SSD Write Endurance 25nm Vs 34nm

[kensiko]

RIP 830, this SSD will not be forgotten for a while!

Hey canthearu, how's the 840 doing?

[Computurd]

Damn Chris that is a crying shame, the things we do for Canada. We mourn the loss of the 830, she was, and is a beast. i would leave her on life support for a while, just in case. Dont pull the plug!

[canthearu]

Sad day for the 830!

However the 840 is still going well. Had it turned off for at least 15 minutes to do firmware update.

Maybe a small increase in performance, and nearing 2000 write cycles.

[teurorist]

sad to hear just followed this thread for a while to see what's with the Samsung 830
looking forward to some 840 pro tests

[Christopher]

The 830 is going to remained powered on but idling for some time. Miracles do sometimes happen, but in this case it went too far and too fast. Even a few minutes was enough to hose it up. I won't be pulling the plug anytime soon.

[gordo]

I'm a bit perplexed by the 830 result. Isn't it just as important to overall longevity when the drive stops being capable of being power cycled reliably rather than just when the NAND happens to be exhausted? Sounds to me like it might have quit petabytes ago in a more real-world-oriented testing scenario.

[canthearu]

True, but a more real life scenario would have taken more then 20 years to put that much data through the NAND.

The huge number is really for fun. The important take home message is that virtually all drives tested are perfectly fine during their rated service life, and that the service life is rather good at the moment. The 830 was some 10 times past it's service life when it died.

I might pull the 840 out after 500 TiB and let it sit around unpowered for a month as a retention test.

[B.A.T]

R.I.P 830. It was a long strugle to the end.

Here are todays update:

m4
2110.8663 TiB
8769 hours
Avg speed 68.44 MiB/s.
AD 183 to 165.
P/E 36477
C3 4206476112 to 209100477
01 3554123 to 9493822(1)
05 12288 to 61440
AA 6 to 30
C4 6 to 30
CE 90
MD5 OK.


Intel X25-E 64GB
1.49 PiB
5233-30=5203 hours
Reallocated sectors : 216 to 255 to 0 to 179,
Available Reserved space: 31 to 25
MWI= 90
MD5 =OK
74.76 MiB/s on avg


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

OCZ Vertex 1 120GB
Will start after an update to the latest firmware.

[canthearu]

Samsung 840 Day 38

Drive Hours: 904
ASU GiB Written: 327,768.56
Avg MB/s: 103.00 (237.01 hours)
MD5: OK

Wear Leveling Count (B1): 2626 raw (1 normalized)

Reallocated blocks (B3,05): 0 (0 sectors)
Failure count (B5, B6): 0 program, 0 erase
Uncorrectable Error Count: 0
ECC Error Rate (C3): 0

[kensiko]

Wow! Seems that the 840 TLC is a really good SSD. Prices are going down too.

[UranusFX]

840PRO & 840's NAND flash has pre-loaded function "read only".
They migrate someday or soon?

source http://akiba-pc.watch.impress.co.jp/.../sp_fline.html

[devsk]

840PRO & 840's NAND flash has pre-loaded function "read only".
They migrate someday or soon?

source http://akiba-pc.watch.impress.co.jp/.../sp_fline.html

Will be interesting to see what canthearu comes up with respect to this comment: "NAND flash bad block delayed occurs whenever you are using. Then, if there is more than the rate at which it will be bad block "bug" and read only if a certain percentage is reached, so that I will not be able to write. Our NAND flash memory has become such a specification."

[canthearu]

Will be interesting to see what canthearu comes up with respect to this comment: "NAND flash bad block delayed occurs whenever you are using. Then, if there is more than the rate at which it will be bad block "bug" and read only if a certain percentage is reached, so that I will not be able to write. Our NAND flash memory has become such a specification."

The translation is pretty damn rough ... I'm not too sure what to make of it.

Maybe the 840 will go into a read-only mode after running out of spare blocks .... maybe not, we will have to see. Right now, the 840 has been continuing to tick over (2810 cycles, 350 TiB written, no bad blocks, speed down to 102.72 average)

I could be here a while!

[canthearu]

First bad block on the 840.

Not due to a erase or program error ... must be a read error

[B.A.T]

Here are todays update:

m4
2156.4621 TiB
9128 hours
Avg speed 39.56 MiB/s.
AD 165 to 130.
P/E 37520
C3 209100477 to 4028072367
01 9493822 to 27935594(1)
05 61440 to 335872
AA 30 to 164(99 to 97)
C4 30 to 164
CE 90
MD5 OK.


Intel X25-E 64GB
1.58 PiB
5590-30=5560 hours
Reallocated sectors : 179 to 255 to 0 to 27,
Available Reserved space: 25 to 14
MWI= 90
MD5 =OK
72.18 MiB/s on avg


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

OCZ Vertex 1 120GB
Will start after an update to the latest firmware.

Been a little too exciting here. Some bad weather cut our power a couple of hours on saturday. Almost killed the m4 but not quite. Still going strong but at some reduced speed.

[canthearu]

Samsung 840 Day 47

Drive Hours: 1116
ASU GiB Written: 403,139.61 (393.69 TiB)
Avg MB/s: 101.97 (449.63 hours)
MD5: OK

Wear Leveling Count (B1): 3232 raw (1 normalized)

Reallocated blocks (B3,05): 12 (99 normalized)
Failure count (B5, B6): 0 program, 0 erase
Uncorrectable Error Count: 0
ECC Error Rate (C3): 0

Looks like the wear is starting to show. Looks like the drive is retiring blocks without being an explicit erase or program failure (possibly excessive raw BER from worn cells). Will be interesting how much wear it can take.

However, it seems to have beat the Vertex 4 at endurance testing.

[Aleks]

Hi guys,

Your tests are very interesting and quite useful for the community, so keep up the good work!

I've noticed that the average speed od the 830 is around 300 MB/s, and for the 840 around 100 MB/s. How is this average speed defined? Is it an average of the read and write? Why is it so much lower for the 840? Does this speed reflect the steady state performance in a proper way?
One more thing, how full are your SSDs during these tests? Do you fill them up entirely and then delete them entirely, or?

I'll be following closely to see what happens with the 840...

Thanks!

[canthearu]

I've noticed that the average speed od the 830 is around 300 MB/s, and for the 840 around 100 MB/s. How is this average speed defined? Is it an average of the read and write? Why is it so much lower for the 840? Does this speed reflect the steady state performance in a proper way?

The 830 is a MLC drive with 256GiB of NAND onboard.

The 840 (not 840 pro) is a TLC drive with 128GiB of NAND onboard.

The 830 has twice as many NAND chips that can be working, as well as a less complex programming procedure ... so it can be programmed quicker, and can handle more wear before failing.

The speeds generally reflect the lower end of steady state performance while under desktop loads. Real world desktop loads should be better because it isn't constant.

Server loads are a completely different matter and server loads generally have a much higher write amplification then these endurance tests. Hence steady state will be slower and more quickly wear the drive out.

One more thing, how full are your SSDs during these tests? Do you fill them up entirely and then delete them entirely, or?

I'll be following closely to see what happens with the 840...

The settings for the test are in the initial post for the 840 test.

The 840 has done pretty well in regards to endurance. As far as I can tell, the 840 is a perfectly adequate desktop drive, but perhaps a bit expensive for what you get right now.

Other TLC based drives may not be as lucky, as Samsung is currently the best at this endurance business. Will have to evaluate other TLC based drives if they ever come out.

[kensiko]

The 830 is a MLC drive with 256GiB of NAND onboard.

The 840 (not 840 pro) is a TLC drive with 128GiB of NAND onboard.

The 830 has twice as many NAND chips that can be working, as well as a less complex programming procedure ... so it can be programmed quicker, and can handle more wear before failing.

The speeds generally reflect the lower end of steady state performance while under desktop loads. Real world desktop loads should be better because it isn't constant.

Server loads are a completely different matter and server loads generally have a much higher write amplification then these endurance tests. Hence steady state will be slower and more quickly wear the drive out.



The settings for the test are in the initial post for the 840 test.

The 840 has done pretty well in regards to endurance. As far as I can tell, the 840 is a perfectly adequate desktop drive, but perhaps a bit expensive for what you get right now.

Other TLC based drives may not be as lucky, as Samsung is currently the best at this endurance business. Will have to evaluate other TLC based drives if they ever come out.

Well said. I chose the 840TLC for my mother in law, it should be pretty reliable. Can't believe how good is this Nand and controller! It's even better than some MLC SSD, as you said.

[Christopher]

After a long vacation, I'm back on the scene.

I was actually running another drive in the rig when the 830 died. That drive died two weeks later. What was it, you ask? It was a 120GB 840. I started it, but never did the first update because I was gone for a couple weeks, and lo! Canthearu busted out with one. So I just finished running it anyway. I won't drop any spoilers, but when Canthearu's dies, we'll compare notes. But his is doing much better it would seem.

[canthearu]

After a long vacation, I'm back on the scene.

I was actually running another drive in the rig when the 830 died. That drive died two weeks later. What was it, you ask? It was a 120GB 840. I started it, but never did the first update because I was gone for a couple weeks, and lo! Canthearu busted out with one. So I just finished running it anyway. I won't drop any spoilers, but when Canthearu's dies, we'll compare notes. But his is doing much better it would seem.

Not sure how much longer mine will hold out for ... maybe a week ... maybe 2 at this stage (has 27 bad blocks now)

[Christopher]

Not sure how much longer mine will hold out for ... maybe a week ... maybe 2 at this stage (has 27 bad blocks now)

Mine made it 8 days after the first read failure. Ended up with a bit less than 650 of them. After 27 read failures, it was about six days until it locked up the system.

[Anvil]

Kingston SSDNow 40GB (X25-V)

1460.63TB Host writes 47861954*32)
Reallocated sectors : 05 768 -
Available Reserved Space : E8 86 (86)
POH 12860
MD5 OK

31.68MiB/s on avg (~147 hours)

Not much to report, except that this sure takes time. (wouldn't surprise me at all if it makes 2 years of testing)

(it was idling a few days during Christmas)

--

[Raephen]

That Samsung 840 is going strong.

I recently got a 250GB one to replace the 128GB M4 in my laptop (which, after some testing, replaced the Agility 3 in my desktop, which in it's turn found it's way into my HTPC).

While I wasn't worried about it failing after a years of normal use, I find the numbers shown with the tested 840 very reassuring.

Keep up the good work!

[canthearu]

Samsung 840 Day 51

Drive Hours: 1216
ASU GiB Written: 437,361.39 (427.11 TiB)
Avg MB/s: 101.10 (549.80 hours)
MD5: OK

Wear Leveling Count (B1): 3508 raw (1 normalized)

Reallocated blocks (B3,05): 405 (86 normalized)
Failure count (B5, B6): 0 program, 0 erase
Uncorrectable Error Count: 0
ECC Error Rate (C3): 0

Wear is accelerating ... it won't be long now!