SSD Write Endurance 25nm Vs 34nm

[Anvil]

I was thinking the RMA word as well, they might be interested in finding what's gone wrong with this particular drive.

[Christopher]

I was thinking the RMA word as well, they might be interested in finding what's gone wrong with this particular drive.

I think I may try and contact Samsung support, but I don't know if I have the patience to explain the situation to whoever the hell is on the other end. But I think you could break every Samsung drive conceivably, under the right conditions, so I'll try anyway. I doubt anyone other than me has encountered this since they'd be yanking them out of Dells and Lenovos overnight if this had happened to too many other people.

It's clearly not some kind of throttling, as reads are more affected than writes, though the writes are pretty bad too.

As such, I believe it to be a firmware bug, something that could only be fixed with a re flash. Possibly not even a firmware update could stop it.

[johnw]

I think I may try and contact Samsung support, but I don't know if I have the patience to explain the situation to whoever the hell is on the other end. But I think you could break every Samsung drive conceivably, under the right conditions, so I'll try anyway

Maybe don't mention that MWI is exhausted until you've explained everything else and let them respond to the other issues. I suspect once you mention that MWI is exhausted (or how much you have written), the support person will blame everything on that, even though there is clearly a firmware bug.

[B.A.T]

Todays update:
Kingston V+100
302.1915 TiB
1503 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
152,7166 TiB
19722 hours
Reallocated sectors : 00
MWI=160 to 159.
MD5 =OK
50.03 MiB/s on avg


m4
93.4232 TiB
338 hours
Avg speed 81.16 MiB/s.
AD gone from 52 to 46.
P/E 1627.
MD5 OK.
Reallocated sectors : 00

[Christopher]

Maybe don't mention that MWI is exhausted until you've explained everything else and let them respond to the other issues. I suspect once you mention that MWI is exhausted (or how much you have written), the support person will blame everything on that, even though there is clearly a firmware bug.

Yes. That's kinda it in a nutshell. That one little fact will over shadow everything.

"...uummmm is 6500PE cycles a lot?"

"Really? You don't say."

"You think the fact that I bought the drive just to destroy it is relevant? We'll just have to agree to disagree."

[Anvil]

Kingston SSDNow 40GB (X25-V)

568.61TB Host writes
Reallocated sectors : 05 14
Available Reserved Space : E8 99
POH 5150
MD5 OK

33.85MiB/s on avg (~52 hours)

--

Corsair Force 3 120GB

01 94/50 (Raw read error rate)
05 2 (Retired Block count)
B1 45 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 554109 (Raw writes) ->541TiB
F1 737553 (Host writes) ->720TiB

MD5 OK

106.61MiB/s on avg (~52 hours)

power on hours : 2132

--

[Christopher]

Samsung 830 64GB Update, Day 15
FW:CXM01B1Q

GiB written:
111566.24

Actual Writes:
112445.4

Avg MB/s
62.27MBs

PE Cycles
6672

Reallocated Sector Count
24576 (12 blocks)
8K pages, 1MiB blocks

358 hours




I have a different view of the Samsung's performance; It could just be dead, but instead it works fine and I don't mind leaving it unattended over the holidays.

[B.A.T]

Todays update:
Kingston V+100
304.3754 TiB
1528 hours
Avg speed 25.37 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
156,5384 TiB
19747 hours
Reallocated sectors : 00
MWI=159 to 157.
MD5 =OK
49.40 MiB/s on avg


m4
100.4250 TiB
363 hours
Avg speed 81.14 MiB/s.
AD gone from 46 to 42.
P/E 1749.
MD5 OK.
Reallocated sectors : 00

[devsk]

Corsair Force 3 120GB

01 94/50 (Raw read error rate)
05 2 (Retired Block count)
B1 45 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 554109 (Raw writes) ->541TiB
F1 737553 (Host writes) ->720TiB

MD5 OK

106.61MiB/s on avg (~52 hours)

power on hours : 2132

--

Looks like F3 is new horse to bet on to break the 1PB barrier!

[minpayne]

Speaking of the X25-E... here's my 32GB one with the most host writes. MWI of 90 still at 580TB written:

The C0 Unsafe Shutdown Count is increasing each time I do a soft (normal) shutdown of Windows. Is the firmware of X25-E flawed or supposed to increment like this?

[Christopher]

My X25-E doesnt do that, but maybe I just havent noticed since it only has about 300GB host writes. But it
does kinda seem like the X25-Es Ive seen have had a lot of unsafe shutdown counts.

[B.A.T]

Todays update:
Kingston V+100
306.6465 TiB
1554 hours
Avg speed 25.38 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
160,5192 TiB
19773 hours
Reallocated sectors : 00
MWI=157 to 156.
MD5 =OK
48.90 MiB/s on avg


m4
107.6948 TiB
389 hours
Avg speed 81.16 MiB/s.
AD gone from 42 to 38.
P/E 1876.
MD5 OK.
Reallocated sectors : 00

[minpayne]

My X25-E doesnt do that, but maybe I just havent noticed since it only has about 300GB host writes. But it
does kinda seem like the X25-Es Ive seen have had a lot of unsafe shutdown counts.

If you still have access to your X25-E, could you please post a screenshot of the SMART readings? Mine looks like this:

[Anvil]

I've got a few E's, will check later tonight, I've seen the issue but haven't checked those drives.

--

Kingston SSDNow 40GB (X25-V)

573.97TB Host writes
Reallocated sectors : 05 14
Available Reserved Space : E8 99
POH 5197
MD5 OK

33.30MiB/s on avg (~100 hours)

--

Corsair Force 3 120GB

01 89/50 (Raw read error rate)
05 2 (Retired Block count)
B1 44 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 563571 (Raw writes) ->550TiB
F1 750143 (Host writes) ->733TiB

MD5 OK

106.69MiB/s on avg (~25 hours)

power on hours : 2179

Somehow I've managed to hit the Stop button while it was running (probably while hitting a button to get the screen out of screen-saver mode) and so it was idling for about 10 hours without me noticing. (next time I'll use the mouse )

I'm pretty confident that this is the first drive that will get past 1PB, it will take another 30 days + 4-5 days of retention testing so about 5 weeks totally.

[minpayne]

Holy ****! Anvil what have you done in the ASU, I think I just ****ed up my 3rd M4 512G (in my 3rd computer)! 4th loop was the death loop (46% application fill), and the computer halted with a **** load of write error displayed in ASU. Is it a problem with the quality control of the flash IC of the 512G model, or is there a flaw of the firmware that doesn't do wear-leveling properly?! I don't think I can be this unlucky to have three (3) bad M4 512G in a row!!!

[Anvil]

There is just one version of the Endurance test that is publicly available and we are all using it.
There is nothing special with the 46% or any of the other compression levels, it's just a write pattern that results in a that particular compression ratio.

It does sound weird that several 512GB are failing, they should behave just like any other m4 drive.

Just for the record, what settings are you using, static file size, min free space etc.

[minpayne]

There is just one version of the Endurance test that is publicly available and we are all using it.
There is nothing special with the 46% or any of the other compression levels, it's just a write pattern that results in a that particular compression ratio.

It does sound weird that several 512GB are failing, they should behave just like any other m4 drive.

Just for the record, what settings are you using, static file size, min free space etc.

The page size of the 256GB/512GB M4 is 8k, while that of the 64GB/128GB M4 is 4k. I have a speculation that the firmware/controller of M4 isn't doing a great job there taking care of the 8k page sized models, which might lead to unexpected wears on certain blocks. Also, it might be a reasonable assumption that the cutting-edge capacity (currently 512GB or 600GB for a 2.5" SATA SSD) always comes with a higher failure rate, just like the mechanical hard drives (e.g. now 4TB drives shouldn't be that reliable at the moment because the manufacturing process isn't that mature yet).

Here I list some details of these three samples of M4 512GB, each on a different machine (I'll try to recall as much as I can and please let me know what you'd like to know). Sample 1 was from an early batch. Sample 2 and 3 were from the same batch at a later time.

Sample 1:

Model: M4 512GB
Factory firmware: 0001
Testing firmware: 0009
Motherboard: ThinkPad X220 SATA III (Intel native). This machine didn't have any problem with Intel X25-E 64GB.
Settings: all default
Static file size: around 380GB
Min free space: 12GB
Avg speed: 100MB/s
Error description: At the beginning I really liked this SSD. Looked very stable. Updated the firmware to 0002 before I installed OS on it. Then one day I started AIDA64 and ran the looped linear read test in the disk benchmark, and the system stability test with disk stress at the same time. The laptop halted within minutes each time I did this. The SSD was not detected by BIOS after reboot and each time I had to use the power cycling method to recover it. Then I updated the BIOS of the laptop from 1.19 to 1.24 and the laptop no longer halted under such tests, and I thought it was just all the laptop's fault. Later on, one day when I was installing a virtual machine, I got 0xF4 BSOD several times, each time upon a burst of writes, and none of the crash dump was successfully written to the SSD, indicating that the SSD went offline. Then I updated the firmware of the SSD to 0009 and seemed to have cured the BSOD problem. I lost trust in the SSD, so I decided to run ASU's Endurance testing on it. Unfortunately with around 380GB static file and the default settings (0-fill, 12GB min free etc), the SSD could be easily forced to go offline within the first loop. The "05 Reallocated Sector Count" and the "C4 Reallocation Event Count" increased crazily. Then I ran "diskpart clean" and used Windows 7 to do a *slow* format (hence 0 static file), and ASU wasn't able to repeat any problem for several loops, which I guess was because Windows marked the bad sectors. Then I did a quick format but ASU wasn't able to repeat any problem for several loops again, which I guess was because Windows copied the information of marked bad sectors. Then I used WinHex to do a DoD fill of the SSD, and it was forced offline easily upon the first loop. Then ASU was able to repeat the problem easily within 1 loop, until several times of failure the SSD was no longer recognized by BIOS regardless of the power cycling method for recovery.

Sample 2:

Model: M4 512GB
Factory firmware: 0002
Testing firmware: 0009
Motherboard: EVGA X58 E760 SATA II (Intel native). This machine didn't have any problem with Intel 320 600GB.
Settings: all default
Static file size: around 320GB
Min free space: 12GB
Avg speed: 80MB/s
Error description: At the beginning the SSD seemed to be stable, until I decided to upgrade the firmware from 0002 to 0009. I noticed 0xF4 BSOD upon bursts of read/write, e.g. verifying the local cache of a Steam game. Then I started AIDA64 and ran the looped linear read test and the system stability test with disk stress at the same time, and I could easily repeat the BSOD within tens of minutes each time. Then I ran ASU's Endurance testing with default settings, and the BSOD could be reproduced very quickly (at the very early stage of the 1st loop). The "05 Reallocated Sector Count" and the "C4 Reallocation Event Count" both stayed at 0.

Sample 3:

Model: M4 512GB
Factory firmware: 0002
Testing firmware: 0002
Motherboard: ASUS P8H67-I SATA III (Intel native)
Settings: all default, except 46% application fill and 1GB min free space
Static file size: around 120GB
Min free space: 1GB
Avg speed: 119MB/s
Error description: This sample didn't seem to be that troublesome but since my partner has been using it and never stressed it like I did I guess it just didn't have the chance to show any problem. I ran ASU's Endurance testing for around 3 loops of 0-fill with 12GB min free space and didn't notice any problem. Then I ran it for another night with 46% application fill with 1GB min free space and the computer lost video signal (and according to the log the computer halted during the 4th loop). At this stage the static data was around 310GB (because ASU didn't have a chance to clean the test files yet) and when I tried to continue running ASU's Endurance testing I got write error straight away (with very slow Avg speed displayed). I clicked "Stop" and all test files were deleted. Then ASU's Endurance testing wasn't able to reproduce the problem for another 3 loops. Still running more tests to see what happens next. The "05 Reallocated Sector Count" ,the "C4 Reallocation Event Count" and the "CA MWI" are all still at 0 at the moment.

[Anvil]

The m4 really should have no problem running the Endurance test, the smaller 64GB did remarkably well and looks to do just as well in round 2.

The amount of static data should be relative to the capacity and so 350-400GB would be OK but 12GB as a minimum of free space would be on the low side for the 256/512GB capacities, you might consider increasing that to at least 16-32GB as that would be more realistic for a drive on the large side.
I keep at least 20-30GB of free space on 256GB drives and that is actually out of my comfort zone.

The relatively slow avg speed might be a pointer to that the drive is struggling with the work-load, I had expected it to be 2x the speed of the m4 64GB.
(the first m4 64GB had an avg speed of 88MiB/s)

Anyways, it should not fail this test in just a few loops!
Personally I'd probably approach the test carefully or stay away from testing that particular drive/model as multiple drives have shown issues with the workload.

So, you could try increasing the minimum free space to 20-30GB as that would be more life-like for client workloads, 1GB is much to low, even for a small drive.
If I can arrange to take one of my C300 or m4 256GB drives out of "production" I'll do a short test sometime during the Christmas holidays.

[B.A.T]

Todays update:
Kingston V+100
308.6083 TiB
1577 hours
Avg speed 25.37 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


Intel X25-M G1 80GB
163,9794 TiB
19796 hours
Reallocated sectors : 00
MWI=156 to 155.
MD5 =OK
48.55 MiB/s on avg


m4
113.9955 TiB
412 hours
Avg speed 81.18 MiB/s.
AD gone from 38 to 34.
P/E 1986.
MD5 OK.
Reallocated sectors : 00

[bulanula]

Any news of those charts, the C300 and Vapor ?

[Christopher]

I don't have access to my X25-E at the moment, and the only pics I have access to are some benchmarks.

But the power count/unsafe shutdown counts are not 1:1, but there are quite a few. Mine only has around 300GB on it though.

[Anvil]

I checked my X25-E drives and all have unsafe shutdowns, looks like 20-30% of the shutdowns are unsafe.

I performed a few tests last night and when Unsafe shutdown increased, the Power Cycle Count did not.
(I had expected Power Cycle Count to increase on every power Off and it didn't)

[Christopher]

I checked my X25-E drives and all have unsafe shutdowns, looks like 20-30% of the shutdowns are unsafe.

I performed a few tests last night and when Unsafe shutdown increased, the Power Cycle Count did not.
(I had expected Power Cycle Count to increase on every power Off and it didn't)

But at the same time, it seems as though most drives don't act this way... the Intel's power cycle may = 1 power on + 1 ATA shutdown. If the unsafe shutdown increases, power cycle doesn't. Other drives just assume that the drive has to be on to get turned off

So the true number of power cycles for an Intel is Power Cycle Count + Unsafe Shutdown.


I believe that if you enter the bios/uefi then exit, unsafe shutdown count will increase. I enter the BIOS and UEFI all the time, so some drives are worse off than others. For example, I had been reflashing and upgrading the FW on all of my Indilinx drives, so I would remove all of the other drives in the system, as it could take a lot of reboots and UEFI changes and I'd prefer not to have other drives in the system when doing FW updates. I don't think the unsafe shutdown counts are problematic, but it's probably best to avoid them if you can do so without much trouble. Basically, anything other than a soft-shutdown with a modern OS would result in an unsafe shutdown count.

All of the drive failures I've experienced have occurred in Windows, not on a restart. The sample size is pretty small though.

[minpayne]

Strange, now I cannot reproduce any problem on the 3rd sample of the M4 512GB. Today when I woke up it was the first time I saw the MWI lost virginity (no reallocated sector though, strange? Is it because MWI is calculated according to wear-leveling count and the P/E of the flash?):

[minpayne]

OK never mind, I've figured it out myself. By looking into the figures in the OP of this thread, it is very obvious that the MWI is simply calculated by the P/E spec of the flash. For M4, the P/E spec is 3000 cycles, and so the MWI of the M4 64GB should last for 59.5GB * 3000 / 1024 = 174.3 TB, which is pretty close to that of the test result of the first M4 64GB sample in this thread (and also the trend of the 2nd sample). It also works the same for the Samsung 470 64GB. I'm just a bit slow to follow you guys here