SSD Write Endurance 25nm Vs 34nm

[B.A.T]

The m4 is completely dead. It's not recognized at startup and nothing I do seems to help.

[Anvil]

Kingston SSDNow 40GB (X25-V)

515.08TB Host writes
Reallocated sectors : 05 13
Available Reserved Space : E8 99

MD5 OK

32.96MiB/s on avg (~145 hours)

* reallocated sectors increased by 1, it's been months since last time it moved.

--

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 54 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 463344 (Raw writes) ->452TiB
F1 618039 (Host writes) ->604TiB

MD5 OK

106.37MiB/s on avg (~239 hours)

power on hours : 1770

B1 is down from last reading, from 56 to 54.

I have paused the test of the F3 and will be starting a retention test later today, as an experiment I'll let it idle for 3-4 hours before starting the test.

[Anvil]

...
However I'd like to see another type of test, i.e. does the nand flash show significant volatility, once the MWI is exhausted? I'd like to have several SSDs of different brands/models to reach the MWI threshold then left powered off for a week or so and see if the contents are lost. By this we can know whether the "MWI Lasted" is indeed the best value to calculate the endurance.

The contents will last for at least 1 year if one stops using the drive when MWI is exhausted, that is the spec.

We are trying to incorporate some sort of retention testing into the test and it was during one of these tests that B.A.T's drive stopped working.

[One_Hertz]

604TB. 7424 reallocations.

Not a single new reallocated sector for over a week now.

[bluestang]

Hey guys,

As we all have noticed, Eric (Vapor) has been MIA for a while now and I know we have all been concerned as to why and have tried to get in contact with him. Well, I sent a PM here at XS to Cameron (skinnee), his partner over at Skinnee Labs and here is some info:

Hi skinnee,

Hate to bother you, but just wanted to ask if Eric (Vapor) is OK. We miss him over in the SSD Endurance Test thread. We hope all is well and if you can pass on the info that we would like to hear from him that would be great.

Thanks for your time.

Best Regards,
Dave (bluestang)

Hey Dave,

No worries, you can bother me any time.

Eric is well, he's just been away tending to some personal business is all. I've been talking to him regularly, so I know he is well. But I'll pass along your kinds words, he'll be glad to hear it.

Regards,
Cameron

Hi Cameron,

Thanks for the news that all is well with Eric and yes, pass along our best. Also, let him know that if he needs anything at all to let us over here at XS know and we'll do all we can.
Thanks again!

Best Regards,
Dave

P.S. -- Is it ok if I pass this along to the guys in the SSD Endurance thread? I know they would like to know everything is ok.

Hey Dave,
Already have, he appreciates it. And yes, feel free to let the guys know he is well and will be back as soon as he can.

Regards,
Cam

[bluestang]

M225->Vertex Turbo 64GB Update:

Well, I'm back to beating the hell out of it...after all the is XtremeSystems

846.45 TiB (930.68 TB) total
~2172 hrs (Torture), 3136 hrs (Power-On)
16490 Raw Wear
117.33 MB/s avg for the last 24.22 hours (on W7 x64)
MD5 every 20 loops on 1.59GB file = OK
C4-Erase Failure Block Count (Realloc Sectors) from 18 to 19.
C5-Read Failure Block Count (uncorrectable bit errors) at 5.
1=Bnk 6/Blk 2406 - Erase Failure C4
2=Bnk 3/Blk 3925 - Erase Failure C4
3=Bnk 0/Blk 1766 - Erase Failure C4
4=Bnk 0/Blk 829 - Erase Failure C4
5=Bnk 4/Blk 3191 - Erase Failure C4
6=Bnk 7/Blk 937 - Erase Failure C4
7=Bnk 7/Blk 1980 - Erase Failure C4
8=Bnk 7/Blk 442 - Erase Failure C4
9=Bnk 7/Blk 700 - Erase Failure C4
10=Bnk 2/Blk 1066 - Erase Failure C4
11=Bnk 7/Blck 85 - Erase Failure C4
12=Bnk 4/Blk 3192 - Erase Failure C4
13=Bnk 7/Blk 280 - Erase Failure C4
14=Bnk 3/Blk 2375 - Erase Failure C4
15=Bnk 7/Blk 768 - Erase Failure C4
16=Bnk 7/Blk 765 - Erase Failure C4
17=Bnk 7/Blk 182 - Erase Failure C4
18=Bnk 5/Blk 939 - Read Failure C5
19=Bnk 5/Blk 1115 - Read Failure C5
20=Bnk 5/Blk 1011 - Read Failure C5
21=Bnk 7/Blk 3549 - Read Failure C5
22=Bnk 7/Blk 3556 - Read Failure C5
23=Bnk 4/Blk 1961 - Erase Failure C4
24=Bnk 7/Blk 1862 - Erase Failure C4

[Anvil]

As we all have noticed, Eric (Vapor) has been MIA for a while now and I know we have all been concerned as to why and have tried to get in contact with him. Well, I sent a PM here at XS to Cameron (skinnee), his partner over at Skinnee Labs and here is some info:

I didn't notice that skinnee was a member here at XS but I figured that it was the same Eric

We do need some charts and so I'll see what I can do, a bit much to do at work right now but I'll try...

M225->Vertex Turbo 64GB Update:
Well, I'm back to beating the hell out of it...after all the is XtremeSystems

604TB. 7424 reallocations.
Not a single new reallocated sector for over a week now.

Looks like I'm catching up on TB's vs the 320, has it been offline lately?

--

I disconnected the F3 ~15 minutes ago, will let it suffer w/o power for a few days

[Christopher]

It would be most difficult to recreate the charts as they were. Barring the ability to get the original files and then update them, we should consider a new chart strategy.

[One_Hertz]

Looks like I'm catching up on TB's vs the 320, has it been offline lately?

It was offline for a few days in total over the past month when I was doing a test to see if it will still work after being unpowered for some time. Other than that, it has been running 24/7.

Keep in mind my average speed is now just 34mb/s

[Christopher]

The Intel 40GBs are still mighty impressive to me.

[sergiu]

Just an idea that hit me... couldn't a MLC SSD be used as a SLC with half capacity after those rated P/E cycles have been depleted? Underneath, both SLC and MLC cells are similar, so binary values 00 and 01 could be easily mapped to 0 while 10 and 11 could be mapped as 1. This would allow higher margins for ADC circuit to interpret the values and could lead to less errors due to data retention, as most probably the charge is not lost completely, but just entered in an undefined zone. So, does anybody know if this would be technically possible? or there are some less known hardware limitations?

Later edit:
Found something interesting on subject: http://www.usenix.org/event/usenix09...rs/lee/lee.pdf

[B.A.T]

Todays update
Intel X25-M G1 80GB
79,7799 TiB
19273 hours
Reallocated sectors : 00
MWI=92 to 89
MD5 =OK
48.75 MiB/s on avg


Other news.
I've got an m4 waiting in my mailbox when I'm back home friday. I'm thinking of redoing the m4 test but with a better eye on retention.

[Christopher]

Can you guys continue the tests on the "dead" M4? Is it possible to use the official firmware update iso image burnt to disc to boot the computer and recognise the disk at all? When I accidentally deleted the firmware on my M4, I was able to force a re-write of the firmware to my M4 after a power cycling fix. I remember the parameter was -a to delete the firmware, then I power cycled it, and used parameter -u to force re-writing the firmware.

I don't think BAT has tried that per se; But word is it's extremely dead. It's extremely doubtful that the drive would show up to flash the FW if it can't be detected in the BIOS.

However I'd like to see another type of test, i.e. does the nand flash show significant volatility, once the MWI is exhausted? I'd like to have several SSDs of different brands/models to reach the MWI threshold then left powered off for a week or so and see if the contents are lost. By this we can know whether the "MWI Lasted" is indeed the best value to calculate the endurance.

At some point after the MWI is exhausted on a consumer drive, the NAND should retain data for a year. As more wear is incurred, that time shrinks down to just a day or two. As such, the current running drives are periodically disconnected for various periods of time to see whether data is retained. Obviously, powering off the drive for a year is out of the question, but all the drives have been taken down for retention at some point. It's not clear at what point retention goes from one year to a couple days.

My Mushkin Chronos Deluxe I was testing died before I got to the point where I was ready to disconnect the drive, but then the Chronos only lasted 50 days. The Intel 40GBs have been running the longest, and both are still able to retain data for quite some time.

[Anvil]

Kingston SSDNow 40GB (X25-V)

517.83TB Host writes
Reallocated sectors : 05 13
Available Reserved Space : E8 99

MD5 OK

33.79MiB/s on avg (~19 hours)

--

Corsair Force 3 120GB

The F3 is on an 48 hour data retention test (disconnected), about 30 hours left of the test.

[bluestang]

M225->Vertex Turbo 64GB Update:

856.19 TiB (941.39 TB) total
~2195 hrs (Torture), 3161 hrs (Power-On)
16667 Raw Wear
117.53 MB/s avg for the last 24.01 hours (on W7 x64)
MD5 every 20 loops on 1.59GB file = OK
C4-Erase Failure Block Count (Realloc Sectors) from 19 to 21.
C5-Read Failure Block Count (uncorrectable bit errors) from 5 to 6.
1=Bnk 6/Blk 2406 - Erase Failure C4
2=Bnk 3/Blk 3925 - Erase Failure C4
3=Bnk 0/Blk 1766 - Erase Failure C4
4=Bnk 0/Blk 829 - Erase Failure C4
5=Bnk 4/Blk 3191 - Erase Failure C4
6=Bnk 7/Blk 937 - Erase Failure C4
7=Bnk 7/Blk 1980 - Erase Failure C4
8=Bnk 7/Blk 442 - Erase Failure C4
9=Bnk 7/Blk 700 - Erase Failure C4
10=Bnk 2/Blk 1066 - Erase Failure C4
11=Bnk 7/Blck 85 - Erase Failure C4
12=Bnk 4/Blk 3192 - Erase Failure C4
13=Bnk 7/Blk 280 - Erase Failure C4
14=Bnk 3/Blk 2375 - Erase Failure C4
15=Bnk 7/Blk 768 - Erase Failure C4
16=Bnk 7/Blk 765 - Erase Failure C4
17=Bnk 7/Blk 182 - Erase Failure C4
18=Bnk 5/Blk 939 - Read Failure C5
19=Bnk 5/Blk 1115 - Read Failure C5
20=Bnk 5/Blk 1011 - Read Failure C5
21=Bnk 7/Blk 3549 - Read Failure C5
22=Bnk 7/Blk 3556 - Read Failure C5
23=Bnk 4/Blk 1961 - Erase Failure C4
24=Bnk 7/Blk 1862 - Erase Failure C4
25=Bnk 7/Blk 111 - Erase Failure C4
26=Bnk 5/Blk 902 - Read Failure C5
27=Bnk 7/Blk 560 - Erase Failure C4

[devsk]

quick question: Is there a Linux version of the tool you guys are using to beat the crap out of the drives? I can write a simple script using 'dd' and /dev/urandom but it would be a waste if we can't get the same data out for graphing purposes. smartctl works well in Linux but translating the attributes is really an art form...

[devsk]

M225->Vertex Turbo 64GB Update:

856.19 TiB (941.39 TB) total

6 more days at this rate for the big 1PB! I am so waiting for next Wed.

[bluestang]

Mostly likely more than that...I don't think I'll run it over the weekend while I'm not here at work.

[Anvil]

quick question: Is there a Linux version of the tool you guys are using to beat the crap out of the drives? I can write a simple script using 'dd' and /dev/urandom but it would be a waste if we can't get the same data out for graphing purposes. smartctl works well in Linux but translating the attributes is really an art form...

There is a Windows version, that's it!
(If there was 48hours in a day I'd do a Linux version as well but currently I'm spending all available time on the Windows version,
it looks like OS X could be next (as I've got a Mac) and on top of that I need to get up to speed with Linux)

Mostly likely more than that...I don't think I'll run it over the weekend while I'm not here at work.

Will it stay connected or are you doing a short retention test?

[Anvil]

...
Keep in mind my average speed is now just 34mb/s

My drive quickly drops in speed as well but it helps restarting the test every few days. (or once a week)

[bluestang]

Will it stay connected or are you doing a short retention test?

Connected, but idle.

[Christopher]

We have a new contender...

Samsung 830 64GB will be up for testing sometime early next week (I think - I mean, as soon as it gets here it's getting trashed).

The 64GBs do come with Norton Ghost and Batman: Arkham City (which I won't play because [A] it isn't Skyrim and [B] uses GFWL).

It's a good deal, and sequential writes are rated at 160MBs.

I expect the 64GB version to average around 110MB/s under endurance testing, but I can't wait to find out if I'm wrong. With practically no 64GB model reviews, I'm just guessing.

Here's hoping it will last a little longer than the Mushkin.

[Eternalightwith]

Been following from Day 1. Looking to buy SSD soon, but data so far doesn't translate to buy this, stay away from that.
Think it's about time to finalize charts and reach some type of consensus?
There is so much data and with new SSD lines coming out. I'm afraid all this hard work will get lost in the deluge of information already obtained.

Sincerely,

ET

[Christopher]

We can put the 64GB 830 and BATs new M4 head to head (should he choose to test the one waiting for him). On paper, the 830 and M4 fw 0009 stack up pretty well.

[johnw]

We have a new contender...

Samsung 830 64GB will be up for testing sometime early next week (I think - I mean, as soon as it gets here it's getting trashed).

Good choice! I was going to do one (or 60GB Intel 520), but you beat me to it. So I guess I'm waiting for 2012 to get an Intel 520.