SSD Write Endurance 25nm Vs 34nm

[B.A.T]

Todays update.
Kingston V+100
289.7360 TiB
1360 hours
Avg speed 25.40 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
128,3653 TiB
19579 hours
Reallocated sectors : 00
MWI=26 to 21
MD5 =OK
51.35 MiB/s on avg


m4
53.5538 TiB
195 hours
Avg speed 81.00 MiB/s.
AD gone from 73 to 69.
P/E 933.
MD5 OK.
Reallocated sectors : 00

[Anvil]

@BAT
So, December the 25th for the Intel X25-M G1 and New Years day for the M4
(according to SSDLife)
I need to install SSDLife onto the current test-rig, it's currently sitting on the AMD rig I used until late this summer.

--

The Intel just got past 550TB and the F3 will be past 700TiB by tomorrow morning.

Kingston SSDNow 40GB (X25-V)

550.87TB Host writes
Reallocated sectors : 05 14
Available Reserved Space : E8 99
POH 4996
MD5 OK

32.05MiB/s on avg (~72 hours)

--

Corsair Force 3 120GB

01 89/50 (Raw read error rate)
05 2 (Retired Block count)
B1 50 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 535621 (Raw writes) ->523TiB
F1 712916 (Host writes) ->696TiB

MD5 OK

102.75MiB/s on avg (~72 hours)

power on hours : 2032

(for some reason both drives have had a slowdown but they are steadily increasing on avg throughput)

--

@bluestang

Looks like I'll have to update the summary!

[B.A.T]

@BAT
So, December the 25th for the Intel X25-M G1 and New Years day for the M4
(according to SSDLife)
I need to install SSDLife onto the current test-rig, it's currently sitting on the AMD rig I used until late this summer.

Looks like an exciting holiday this year

[Anvil]

I bet that you can disregard those dates , still it can be exciting, anything can happen of course.

[B.A.T]

I planned on starting my retention test from mwi=1%. Stop every 100GB and try like you did with the X25-V.

[Christopher]

Those Intel's 40GBs are a model of good behavior. Maybe if you're writing slow and steady, you can do so almost indefinitely.

[Anvil]

The 40GB Intels are writing as much/fast as they can, they are slow though

I have considered making an option for "slowing down"/controlling writes.
There are lot'a of possible options to consider though, like setting a fixed number of MB/s or a fixed number of GiB per hour or TiB per day and so the options will make it possible to have a few pauses.
These could be useful options but you can of course run the test just like before, w/o pauses.

[Christopher]

One mystery solved --

The Samsung Toggle NAND uses 8K Pages, 1MiB blocks. the reallocated sector count really are sectors. Right now, the Samsung has 10 replaced blocks and reports 20480 reallocated sectors.

512 bytes * 20480 = 10485760 bytes, or 10MiB. Each block is 1MiB, so 10MiB of blocks have been flagged as bad and replaced.

[Anvil]

Kingston SSDNow 40GB (X25-V)

552.24TB Host writes
Reallocated sectors : 05 14
Available Reserved Space : E8 99
POH 5008
MD5 OK

32.16MiB/s on avg (~84 hours)

--

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 49 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 539011 (Raw writes) ->526TiB
F1 717426 (Host writes) ->701TiB

MD5 OK

103.30MiB/s on avg (~83 hours)

power on hours : 2044

The Force 3 is now idling and will do so for a few more hours in wait for a data retention test.

--

[Christopher]

Samsung 830 64GB Update, Day 9
FW:CXM01B1Q

GiB written:
80990.91

Avg MB/s
105.43

Per-day average:
8990GB

PE Cycles
129

Reallocated Sector Count
20480 (10 blocks)
8K pages, 1MiB blocks

219 hours





The min free space is set to 12GB. If you look at the ASU shot above you'll see the instant MBs in the 60MBs range, but the average hasn't gone down too much (yet)


Measured WA -- 1.05

Since moving to 12GB min free from 16GB, the drive was only able to write 8284GB, about 700GB less than the previous daily average.

[bluestang]

My Last Official M225->Vertex Turbo 64GB Screenshot and Stats:



879.17 TiB (966.66 TB) total.
~2245 hrs (Torture), 3304 hrs (Power-On).
17092 Avg. Raw Wear.
B8-Initial Bad Block Count at 8
C3-Program Failure Block Count at 2
C4-Erase Failure Block Count (Reallocations) at 25.
C5-Read Failure Block Count (uncorrectable bit errors) at 27.

On 12/08/11 Drive threw 16 more C5 for a total of 43.
So a total (B8+C3+C4+C5) of 78 Bad Blocks.

I was able to write down SSD Life total of 901479.9 GiB (880.35 TiB / 967.96 TB) before it died on 12/08/11.

This drive managed 116.48 MB/s over the entire endurance test.

*** M225 -- R.I.P. December 8, 2011 at 968 TB of Writes ***

[Anvil]

Thanks for all your work on this bluestang!
(first post updated)

--

The Corsair Force 3 has been disconnected for about 6 hours, another 48 hours and it will be back online.

[Christopher]

Bluestang,

Are you going to put the other M225 into service? Or put it next to the expired one in the SSD Hall of Fame?

[johnw]

The Samsung Toggle NAND uses 8K Pages, 1MiB blocks. the reallocated sector count really are sectors. Right now, the Samsung has 10 replaced blocks and reports 20480 reallocated sectors.

512 bytes * 20480 = 10485760 bytes, or 10MiB. Each block is 1MiB, so 10MiB of blocks have been flagged as bad and replaced.

Did you ever see the RAW value increment by less than 4096? In the data you posted, I thought it went by at least 4096 increments.

4096 * 512 = 2 MiB

[B.A.T]

Todays update.
Kingston V+100
291.9157 TiB
1385 hours
Avg speed 25.39 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
132,7932 TiB
19604 hours
Reallocated sectors : 00
MWI=21 to 15
MD5 =OK
51.32 MiB/s on avg


m4
60.5541 TiB
220 hours
Avg speed 81.11 MiB/s.
AD gone from 69 to 65.
P/E 1055.
MD5 OK.
Reallocated sectors : 00


@Bluestang
R.I.P M225

You did a good job babysitting it. To bad it didn't last all the way to 1 PiB.

[Christopher]

Did you ever see the RAW value increment by less than 4096? In the data you posted, I thought it went by at least 4096 increments.

4096 * 512 = 2 MiB

Yes. It increments in 2048. That's what I thought as well initially, but it was just because two blocks went out at once. Samsung's web page lists the 8K pages 1MiB block size. If you look at the smart data from today, the number of replaced blocks is listed as 10. Still, I'm a little confused by the erase fail and total runtime bad block counts. It does seem like one doesn't increment without the other.

[Anvil]

Kingston SSDNow 40GB (X25-V)

555.03TB Host writes
Reallocated sectors : 05 14
Available Reserved Space : E8 99
POH 5032
MD5 OK

33.54MiB/s on avg (~18 hours)

--

Corsair Force 3 120GB

It has been disconnected for 18 hours.

[bluestang]

Thanks for all your work on this bluestang!

Thanks! ...and same to you since this would not be as possible without ASU.

Bluestang,

Are you going to put the other M225 into service? Or put it next to the expired one in the SSD Hall of Fame?

Thanks! ...I might mount this one on a plaque and hang it in my office.

Not sure yet on running the other one, if I do it will be with FW 1.7.

@Bluestang
R.I.P M225
You did a good job babysitting it. To bad it didn't last all the way to 1 PiB.

Thanks! ...and yeah, I'm a little upset with not reaching 1 PB.



If I do put the other M225 into the test, it won't be till after the New Year.

[Christopher]

So I modified the Min GB Free the other day. It took a while, but the 830 is now unable recover its speed. I ran the Magician Optimization 6 times, but no increase in speed. I'm going to have to SE.

I was using 18.3GB of static data, 16GB min Free, so that leaves 26GB per loop. If I move it down to 12GB Min Free, it can't keep its speed up, and worse, can't be fixed.



Before SEing the drive, I tried a few things. The Samsung Magician's performance optimization is worthless. Those numbers above are abyssmal, and it all started when I lowered the min free space to 12GB. It took about a day, but the controller seems to have backed itself into a corner here. Something is wrong though... it's not fixable with traditional means, so I'm going to be forced to SE it. Just deleting the volume didn't do much. Large sequential writes didn't do much. Samsung's software didn't do anything, even when run 8 times at about 4 minutes each.

EDIT


This is after a SE. It looks pretty sad.

[Christopher]

Major update. For real this time. Sort of...

but first, today's update:


Samsung 830 64GB Update, Day 10
FW:CXM01B1Q

GiB written:
87061.03

Actual Writes:
87476.6

Avg MB/s
81.54 down from 105MBs

Per-day average:
8706GB

PE Cycles
138

Reallocated Sector Count
20480 (10 blocks)
8K pages, 1MiB blocks

243 hours





Okay. Notice the 138 PE cycles?

From the reboot, or the SE, things changed. You can see the before and after SE CDM numbers.

Now look at this:


The PE cycles are now at 5444. That's 1/1/5444.

So the PE cycle/wear leveling does work properly and can be used as a life indicator.

But I can't get performance back to near-new levels to save my life.

[johnw]

The PE cycles are now at 5444. That's 1/1/5444.

That gives a write amplification of 4.00 : 5444*64*1024^3 / ( 85.0205*1024^4)

The 470 was about 5.0. It seems Samsung may have tweaked the firmware algorithms slightly.

[Christopher]

That gives a write amplification of 4.00 : 5444*64*1024^3 / ( 85.0205*1024^4)

The 470 was about 5.0. It seems Samsung may have tweaked the firmware algorithms slightly.

Truth it, I don't put a lot of faith in either number (100+ PE cycles, then a jump to 5444). Looking at the total writes to the total count of write sectors, you don't see much WA. Around 1.06 when I calcluated yesterday.

I'm more concerned that a SE didn't fix the profound performance drop.

[Christopher]

One possible, if not probable, explaination is several of the cells are borderline and the controller is having a hard time dealing. Maybe the Samsung SE isn't real SE. Who knows? I deleted the volume first and that made no performance difference. See the a couple posts up for before and after pics.

[johnw]

Truth it, I don't put a lot of faith in either number (100+ PE cycles, then a jump to 5444). Looking at the total writes to the total count of write sectors, you don't see much WA. Around 1.06 when I calcluated yesterday.

No, the WA is definitely not 1.06. The 5444 erase cycles makes sense (and 100 is definitely wrong) since the normalized value has already reached threshold (it must have done at either 3000 or 5000).

[Sirakuz]

So I modified the Min GB Free the other day. It took a while, but the 830 is now unable recover its speed. I ran the Magician Optimization 6 times, but no increase in speed. I'm going to have to SE....

Try to reboot..
Samsung 470 at for example, this behavior is a known bug, similar to the transition mode of the controller in sata 1.5Gbps