SSD Write Endurance 25nm Vs 34nm

[Christopher]

The MTRON FW is probably not very rugged. The other Pro 7000 is the system drive in the endurance rig.

I'll give it a shot, but I'd say it's probably very dead.

UPDATE
It can be detected (sort of), but causes system instability and it never shows up in Explorer. Also, there are two LEDs on the PCB. One is the activity light (which isn't on) and the other is probably a fault LED (which is lit).
a
-------------------
And the new Turbo has been working for 14 hours now without popping a bad block, so it's doing better than the last one.

[bulanula]

Typical controller failure.

Seems to be very common with SSDs of today ( especially SF drives ).

Maybe in the future SSDs will mature just like HDDs did ( they had problems in the early days too ) ???

[canthearu]

With sandforce drives, I think it is less outright failure of the controller, but either:

a) NAND chip failure
b) Firmware data tracking failure (probably most common)

Causes the controller to go into a panic mode.

Intel 320 drives have a special panic mode that you can secure erase from to recover a usable drive, sandforce drives don't.

[canthearu]

Ao1,

When my MWI indicator hits 90, I'll supply the smart log for my 520 so you can have relatively precise indicators for when the MWI changes.

I predict at least 250TB writes before MWI is exhausted on the 520 (but secretly hoping for 300TB odd or more)

[Christopher]

Yeah, I don't think the controller is failed; the firmware just got lost somewhere on the way. You can have the fastest drive in the world, but if the FW is junk, so is the whole drive.

Oh and SF's RAISE is supposed to prevent failure from one NAND device, so if that is killing SFs, then it lends more credence to my "RAISE sucks" theory.

[canthearu]

Yeah, I don't think the controller is failed; the firmware just got lost somewhere on the way. You can have the fastest drive in the world, but if the FW is junk, so is the whole drive.

Oh and SF's RAISE is supposed to prevent failure from one NAND device, so if that is killing SFs, then it lends more credence to my "RAISE sucks" theory.

No, RAISE cannot protect against an entire NAND device breaking, only from an individual NAND chip. (each device can have multiple NAND chips stacked together) There isn't enough spare space for an entire NAND device to die

[Christopher]

No, RAISE cannot protect against an entire NAND device breaking, only from an individual NAND chip. (each device can have multiple NAND chips stacked together) There isn't enough spare space for an entire NAND device to die

I use the term "devices" frequently; Note, this is not the whole package, which may consist of many dice (at least the way I use it), but rather another name for a die. A 512GB Samsung 830 has eight 8-die packages, or 64 8GB (64gbit) "devices" total for example.

Note -- I may just be using "device" wrong, but we were saying the same thing.

So in the case of a 120GB Toggle NAND equipped Mushkin Chronos Deluxe, it has 32gbit devices. So it could supposedly withstand the loss of one of those (4GB worth) through RAISE. Not a whole package, which might be several 4GB dies.

Sorry didn't mean to confuse -- we should create a master glossary of SSD terminology! Maybe I'll get started on that.

[Christopher]

Vertex Turbo 64GB Day 1

7,793.97 GB
82.11 MB/s

0 Erase Failure
0 Program Failure
0 Write Failure

27 Hours


The Vertex is using the same type and amount of static data as the old Turbo and 830.

[canthearu]

Patriot Torqx-2 64GB - Day 13

Drive hours: 311
GiB written: 16,722.24 (16.3303 TiB)
Avg MB/s: 17.08 MB/s

Bad blocks: 83
Wear cycle counter: 0/1092/1646

Intel 520 60GB - Day 4

Drive hours: 119
Avg MB/s: 75.09 MB/s

Host GB written (F1): 27,686.93 GiB (27.04 TiB , 885982 raw)
NAND writes (F9): 19,585 GiB (19.13 TiB)

Reallocated sectors (05): 0
Failure count (AB, AC): 0 program, 0 erase
Media Wearout Indicator (E9): 96

[Christopher]

Why am I having such bad luck with drives as of late? Hopefully, the new Turbo is the first in a new wave of more successful drives.

[canthearu]

Why am I having such bad luck with drives as of late? Hopefully, the new Turbo is the first in a new wave of more successful drives.

Because the drives don't have the name Intel in them :P

Well, the vertex turbo (barefoot) should be returned under warranty if you can and haven't already. Performed well below spec. Hopefully this one will give a good show. Unfortunately, the barefoot controller has always had a lot of weird and wonderful behaviours.

The 2nd MTRON drive was likely just bad luck or bad programming on the effort of the drive's creators.

If I was going to add a new drive to the burning, it would have to be an ocz petrol or a drive with 20nm NAND. A corsiar performance pro would also be a good drive, but they are only 128gig and the pricing is kinda scary for something that is going to be endurance tested to death.

[Christopher]

Because the drives don't have the name Intel in them :P

Well, the vertex turbo (barefoot) should be returned under warranty if you can and haven't already. Performed well below spec. Hopefully this one will give a good show. Unfortunately, the barefoot controller has always had a lot of weird and wonderful behaviours.

The 2nd MTRON drive was likely just bad luck or bad programming on the effort of the drive's creators.

If I was going to add a new drive to the burning, it would have to be an ocz petrol or a drive with 20nm NAND. A corsiar performance pro would also be a good drive, but they are only 128gig and the pricing is kinda scary for something that is going to be endurance tested to death.

Well, I email my way into some FW updates for the MTRONs. I'm hoping I can flash the PRO 7025 back into business.

The Indilinxes are strange and wonderful sometimes, aren't they? The old one was finding at least one bad block every day, and after 40 hours the new one is still clean. That gives hope that the future is looking better for the new one. It is my hope that the Turbo can go where no 64GB has gone before, and perhaps I can learn something on the way.

UPDATE
They didn't send me the MTRON fw, just the flashers for the M-Classes. So that's a no go. If anyone knows where I can get MTRON firmware and how I can flash it, that would be helpful.

[canthearu]

It is my hope that the Turbo can go where no 64GB has gone before, and perhaps I can learn something on the way.

In theory, the Turbo using 51nm NAND should be able outclass all of more modern devices (at least in this endurance testing). The M255 went pretty well, but I was surprised it didn't hang in their a bit longer.

However, I'm putting my money on my Intel 520. Intel NAND has always shown itself to be highly durable, and with a supposedly debugged controller, it could go a damn long way. Of course, realisticly, it is still limited by the 25nm process technology that ended up being the undoing of the brilliant run of the Intel 320 40gig.

[Christopher]

I think the old turbo just had a bad batch of flash in it.

[Christopher]

Vertex Turbo 64GB Day 2

12575.45 GB
82.26 MB/s

MWI 96

0 Erase Failure
0 Program Failure
0 Write Failure

Avg Erase Count
226

43 Hours

[canthearu]

Patriot Torqx-2 64GB - Day 14

Drive hours: 336
GiB written: 18,153.33 (17.7281 TiB)
Avg MB/s: 16.87 MB/s

Bad blocks: 83
Wear cycle counter: 0/1180/1775

Intel 520 60GB - Day 5

Drive hours: 144
Avg MB/s: 75.21 MB/s

Host GB written (F1): 34,212.25 GiB (33.41 TiB , 1094792 raw)
NAND writes (F9): 24,199 GiB (23.63 TiB)

Reallocated sectors (05): 0
Failure count (AB, AC): 0 program, 0 erase
Media Wearout Indicator (E9): 95

[Ao1]

So it seems that the 520 performs very much like any other SF2xx based drive. Not too surprising as the MWI is based on theoretical P/E cycle capability, which will be the same value depending on nand geometry. Specially binned nand only starts to shine once the theoretical MWI has expired, but realistically who is going to keep on using their SSD once the MWI gets below 0

It really looks like Intel have done very little to the SF f/w. Turning off raise was already an option. All they seem to have done is debugged the f/w and modified the SMART reporting attributes.

[Anvil]

Kingston SSDNow 40GB (X25-V)

754.75TB Host writes
Reallocated sectors : 05 27 up 2 from last update!
Available Reserved Space : E8 99
POH 6783
MD5 OK

33.26MiB/s on avg (~335 hours)

So, reallocated sectors are increasing, could be a trend.

[canthearu]

So it seems that the 520 performs very much like any other SF2xx based drive. Not too surprising as the MWI is based on theoretical P/E cycle capability, which will be the same value depending on nand geometry. Specially binned nand only starts to shine once the theoretical MWI has expired, but realistically who is going to keep on using their SSD once the MWI gets below 0

It really looks like Intel have done very little to the SF f/w. Turning off raise was already an option. All they seem to have done is debugged the f/w and modified the SMART reporting attributes.

Well, I didn't expect any really huge differences between the intel 520 and other sandforce drives.

What will be interesting to see is:
a) Where MWI=1 happens, which I should be able to predict soon.
b) If the device dies like other sandforce drives tested here do, at about 2x of MWI in a horrible flaming insta-death, or if it manages to hang on like the m4 does.

Haven't seen a 60gig sandforce drive hit 700TiB like we have with the m4, would be nice to see if it were possible!

[Anvil]

Most (if not all) 2281 based SandForce drives tested earlier had the flawed FW so starting off with a working firmware might help getting it up there with the m4.
(and then there is the difference on RAW vs HOST which in theory should put the SF drives ahead of other non-compressing controllers)

[Anvil]

Endurance_cr_20120301.png

[Christopher]

So, the 520 just goes to MWI = 1? The SF2281s get to 10, but as blocks are replaced they will get to 1 eventually. At MWI = 1, the drive (any 2281 besides the 520) is supposed to be read-only. How often this occurs in real-life is not known.

[canthearu]

So, the 520 just goes to MWI = 1? The SF2281s get to 10, but as blocks are replaced they will get to 1 eventually. At MWI = 1, the drive (any 2281 besides the 520) is supposed to be read-only. How often this occurs in real-life is not known.

Hmmm, has any drive entered a readonly state though, is it documented in any technical manual?

Does such a state actually exist for SSDs, or is it an urban legend created by the fact that drives, once they reach their P/E limits, still have reasonable read retention?

I know that a normal windows install certainly wouldn't boot from a read-only volume.

But yeah, the 520 works exactly the same as other intel drives with respect to the normalized value of media wearout indicator attribute E9 (my raw value for E9 is always 0)

[Christopher]

The MWI on SF is a normal MWI from 100 to 10. Then it becomes a
available reserved space counter. According to my research, when MWI is 1, the drive is read only.

[Christopher]

I should note that in this test, the chances of a SF becoming RO are about the same as me becoming the Prime Minister of Japan. It's impossible and inconceivable. The chances of any drive becoming RO are slim at best. Maybe two or three of all drives tested in a year.