SSD Write Endurance 25nm Vs 34nm

[bluestang]

M225->Vertex Turbo 64GB Update:

348.22 TiB (382.87 TB) total
1000.57 hours
7445 Raw Wear
117.84 MB/s avg for the last 16.17 hours (on W7 x64)
MD5 OK
C4-Erase Failure Block Count (Realloc Sectors) at 4
(Bank 6/Block 2406; Bank 3/Block 3925; Bank 0/Block 1766; Bank 0/Block 829)

I think the added pauses in the new Beta8 has helped out speed alot. Especially consistent speed through the loop.
I'm now doing loops at 98-100 seconds compared to 114-117 seconds with the old settings.

[bulanula]

These Indilinx drives are quite good, it seems ! I thought they were pretty bad at first. Nice to see they are fast and endurant BUT I think their 4K scores suck ?

[Vapor]

C300 Update

408.408TiB host writes, 1 MWI, 6895 raw wear indicator, 2048/1 reallocations, 62.8MiB/sec, MD5 OK


SF-1200 nLTT Update 1

249.75TiB host writes, 178.844TiB NAND writes, 10 MWI, 2861.5 raw wear (equiv), 55.7MiB/sec, MD5 OK

SF-1200 nLTT Update 2

262.5TiB host writes, 193.844TiB NAND writes, 10 MWI, 3101.5 raw wear (equiv), 55.65MiB/sec, MD5 OK


Seems both SF-1200s are 'stuck' at 10 MWI right now.

Charts will be updated when B.A.T returns from his vacation in a few days...both SF-2200s will be new additions

[Christopher]

These Indilinx drives are quite good, it seems ! I thought they were pretty bad at first. Nice to see they are fast and endurant BUT I think their 4K scores suck ?

The 4K throughput for 60GB - 120GB Indilinx drives is about 12MB/s reads and 7MB/s writes. Or is it the other way around...? Bluestang's M225 --> Vertex Turbo is an excellent specimen, but probably won't be much past that. Some Indilinx drives were just better than others.

EDIT.....
Speaking of the Bluestang special, look at how few bit errors it has. My Indy drives have 1000x times as many, and two of them are brand new (maybe...).

[sergiu]

It looks that way, it has now changed to 82. (15-20 minutes ago it was 94)

I'll still report it for a few more days in case it starts making sense. (can't see any connections to other things though)

I am following these parameters for almost one year on my model and I have found no direct link with anything. For example, I watched the evolution of the value very closely (refresh every 20-40s) both when drive was in idle and was running Anvil's storage utility. The raw value increased around the same rate. Also I have not noticed any ascending trend over the year but also, the drive is "brand new" compared to what we have here. As raw value is reset to 0 when drive enters a power down cycle, it might be useful to monitor and log the evolution during the entire test. We might notice an ascending trend at the end of the lifetime.

[bluestang]

The 4K throughput for 60GB - 120GB Indilinx drives is about 12MB/s reads and 7MB/s writes. Or is it the other way around...? Bluestang's M225 --> Vertex Turbo is an excellent specimen, but probably won't be much past that. Some Indilinx drives were just better than others.

[Christopher]

Here's tonight's update today.

Now that the Bit error count is up in the 8 digit range, Raw read error rate has stabilized.

Mushkin Chronos Deluxe 60
F1 Host Writes 20015
E9 NAND Writes 15402
E6 Life Curve 100
E7 Life Left 95
Average 122.57MBs last 12 hours.

[Christopher]

Here's AS SSD for my Indilinx 60 with Intel NAND



I think those M225's are just awesome.

Once you can start doing 30 to 40 MB/s 4K reads it feels a lot less like a HDD.

[Anvil]

I think the added pauses in the new Beta8 has helped out speed alot. Especially consistent speed through the loop.
I'm now doing loops at 98-100 seconds compared to 114-117 seconds with the old settings.

Yes, in general it helps the OS/filesystem to keep up with the writes, without the small pauses there were small freezes while deleting files.
(especially noticeable on the F3 due to the number of files and the SF TRIM "bug")

..
Seems both SF-1200s are 'stuck' at 10 MWI right now.

For how long?
(most likely a side-effect of the nLTT FW, let's see what happens to mine)

I am following these parameters for almost one year on my model and I have found no direct link with anything. For example, ...

I save a screenshot of CDI or SSDLife every day so if there is some correlation to other attributes it might be possible to pick up, not sure that it is important though.

...
Mushkin Chronos Deluxe 60
...
Average 122.57MBs last 12 hours.

With those speeds I'd be surprised if others aren't picking up on the toggle mode NAND on the 60GB drives

[Christopher]

With those speeds I'd be surprised if others aren't picking up on the toggle mode NAND on the 60GB drives

Well, it's super impressive at the 120GB size, but don't forget that it's still only writing 75% of that. In reality, it's not much faster at these endurance loads than my Agility 60.

For a 60GB SF drive, there's no question that's it's the champion of the universe. I'm surprised that it's the ONLY 60GB SF2281 toggle NAND drive out there. It's impossible to find the 60 Deluxe at the moment, but hopefully it'll catch on with higher availability.

Plextor makes a Marvell with Toshiba NAND at the 60GB mark as well -- Corsair used to -- It's just not price competitive vs. the 120GB versions.

Here's the ASU benchmark I meant to post the other day.




I like fast, smaller drives.

[Christopher]

UPDATE

After a few days of flawless performance, I seem to have had some kind of crash. I was away for a little while when it happened, and when I came back, the system had restarted. Not sure what happened, but I'm going to err on the side of caution and make some changes to RAM timings to ensure that it was not the cause.

It could be the mysterious SF2281 instability, which is why I want to ensure that there can be no question about system stability... not that I think my system idling caused the crash in any event.

EDIT.

I started back up again, and started getting a couple of these Write Error 0s

[Christopher]

Look at this



Those ECC numbers were in 8 figures earlier, and are now at 0

What the hell?!?

[Christopher]

And now, like magic, they're back, but not at the same level.
I bet this smart attribute is really some kind of timer or something (who knows?).
I don't think it's anything like the label would imply.

[Computurd]

weirdness...that is very curious. wth is going on?

[Christopher]

weirdness...that is very curious. wth is going on?

I think the firmware took a dump, or the sudden system reboot f'ed up the smart data recording. Everything seems to be fine now. No data loss, no problems except for that write error 0 in ASU.

[Vapor]

For how long?
(most likely a side-effect of the nLTT FW, let's see what happens to mine)

Been a few days now, normally each MWI tick was ~2.5TiB and it's been stuck at 10 MWI for over 14TiB. SynbiosVyse's last update had 10 MWI when it could/should have been 9 (based on extrapolation). I would be very surprised if this were due to the nLTT setting, I think it's just a Sandforce oddity (like their first 3-5% expected lifetime being stuck at 100 MWI).

And now, like magic, they're back, but not at the same level.
I bet this smart attribute is really some kind of timer or something (who knows?).
I don't think it's anything like the label would imply.

I've noticed RRER jumping around on my 4 Sandforce drives, as well. Never thought much of it though--normalized threshold is at 50 and I don't think I've ever seen the normalized value worse than 85.

I wonder if there were some way to log it every minute-ish and see if potential drastic dips align with potential future crashes.


@Anvil and @Christopher, what kind of static data are you using? Your wear range deltas are already way past mine (3) and seem to be scaling like SynbiosVyse's (which is over 100, as of last update). SF-1200 and SF-2200 aren't exactly the same beast, but I would be surprised if wear leveling performance were reduced on the SF-2200 (unless it were for a tangible performance gain). The odd thing with my WRD vs. SynbiosVyse's is that he has just a few MB of static data whereas over 50% of my drive is filled with static data--one would expect wear leveling to be more effective on an empty drive.

[Christopher]

@Anvil and @Christopher, what kind of static data are you using? Your wear range deltas are already way past mine (3) and seem to be scaling like SynbiosVyse's (which is over 100, as of last update). SF-1200 and SF-2200 aren't exactly the same beast, but I would be surprised if wear leveling performance were reduced on the SF-2200 (unless it were for a tangible performance gain). The odd thing with my WRD vs. SynbiosVyse's is that he has just a few MB of static data whereas over 50% of my drive is filled with static data--one would expect wear leveling to be more effective on an empty drive.

I'm using the Mushkin with 17.5GB of data (It's my main system's image, cloned to the Mushkin, which I'm using as the boot drive).

Out of the 55.8GB, 17.5GB is comprised of Windows, apps, and data. No system restore points, swap file, no hiberfil.sys.

I thought it would be more realistic to actually use the drive as my main system drive, taking the place of my Intel 510.
It's amazing how responsive the system still is while pounding the system drive with writes.

[sergiu]

Those ECC numbers were in 8 figures earlier, and are now at 0

What the hell?!?

Numbers are reset to 0 after each restart/shutdown/standby/power loss. Value is not cumulative. The smart parameter might actually be an indicator for an average number/hour

[Anvil]

I think the firmware took a dump, or the sudden system reboot f'ed up the smart data recording. Everything seems to be fine now. No data loss, no problems except for that write error 0 in ASU.

It happened to me 12-13 hours ago as well, I'll try to tweakl the end of the loop some more (the cleaning up part) as it looks like it occurs while there is a lot of stress on the controller. (TRIMming)

As my drive is a spare I don't have to reboot, the drive has to be reconnected though. (it is disconnected)

I excpect this is the major bug. (if the drive was my boot drive it would surely have rebooted)

[Anvil]

Kingston SSDNow 40GB (X25-V)

340.24TB Host writes
Reallocated sectors : 7
MD5 OK

36.86MiB/s on avg (~12 hours)

--

Corsair Force 3 120GB

01 90/50 (Raw read error rate)
05 2 (Retired Block count)
B1 27 (Wear range delta)
E6 100 (Life curve status)
E7 91 (SSD Life left)
E9 43940 (Raw writes)
F1 58566 (Host writes)

109.19MiB/s on avg (~12 hours)
Uptime 167 hours. (power on hours)

SSDLife estimates lifetime to be 1 month 18 days.

[Anvil]

...
@Anvil and @Christopher, what kind of static data are you using? Your wear range deltas are already way past mine (3) and seem to be scaling like SynbiosVyse's (which is over 100, as of last update). SF-1200 and SF-2200 aren't exactly the same beast, but I would be surprised if wear leveling performance were reduced on the SF-2200 (unless it were for a tangible performance gain). The odd thing with my WRD vs. SynbiosVyse's is that he has just a few MB of static data whereas over 50% of my drive is filled with static data--one would expect wear leveling to be more effective on an empty drive.

"The drive is filled with W7 files (OS), one Virtual Machine, 22GB of mp3 files + some miscellaneous files (mostly incompressible), this leaves ~49GB of free space." (from a few pages back )

It's growing steadily, there's not much time to do GC nor static wear leveling. (edit)

I'm preparing a new module for doing more extensive performance testing, it includes full span seq/random IO, I'll play a bit just to see how long it takes for GC to kick in vs filling the drive completely and letting TRIM do the job.

This is what it looks like after a period of Endurance testing. (w/o time for GC to kick in)
24-September-2011_00-00.png

This is after filling the drive and letting TRIM do it's job.
24-September-2011_00-13.png

[Christopher]

It's been running fine for the last 18 hours.

I wasn't really aware of any TRIM bug, nor have I ever heavily watched SandForce SMART data -- at least before three days ago -- so every attribute looks funny to me.



This was my first SF drive and I wasn't really sure what to expect.

The first thing I noticed about this drive was that it writes as fast with incompressible data as a C300 128GB does... as in pretty damn fast. Seriously -- I saw a CDM screen shot of a C300 128 - the writes were identical. The 120 version of the Chronos Deluxe must be awesome.

I love this thing. Considering how nerfed most 60GB SF2281 drives are, it's kinda in a class by itself. If you could say one drive is more "fun" than another, then this drive has been a lot of fun so far.

[Ao1]

Corsair Force 3 120GB


E7 91 (SSD Life left)
E9 43940 (Raw writes)
F1 58566 (Host writes)

So by the looks of it Corsair do not deploy LTT.

[Christopher]

Here's my update. Notice that I'm now at the same MWI as the Anvil's Corsair.


Mushkin Chronos Deluxe 60
F1 Host Writes 29380
E9 NAND Writes 22610
E6 Life Curve 100
E7 Life Left 91
Average 121.57MBs last 18 hours.

[Christopher]

If there's some kind of throttling going on, I can't tell. I thought life throttling kept MWI above life curve anyway.

If there is no movement in life curve, I think there's no LTT -- I could be completely wrong about this, but then I usually am.

Another item that catches my eye is the Threshold vaule for MWI. It's set at 10, so maybe these drives aren't going to hit 9, and they'll simply stay at 10 for the rest of eternity.