 64 Update Day 0

TiB 3.4471
GiB 3529.87
Avg MB/s 82.50

Avg Erase Count
62 up from 1

MWI 97 down from 100


1 Program Fail
13 Erase Fail
0 Read Fail
12 Hours

[Christopher]

I've also decided to start testing one of the MTRON 16GB SLCs. I have the 3.5" 7000 PRO, and the MOBI 3000 should be here in a couple days. I think I'll test the MOBI since it's a 2.5" drive and will be easier to cram in the endurance rig.

Neither drive will have much SMARt info... but I don't really have a use for a 16GB drive without NCQ, TRIM, or much speed other than endurance testing. I think performance between the drives should be almost identical. They're not really good for much, and installing Win7 on the 7000 PRO took hours. Booting takes several times longer than it should, though it doesn't stutter and does have pretty high write speeds vs. capacity. I'll take the drive apart to verify how much and what kind of NAND is onboard - it's not clear whether the drives are using OP of any kind, but based on it's Korean origin I would assume Samsung SLC similar to what is in my Vertex EX.

[Ao1]

Host writes: 12,632 GiB
MWI: 37
P/E Cycles: 2,283
POH: 306
Relocated sectors: 4,096
MB/s: 5.97

[iceblade]

New firmware for Samsung 830 available today from Samsung: CXM03B1Q

[bluestang]

Christopher
Did that VT64 come brand new, sealed with FW 1.7 already on it?
Also, if it came with 13 bad blocks out of the box, they should have been listed under "Initial Bad Block Count". No?

[Christopher]

Bluestang,

As a matter of fact you are correct. But -- mine when unsealed had 18 initial bad blocks and 5 erase failures. I had a picture of this, but the earliest one I can find was after a few GB of writes from ASU and CDM.

It only had a couple KB of writes on it new out-of-the-box. But when combined with the runtime bad blocks is why MWI was at 97 out of the box. I've been trying to get some answers about this for quite some time over on the OCZ forums, but didn't get much in the way of help.

For instance, I didn't know that bad blocks factored into MWI -- but that's clearly what happened. I also didn't know that the purpose of the NAND cleaner was to mark bad blocks as good so that FW could be reflashed -- which makes me wonder if you could have gotten a couple more GB out of the M225 with a NAND cleaning and D-flash as at one point I couldn't dflash as there were too many bad blocks -- which doesn't make any sense. If they're marked as bad from the factory, it's like they never existed. If they're not and pop up in runtime, it does affect MWI.

Anyway, I screwed up last night and accidentally stopped the Vertex, so it's lost about 8 hours of running time. It's averaging about 82MBs. But after the NAND cleaning, it averages about 140MBs for a couple of loops.

[Anvil]

Kingston SSDNow 40GB (X25-V)

645.92TB Host writes
Reallocated sectors : 05 21
Available Reserved Space : E8 99
POH 5830
MD5 OK

33.03MiB/s on avg (~61 hours)

--

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 52 (Wear range delta)
B6 1 (Erase Fail Count)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 738755 (Raw writes) ->721TiB
F1 983142 (Host writes) ->960TiB

MD5 OK

106.83MiB/s on avg (~61 hours)

power on hours : 2803

[Ao1]

Host writes: 13,162 GiB
MWI: 33
P/E Cycles: 2,446
Relocated sectors: 4,096
MB/s: 6.27

[Ao1]

OK I have a theory. Samsung support stated that:

The raw value on this attribute [177] is how many times it can prolong the life of a specific block.

Assuming the normal value is the MWI the raw value should be 3,000 by the time it gets to 1. It’s not quiet working out that way. I will end up with more than 3,000 by the time I get to MWI 1. I think the raw value above 3,000 might therefore reflect the amount of times that a P/E cycle was avoided due to block management.

What do you reckon?

For my data retention test I have to decide; do I get to MWI 1 or do I use the RAW value of 3,000?

[minpayne]

OK I have a theory. Samsung support stated that:

The raw value on this attribute [177] is how many times it can prolong the life of a specific block.

Assuming the normal value is the MWI the raw value should be 3,000 by the time it gets to 1. It’s not quiet working out that way. I will end up with more than 3,000 by the time I get to MWI 1. I think the raw value above 3,000 might therefore reflect the amount of times that a P/E cycle was avoided due to block management.

What do you reckon?

For my data retention test I have to decide; do I get to MWI 1 or do I use the RAW value of 3,000?

I would do it with MWI 1. Two reasons:

a) It is likely to be a more strict test;

b) The P/E spec might be more than 3000 (e.g. 5000 like the Intel 320), and if taken WA into account it might end up with more than 3000 but less than the P/E spec.

[johnw]

OK I have a theory. Samsung support stated that:

The raw value on this attribute [177] is how many times it can prolong the life of a specific block.

What is the SMART reported threshold value for attribute 177?

[Ao1]

The threshold is 0

[Christopher]

Samsung 830 64GB Update, Day 47
B]FW:[/B]CXM02B1Q

TiB Written:
285.9084

GiB written:
292770.25

Avg MB/s
125.90

PE Cycles
14881 <-- not the true count as it stopped realtime updates
Reallocated Sectors

40960
20 Blocks, holding steady

1106 hours




OCZ Vertex Turbo 64 Update Day 3

TiB 11.3107
GiB 11582.13
Avg MB/s 82.86

Avg Erase Count
205 up from 62

MWI 96 down from 97


1 Program Fail
14 Erase Fail
0 Read Fail
49 Hours

[Christopher]

The Vertex Turbo is slow, and the lack of new failed blocks on the Samsung puzzles me greatly.

It's been almost a month since the last block got reallocated on the Samsung...

[bluestang]

How does Vertex Turbo come with 1.7 already loaded? Isn't 1.7 fairly recent, and the Turbo is an older drive??
You might want to pop that sucker open and take a looksee!