SSD Write Endurance 25nm Vs 34nm

[Vapor]

Just 2.25TiB (after static data, etc) into the test, Kingston V+100's WA is already up to an apparent 4.17x

For reference, after 8.59TiB host writes and with more precise measurement, the Samsung 470 only had an apparent WA of 3.53x...and it ended up at 5.11x.

The M225->Vertex Turbo also had its WA initially climb then stabilize. Basically, wouldn't be surprised to see the V+100's WA go even higher then stabilize over the next few days.

(it is entirely possible my normal method of deriving PE cycles from MWI depletion is also not applicable here)

Current expected MWI Exhaustion is just 74.3TiB host writes, 1276 normalized (64GB) writes, and 11.4 write days.

Preliminarily, it looks like we may have a drive similar to the Samsung 470

[Anvil]

The WD ended up being a total failure (in this regard) it works just fine for normal tasks.

The latest session ended up failing due to the SF2 controller issue, the Endurance app was still running , everything else had shut down, I was unable to save the screenshot though.

So, 1 hour of writes ended up in an average throughput of 21MiB/s.
The drive is capable of about ~140MB/s during normal large block IO.

This is a screenshot from one of the other sessions, it is just not suitable for this test. (I wasn't planning on using it, just checking the performance)

WDC SSC-D0064SC-2500_64GB_endurance-20110823-1931.png

So, compared to this B.A.T's Kingston is doing just fine.

[B.A.T]

m4 update:

362.0994 TiB
1191 hours
Avg speed 90.60 MiB/s.
AD gone from 147 to 145.
P/E 6345.
MD5 OK.
Still no reallocated sectors


Kingston V+100:

5.9450 TiB
23 hours
Avg speed 78.50 MiB/s.
AD gone from 194 to 184.
P/E ?
MD5 OK.

[Anvil]

256.29TB Host writes
Reallocated sectors 6
MD5 OK

Everything is normal.

[[XC] Oj101]

I have another question, would a drive ever reach a stage where data starts to become corrupt?

[Vapor]

I have another question, would a drive ever reach a stage where data starts to become corrupt?

NAND/SSD makers have said that even if a drive fails due to NAND failure, it should still be readable for a year. Nobody has had any read failures yet (that's what the MD5 checks are about), and even the Samsung 470 that had NAND failure was still readable, so we have no evidence of any data corruption yet.



C300 Update

248.648TiB host writes, 16 MWI, 4202 raw wear indicator, 2048/1 reallocations, 62.8MiB/sec (finally back to normal-ish), MD5 OK


SF-1200 nLTT Update (from much earlier today, been awhile since MWI dropped)

118.125TiB host writes, 80.5TiB NAND writes, 62 MWI, 1288 raw wear (equiv), wear range delta 3, 56.25MiB/sec, MD5 OK

I had my endurance test system offline last night--was expecting a power outage this morning for electrical repairs for my neighborhood and didn't want to have another cropped C300 log--so have only had ~12 hours of endurance test progress since last update.

[bulanula]

Wondering if the Samsung 470 will really retain the data for 1 year. Maybe it will retain it for even more than one year ?

[bluestang]

M225->Vertex Turbo 64GB Update:

136.14 TiB
518.2 hours
MWI 30 (drops by 1 for every 50 raw wear)
3500 Raw Wear
110.72 MB/s avg for the last 16.11 hours (on W7 x64)
MD5 OK
C4-Erase Failure Block Count(Realloc Sectors) at 2 (Bank 6/Block 2406; Bank 3/Block 3925)

[Anvil]

258.91TB Host writes
Reallocated sectors 6
MD5 OK

34.0MiB/s on avg (~27.5 hours)

[bulanula]

Wonder how the C300 will end up performing compared to the Intel or the Samsung. The no-LTT Sandforce is also very interesting as well.

[One_Hertz]

339TB. 69 reallocations. MD5 OK. My SSD is actually speeding up by about 0.1mb/s per day. It is 0.4mb/s faster now than it was late last week.

[B.A.T]

m4 update:

369.8099 TiB
1215 hours
Avg speed 90.34 MiB/s.
AD gone from 145 to 140.
P/E 6481.
MD5 OK.
Still no reallocated sectors



Kingston V+100:

12.5365 TiB
48 hours
Avg speed 77.15 MiB/s.
AD gone from 184 to 179.
P/E ?
MD5 OK.

[[XC] Oj101]

Why's the Kingston so slow? It's rated at 145MB/s. If applicable, surely LTT wouldn't kick in at only 12.5TB?

Seems it's actually 180MB/s on writes

[B.A.T]

The Kingston doesn't have LTT because it uses a Toshiba controller, not Sandforce.
It might be a combination of things like a aggressive GC and little time for the controller to shuffle data.

[[XC] Oj101]

The Kingston doesn't have LTT because it uses a Toshiba controller, not Sandforce.

I thought as much, hence saying "if applicable." Would GC really hurt so badly?

[B.A.T]

Well, it sure looks like it. I don't have any better explanations, maybe some of the others in the thread can help with some understanding.

[Anvil]

Keep in mind that the 180MB/s is based on a rather large block size, the endurance test is writing a mix of block sizes and >50% is less than 128KB in size.

It looks like the controller struggles with this mix.

[Vapor]

Kingston V+100's apparent WA gets weirder...after being exactly the same twice in a row (4.166666x), it's down to 2.61x overall. At this point I'd guess the SMART attribute that's declining isn't MWI as we know it (i.e., not a linear scaling--relative to P/E cycle--counter) and therefore my WA calculations are basically useless

It was at 92 (equiv) after 6TiB and is at 89.5 (equiv) after 12.59TiB...so yeah, definitely not linear and seemingly not related to P/E cycles (unless recent WA has gone from ~4.17x to ~1.19x, which I suppose is possible).

[Anvil]

There are a few more attributes that could start changing, nothing like a P/E counter though. (unless there is some trick to interpreting AD)

[Vapor]

C300 Update

254.5TiB host writes, 14 MWI, 4300 raw wear indicator, 2048/1 reallocations, 63.0MiB/sec (going back up again...), MD5 OK


SF-1200 nLTT Update (from much earlier today--I seem to be out of sync with the MWI movement)

123.188TiB host writes, 84.47TiB NAND writes, 60 MWI, wear range delta 3, 1351.5 raw wear (equiv), 56.15MiB/sec, MD5 OK

[bulanula]

The M4 is going quite far already but I bet the C300 will go even further because of 34nm NAND and same controller type.

[Anvil]

260.63TB Host writes
Reallocated sectors : 6
MD5 OK

33.36MiB/s on avg (~43 hours)

[B.A.T]

m4 update:

375.405 TiB
1234 hours
Avg speed 90.34 MiB/s.
AD gone from 140 to 137.
P/E 6581.
MD5 OK.
Still no reallocated sectors



Kingston V+100:

17.3295 TiB
66 hours
Avg speed 76.59 MiB/s.
AD gone from 179 to 177.
P/E ?
MD5 OK.

[bluestang]

The M4 is going quite far already but I bet the C300 will go even further because of 34nm NAND and same controller type.

Well, in that case the M225 will go even farther with it's 51nm NAND (10k P/E).

[Vapor]

SMART attribute AD is definitely not MWI-as-we-know it. If it were, WA would have been .65x over the past ~4.8TiB. This Kingston V+100 is a weird one