SSD Write Endurance 25nm Vs 34nm

[Anvil]

Well, we've started the quest to find out how long an SSD can last.

I'm using the Kingston SSDNow 40GB, a rebranded Intel X25-V and One_Hertz is using the new 320 Series 40GB SSD.

I'll be posting updates every day, well, thats my intention at least

This is the status of my SSD just before the test started.

V40_before_endurance_testing.PNG

My SSD is currently running as the C drive on a laptop and it will continue to do that at least until it's up to 35TB, which is the expected life based on 20GB per day for 5 years.

So, 1/3 of the drive is filled with static data (W7 x64) an the rest is used for writes. The app fills the drive with random sized files ranging from 1KB up to 12MB in size.
More than 50% of the files are 128KB or smaller in size.
TRIM is active.

I'll be posting the first update within an hour or two, it's almost 24h since I started and thought that an 24hour update would be interesting.

Summary (in the works)











16th of May, start of test
20th of May Ao1 entered the test with a Vertex 2 40GB using 34nm NAND
21st of May Ao1 produced a chart showing the SF "hang" during deletes (page 3) Link
21st of May, Vapor on compression Link
22nd of May, Ao1 summary/thoughts on the SF drive so far Link
22nd of May, Ao1 tries a Samsung F3 Link
23rd of May, overthere joins in and explains on Ao1's hIOmon testing Link
24th of May One_Hertz's 320 is at 26TB
24th of May Anvil's Kingston (X25-V) is at 22.5TB
24th of May One_Hertz's 320 got the first reallocated sector
26th of May, TRIM on Ao1's V2 is now up to 20 seconds. Link
27th of May Ao1 chart on compression/throttling Link
Link to white paper on impact of the frequency of writes. (recovery)
Read disturb
28th of May, Ao1's V2 started LT Throttling (7 days and 30TB), throughput is down to ~5MiB/s Link (page 10)
One_Hertz's 320 is at 43TB
Anvil's Kinston is at 35TB
30th of May, Kingston moved from Dell Latitute to AMD rig. (41TB, MWI 76)
We started using random writes "module"
2nd June, One_Hertz is at 60TB (MWI 70%)
Ao1 tries to revive the V2, it still throttles. Link
Discussion on how LTT is regulated, power on hours vs host writes
4th of June, Kingston is at 50TB MWI 71
Trendline for MWI exhaustion
7th of June One_Hertz's 320 78TB (MWI 60) which is 2x the advertised "endurance"
Ao1 ran the test for 6 hours, write speed 6MiB/s which equals about 500GB per day.
9th of June Ao1 summary and end of V2 test Link
11th of June
Kingston at 70TB, MWI 61
Intel 320 at 92TB MWI 53
13th of June Chart on MWI/Host writes Link (page 16)
15th of June
Kingston at 81TB MWI 55
Intel 320 at 107TB MWI 45
Discussion on Write amplification/wear leveling Link
16-17th of June, MD5 checking being implemented
Kingston is at 86TB MWI 52
Intel 320 is at 113TB MWI 41
Spec for Crucial 64GB drives is 72TBW (page 18)
21st of June, One_Hertz performed a full corruption test on his 320, no errors. Link
Discussion on testing, nand quality
22nd June
Kingston at 98TB MWI 46
Intel 320 at 131TB MWI 32
Chart by Ao1 on Host writes/MWI Link
Discussion on SLC, is it dead?
24-25th of June
Heated discussion on throttling which resulted in a new thread by Ao1
Intel 320 142TB MWI 27
Kingston 107TB MWI 41
"Up to 20GB Host writes per day for a minimum of 5 years" Link (page 22)
26th of June Chart by Ao1 Link
28th of June, Johnw enters the test with the Samsung 470 64GB SSD Link (page 23)
Discussion on Compression
29th of June Johnw's drive produces 12TiB in 24hours Link
Vapor and B.A.T announces they will be entering the test shortly
30th of June
Vapor on compression Link (page 25)
Pause after each loop is introduced.
Charts by Vapor
1st of July
B.A.T started testing the m4 Link
SMART table for Intel
SMART table for Micron
4th of July
Updated charts (page 29)
5th of July
The Samsung 470 sa177 attribute is already down to 1 (~63TiB) Link
Vapor received the C300 and has issues with SMART attributes, too many? (page 31)
6th of July
NAND info found by Ao1 (Intel 25nm, Micron 34nm, Samsung,
7th of July
MD5 testing introduced (page 32)
Updated charts, including C300
8th of July
1st week of m4 summary
The Intel 320 is down to 3% (186.5TiB)
9th of July
Image by B.A.T of the NAND on the m4
Listing of NAND by Ao1
The Intel 320 is down to 1% (190TB)
Updated charts (page 34)
11th of July
The Samsung 470 is at 110TiB
13th of July
I'll be impressed if the Samsung 470 makes it past 200TiB says johnw (page 36)
Approximate Write Amplification chart added
15th of July
C300 is past 50TiB writes + updated charts
Some drama, My drive (X25-V) dropped out on the AMD rig, moved to an Intel rig for further testing. (162TB)
m4 is past 100TiB
16th of July
Samsung 470 is past 150TiB
17th of July
My drive (X25-V) was off for the weekend, no issues.
The Intel 320 has 48GiB of flash
Intel 320 is at 217TiB, 24 Reallocated sectors.
Picture of the 320
20th of July
Some info on normal usage (page 41)
21th of July
The Samsung 470 is past 200TiB avg speed 112MB/s
23rd of July
The X25-V is past 180TB host writes (reallocated sectors at 6)
24th of July
X25-V MWI is at 1
25th of July
C300 is past 100TiB + updated charts
nn_step on his Endurance testing (not part of this test)
26th of July
m4 entered Bad status state (MWI 10)
New chart by Vapor
27th of July
The app goes Beta
28th of July
m4 goes past 3000 P/E cycles
bluestang started testing the Crucial M225
29th of July
m4 shows up as Good as attribute wraps around
Intel 320 is past 250TB, 33 reallocated sectors
bluestang needs to use Wiper on the Indilinx drive
Charts updated
30th of July
Vapor got an Vaporized SF-1200 drive w/o LTT
1st of August
BAT's m4 got past 200TiB Link
johnw's Samsung 470 got past 300TiB Link
4th of August
Charts updated
6th of August
A few bets are on? Link
8th of August
Vapor on Wear Range Delta on the SF-1200 Link
The Samsung is at 999 hours
10th of August
Ao1 on MWI on the V2 and the V3 he tested separately Link
Whitepaper on ECC
11th of August
SynbiosVyse entered the test with an Corsair Force 40A Link
Charts updated
B.A.T announced that he will be entering the test with an Kingston SSDNow V+ 100 64GB.
12th of August
The Samsung is about to go 400TiB
Adding TRIM to the formula changes WA for the M225, it also broke the 50TiB mark.
14th of August
The M225 really flyes on W7, MB/s almost doubled from XP
Chart posted by Ao1 on recycling (page 52)
P/E Interleaving/Channels/Dies
15th of August
Vapor's C300 is past 210TiB, Vaporized is at 85TiB
M225 is at 75TiB, Samsung 470 is at 433TiB, m4 64 is at 304TiB, F40A is at 25TiB
16th of August
Charts updated
17th of August
Calculating WA for the Charts
M225's Erase failure count from 0 to 1 Link
Samsung 470 is showing signs of failure at 450TiB and is slowing down Link
19th of August
M225 past 100TiB, m4 is past 330TiB
20t of August
Samsung has less than a day left? at 472TiB
There is no LTT on the Corsair F40A (it is past 48TiB)
A bit of WA
21st of August
Samsung 470 is out of the test, 478TiB
Charts updated
The X25-V has finally reached 250TiB, it took >3 months.
22nd of August
B.A.T started testing the Kingston V+100 64GB
24th of August
Intel 320, the speed is increasing (339TB, 69 reallocations)
25th of August
m4 is at 375TiB, Kingston V+100 is at 17TiB, C300 is at 255TiB, nLTT = 123, X25-V = 260TB, Corsair F40A = 100TiB (0Fill)
29th of August
m4 is past 400TiB and the Kingston dropped out. (was quickly back up again)
Testing SSFLife
1st of September
C300 = 290TiB, nLTT = 126TiB, F40A = 126TiB, m4 = 424TiB, V100+ = 35TiB, X25-V = 280TB, M225 = 190TiB
(and I've ordred an F3 120GB)
2nd of September
Intel 320 40GB 370TB and 85 reallocated sectors.
4th of September
M225 reached MWI 0
9th of September
X25-V gets past 300TB
Charts updated
10th of September
Development on the Intel 320 series, reallocated sectors from 105 to 4071 (could fail shortly)
Great post on P/E rating by frostedflakes
12th of September
Christopher is looking at candidates for testing...
m4 is at 511TiB
14th of September
jcool lists a X25-E with 580TB Host writes! Notice that MWI is still at 90
15th of September
m4 = 532TiB, C300 = 367TiB, nLTT = 226TiB,
17th of September
Ao1 lists endurance info from Intel Link
4K writes 100% span Intel Link (page 68)
The X25-V drops out on the AMD rig, works fine on the Intel rig?
4K Endurance?
19th of September
Loop write speed (bluestangs M225) Link
SSDLife predicts that the F3 120GB will last for 6 months 10 days (April 8th 2012)
20th of September
Corsair F40A = 190TiB, X25-V = 330TB
21st of September
F3 = 35TiB,X25-V = 333TB, both my drives are now running off an Z68 (ASRock) consuming 72W
22nd of September
The Mushkin Cronos Deluxe makes it's first appearance Link
23rd of September
Intel 320 = 438TB
SSLife now predicts that the F3 will last for 1 month 20 days.
C300 = 408TiB, nLTT = 263TiB w MWI stuck at 10 (proved to be normal) Link
24th of September
Christopher caught the SF2 bug Link
It happened to the F3 as well so both SF2 drives are affected by the bug. (mine lasted for 1 week)
25th of September
Christopher said about the Mushkin "I don't think this drive is going to die anytime this year" (famous last ...)
SynbiosVyse's F40A has gone past 206TiB
The F3 continues to disconnect
26th of September
The Mushkin disconnects
The F3 continues to disconnect
27th of September
Factory Bad Block count
SF2 discussions
B.A.T secure erased the Kingston after having issues.
28th of September
Charts updated
30th of September
m4 = 561TiB, Kingston V+100 = 81TiB, X25-V = 355TB, Force 3 = 101TiB, Mushkin Chronos Dlx = 72TiB, Corsair F40A = 229TiB
1st of October
C300 = 446TiB, nLTT = 296TiB
Workloads on Intel drives Link (page 77)
The Force 3 has been moved to another Z68 w/o results and is now running off an X58 ICH10R
The m4 gets past P/E 10000
2nd of October
command for smartmontools
1.3.2 fw released for the Corsair
3rd of October
The Force F40A died
The Force 3 lasted for 51hours on the X58 and then disconnected, it is not related to chipsets imo, updating firmware is next.
4th of October
johnw checked the Samsung on September 20 and it did not detect and managed to bend the connector, possible data retention failure!
One_Hertz will check the drive at his lab.
Jedec spec
1) The SSD maintains its capacity
2) The SSD maintains the required UBER for its application class
3) The SSD meets the required functional failure requirement (FFR) for its application class
4) The SSD retains data with power off for the required time for its application class
5th of October
The Force 3 disconnected again. (completely froze the system)
Reconfigured the computer to no C states (no power savings)
Voodoo Science
m4 = 604TiB, Kingston V+100 = 118TiB, Mushkin = 129TiB, (X25-V) 371TB, Force3 = 149TiB
6th of OCtober
Looks like C states combined with the new FW is what it takes to get it stable? (>46 hours without issues)
New thread created for the SF2XXX issues Link

Wear Range Delta is a topic (page 82)
8th of October
Force 3 > 83 hours without issues
More on WRD
SynbiosVyse confirms that the F40A can't be detected.
9th of October
The Kingston V+100 drops out again.
Link on Static wear Leveling
10th of October
The boot drive (Force 3 GT disconnected) so test had to be restarted
The Mushkin has been running for 45 hours w/o issues
WRD Charts by Ao1
11th of October
One_Hertz gives his 320 another week or two (~488TB written)
12th of October
Reallocated sectors (4904) are still increasing on the 320
13th of October
Windows issues a warning on the 320 as E8 is below threshold.
Charts on WRD/MWI by Ao1
M225 Milestone, Erase count is past 1000
14th of October
Intel 320 is at 499TB
Formulas linked by Ao1 (wear leveling/ P/E)
Force 3 uptime ~94 hours
15th of October
Intel 320 500.71 TB. 1 wear leveling reserve left!
Force 3 uptime 119 hours
The Intel 320 is still going strong!
16th of October
The Kingston V+100 dropped out again
Formula for calculating life-time
X25-V has 40GiB of flash. 40GB 320 has 48GiB of flash
17th of October
One_Hertz received the Samsung from johnw
The (X25-V) gets past 400TB
18th of October
No disconnects for over a week for the Force 3 (on fw 1.3.2)
(SF announces that the BUG is fixed?)
Chart on WRD/MWI by Ao1
Speed is decreasing on the Intel 320 (down by ~3MB/s)
The m4 = 700TiB
19th of October
The Samsung 470 can't be revived
B.A.T announces that he will be entering 2 more drives
21st of October
(X25-V) 209TB, Force 3 271TiB, Muskin Chronos 207TiB, M224 Turbo 604TB, m4 724TiB, Kingston V+100(offline)
22nd of October
The Force 3 is idling to check if WRD changes, no changes were seen during 10 hours of idle time.
Intel 320 523TB 6977 Reallocated sectors
23rd of October
F3 refreshed the drive by moving all data off the drive and then back again. Link
WRD started decreasing when the test was restarted.
24th of October
B.A.T startes testing the Intel X25-M G1 80GB
B.A.T's MTron SLC fails do co-operate and won't be joining the test (for now).
25th of October
One_Hertz initiates a retention test on the Intel 320
The m4 gets past 750TiB
27th of October
The X25-M G1 is slowing down and needs attention to perform.
* Uploading images is no longer supported on the forum??
28th of October
X25-V 431TB, Force 3 336TiB, M225 662TB,
31st of October
The Mushkin has had 128 hours uptime,
The test-rig (Z68) for the Force 3 has been up for 12 days, the F3 has worked for 179 hours
1st of November
M225 is past 701TB
2nd of November
bluestangs logfile
3rd of November
X25-V 446TB, Force 3 383TiB (uptime > 251 hours), Mushkin Chronos 404TiB (uptime 207 hours), M225 722TB,
4th of November
Charts by Ao1
The Mushkin has been running for 240 hours, the Force 3 for > 275 hours.
The Mushkin has started generating a few more retired blocks
5th
JEDEC states (on Endurance)
The Intel 320 is back, no issues. (offline since October 25th)
6th
The Mushkin starts having issues (at 432TiB)
B.A.T is back with a new born
Unfortunately the m4 is bricked, it was disconnected for 9 days and it did not survive, data retention is to blame.
7th of November
The thread has had more than 500.000 views
The Samsung 830 carries a SMART attribute for host writes
8th of November
M225 gets past 700TiB
Kingston V+100 gets past 200TiB, Intel X25-M ~25TiB,
Link to JEDEC
9th of November
Christopher discovers that RRER is below threshold on the Mushkin and this is the cause of CDI is reporting errors (465TiB)
The Corsair Force 3 is at MWI 10 (which is really 0) at ~438TiB
10th of November
The Mushkin died at 475+TiB
M225 794TB, X25-V 466TB, Force 3 447TiB, Kingston V+100 215TiB, X25-M G1 33.5TiB,
11th
Small summary by sergiu (on causes of death)
M225 is past 800TB
13th of November
The Kingston V+100 still disconnects, frequently.
Christopher shows signs of re-entering the test with some drive
The Force 3 is about to overtake the X25-V (which has been running for almost 6 months)
15th of November
B.A.T orders another m4
The M225 overtakes the m4 on TiB Host writes (768.91TiB)
17th of November
Intel 311, just a teaser
Quick retention test on the Force 3 and the X25-V
18th of November
Completed retention testing both drives for 11 + 24 hours, all is well.
Recovery between P/E cycles (linked by Ao1)
M225 gets past 800TiB
The Force 3 gets past 500TiB
19th of November
Intel 320 -> 573.3TB. 7424 reallocated sectors (page 109)
Kingston V+100 263TiB, X25-M G1 68TiB
The Kinstson has issues and has slowed down to 1/3 of it's original speed, < 30MiB/s
21st of November
The M225 is reported dead? it is later revived (823TiB, 905TB total)
It *fails* the MD5 test later that day.
bluestang had a power outage, it still works
25th of November
The X25-V gets past 500TB
30th of November
Intel 320 604TB, Force 3 604TiB, X25-V 515TiB, M225 846TiB, X25-M G1 80TiB
(The Force 3 is having another retention test today)
1st of December
Christopher announces that he will be entering with an Samsung 830 64GB
2nd of December
The Force 3 survived the 54 hour retention test.
5th of December
The battle of the... is about to start (Samsung 830 + Crucial m4)
The M225 has issues having idled the weekend
6th
M225 struggles again and it's close to 1PB
First update on the Samsung 830 and the new m4 64GB
M225 876.17 TiB (963.36 TB) total, bluestang will babysit the little drive

(page 116/117)

....




Updated charts


Host Writes So Far






Normalized Writes So Far
The SSDs are not all the same size, these charts normalize for available NAND capacity.






Write Days So Far
Not all SSDs write at the same speed, these charts factor out write speeds and look at endurance as a function of time.






Host Writes vs. NAND Writes and Write Amplication
Based on reported or calculated NAND cycles from wear SMART values divided by total writes.







--