SSD Write Endurance 25nm Vs 34nm

[Christopher]

Oh.

That makes me sad.

You could try d flashing it. The 1.7 dflash is public on the OCZ forum.

[Ao1]

@ Anvil. I agree. For specific applications SF based drives are a great solution. The downer is that those specific applications usually operate in an environment where reliability is of critical importance. SF drives suffer in that regard, not so much due to SF but due to the limited ability of SF vendors to do their part of the validation processes.

My SF2xxx is also dead by the way. It clapped out over the weekend.

[bluestang]

MD5 test on static data just hangs at the 2.6 GiB point of 32.6 GiB total. I''m leaving work in about 90 minutes, so I'll let it run overnight to see what happens. CDI C6 "Total Count of Read Sectors" keeps ticking upwards, just slowly.

[devsk]

@bluestang: D-flash it. Vertex drives do have propensity to get into this mode. If D-flash doesn't help, then its really dead! 900TB is nothing short of a miracle anyway.

[bluestang]

Oh.

That makes me sad.

You could try d flashing it. The 1.7 dflash is public on the OCZ forum.

@bluestang: D-flash it. Vertex drives do have propensity to get into this mode. If D-flash doesn't help, then its really dead! 900TB is nothing short of a miracle anyway.

If the MD5 test is still not finished when I get back in tomorrow morning, then I will try and D-Flash it to 1.7.

EDIT:
Update: MD5 test finally finished and test on 32.6 GiB static data failed.

[Anvil]

Not all motherboards have the problem with lower 4KB random reads when processor C-states are enabled. But I'm not sure which motherboards are which.

Also, remember that CDM sometimes gives strangely high results for 4KB read on Sandforce SSDs.

What does AS-SSD measure?

I'll say that it's the MB's using the PCH that are mostly affected. (I'll do some tests on the ICH to confirm)

CDM is not using threads and I can't find anything wrong with the code.
ASU and AS SSD uses threads and should be close but it's plain to see that AS SSD uses a different approach when measuring 4K in particular.
(AS SSD takes forever on HDD's and uses multiple files when testing QD, ASU and IOmeter uses a single file for QD testing)

As a result of the different approach they will never compare 100%, they should be relatively close, and they are close most of the time.

[johnw]

I added AS SSD results above. I don't think the Chronos' 4K reads were particularly high with CDM, but AS SSDs seem excessively low. My M4 only scores about 12MBs reads @ 4K with AS SSD.

LOL. Anytime you see 4KB random reads over 30MB/s, that is quite high. The only SSDs I have seen achieve that level consistently are Crucial C300s and I believe some Indilinx (well, iodrives are much higher, but I am talking normal SSDs).

For some reason I do not understand, CDM shows anomalous 4KB read results on Sandforce 22XX SSDs. If you use AS-SSD, IOMeter, or ASU (incompressible data), they will usually measure in the low 20s MB/s. But the same Sandforce SSD on CDM often measures low-to-mid 30s MB/s. So it is CDM that is the anomalous one for 4KB random read tests.

If you disable CPU C-States, you will probably see your m4 4KB random read result go up to low-20 MB/s.

[johnw]

CDM is not using threads and I can't find anything wrong with the code.

Right, we have discussed this before. No one has an explanation for it, but many SF22XX SSDs get low-to-mid 30s MB/s 4KB random reads on CDM, but low 20s MB/s random reads with AS-SSD, IOMeter, ASU.

[johnw]

For client applications you get a great boost from the spare area that is made from the OS and application installs, which saves around 4GB of NAND writes. After that however savings start to evaporate. We know that SF can easily compress zeros, but they struggle to compress anything else in client based applications. For sure SF cannot compress anything close to the theoretical compressibility of data in client applications with a low QD, so I would argue quiet strongly that the theoretical compressibility of application data is nothing like what can be achieved in real life. No-one (to my knowledge) using SF drives for normal client based activities has been able to demonstrate a significant difference between host and NAND writes.

Right. For flash longevity, Sandforce compression offers almost no benefit, since as you say, only the OS install and application installs have data that the Sandforce SSDs can compress. Most user data saved to the SSDs after the initial installs will be hardly compressed at all. And since any SSD used to write hundreds of TBs is unlikely to be installing the OS over and over, it is likely that most of that data written will not benefit much from compression.

[sergiu]

Update: MD5 test finally finished and test on 32.6 GiB static data failed.

Do you still have the original file? if yes, then could you do a bit to bit compare? I would be interested in how many mismatches are and if there are inside a single page or spread across multiple pages

[johnw]

Do you still have the original file? if yes, then could you do a bit to bit compare? I would be interested in how many mismatches are and if there are inside a single page or spread across multiple pages

Also it would be good if he can clarify whether he was able to compute the MD5 checksum at all, or whether the failure was due to part of the file being unreadable.

[Anvil]

Kingston SSDNow 40GB (X25-V)

491.71TB Host writes
Reallocated sectors : 05 12
Available Reserved Space : E8 99

MD5 OK

34.71MiB/s on avg (~33 hours)

--

Corsair Force 3 120GB

01 88/50 (Raw read error rate)
05 2 (Retired Block count)
B1 52 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 406317 (Raw writes) ->397TiB
F1 540818 (Host writes) ->528TiB

MD5 OK

106.34MiB/s on avg (~33 hours)

power on hours : 1565

B1 is down again from 53 to 52.

[Christopher]

You know, I wonder how much that M225 actually had written to it before it was refurbished and repackaged. The WA of earlier Indilinx FW was pretty high, so it could have a significant amount of writes on it before it was sent to Bluestang and became a M225 Turbo Edition thanks to a little flashing.

[bulanula]

Sad to see the Indilinx die too. So far none of the drives were able to be read-only after they failed. So much for the "SSD do not wear out and you don't lose data it just becomes read only" myths. I bet the C300 would have made it to 1 PB easily if not more.

BTW, does anybody know what happened to Vapor and his charts and his C300 ? Anyone got a C300 to prove my point ? Thanks !

[B.A.T]

Todays update:
Kingston V+100
267.6568 TiB
1103 hours
Avg speed 25.52 MiB/s
AD still 1.
168= 1 (SATA PHY Error Count)
P/E?
MD5 OK.
Reallocated sectors : 00


My write speed is still low. I've tried to format and let trim work and just deleting all files before I copy them back but with no luck. I'm away for over a week so my only options is through teamviewer.

Intel X25-M G1 80GB
75,8924 TiB
19058 hours
Reallocated sectors : 00
MWI=110 to 92
MD5 =OK
49.23 MiB/s on avg


@bluestang
There is a wave of malfunction ssd's right now. The M225>Vertex Turbo gave it a good run for the money

[Christopher]

Hmm. I think we need to have a discussion concerning the charts and graphs given that Vapor is MIA.

There have been monumental changes and addition since the last update, and I think at a bare minimum we need some basic chart on the first page reflecting these changes.

BTW Ao1,

I don't necessarily disagree with your contention that SandForce drives (in a consumer environment) don't really benefit from compression after the OS and applications are installed. However, I look at uncompressible perfomance. I don't just think that I'm always going to get the 555MBs+/500MBs+ a drive is rated for. I believe that a 2281 with toggle nand would still be fast without compression of any kind. Having compression is just the icing on the cake, and one more reason to choose a synchronous equipped model over an async. Just because you might not benefit as much from dedupe doesn't mean there's no benefit at all .

Which 120GB drive is as fast as a Chronos Deluxe/MaxIOPS/Wildfire/OWC Mecury Pro 120GB even without compression speeding things along?

[johnw]

However, I look at uncompressible perfomance. I don't just think that I'm always going to get the 555MBs+/500MBs+ a drive is rated for. I believe that a 2281 with toggle nand would still be fast without compression of any kind.

No, the problem is that the performance of Sandforce SSDs for incompressible data is WORSE than other SSDs that do not do compression. That is what Ao1 was talking about. There is minimal longevity advantage from compression for most user data, and there is a performance hit for reading back the compressed data. For details, look in some of the other threads where Ao1 does a lot of measurements.

[squick3n]

There's definitely a performance hit relative to compressible data, but don't the toggle and sync NAND SF drives still outperform most non-SF drives when using incompressible data?

[johnw]

There's definitely a performance hit relative to compressible data, but don't the toggle and sync NAND SF drives still outperform most non-SF drives when using incompressible data?

Not really. A Crucial m4, Samsung 830, or Intel 510 will beat a Sandforce 22XX SSD in many benchmarks when using incompressible data. Of course, it depends on the exact benchmark, and the Sandforce SSDs are faster in some specific areas. But I'd rather have any of those other three in most cases.

[Anvil]

Kingston SSDNow 40GB (X25-V)

493.19TB Host writes
Reallocated sectors : 05 12
Available Reserved Space : E8 99

MD5 OK

34.16MiB/s on avg (~46 hours)

--

Corsair Force 3 120GB

01 85/50 (Raw read error rate)
05 2 (Retired Block count)
B1 51 (Wear range delta)
E6 100 (Life curve status)
E7 10 (SSD Life left)
E9 410014 (Raw writes) ->400TiB
F1 545737 (Host writes) ->533TiB

MD5 OK

106.34MiB/s on avg (~46 hours)

power on hours : 1578

B1 is down again from 52 to 51.

--

We do need to find someone to do the charts, I've had no response at all.

[bluestang]

Do you still have the original file? if yes, then could you do a bit to bit compare? I would be interested in how many mismatches are and if there are inside a single page or spread across multiple pages

I still have the original file.

Also it would be good if he can clarify whether he was able to compute the MD5 checksum at all, or whether the failure was due to part of the file being unreadable.

Yes I was able to compute the checksum. It took 2+ hours to finish and the checksum did not match the original. However, drive is still accessible and the MD5 test ASU was performing every 50 loops during testing does pass...



I'm was also able to run an ATTObenchmark on it this morning...



I believe that there are areas that just are not writable and/or readable anymore. I'm open on ideas of what to do next.

[bluestang]

For the hell of it I ran W7 CHKDSK...

Chkdsk was executed in read/write mode.  

Checking file system on X:
Volume label is M225_64_NS.

CHKDSK is verifying files (stage 1 of 5)...
Cleaning up instance tags for file 0x2a.
  9728 file records processed.
  File verification completed.
  3 large file records processed.
  0 bad file records processed.
  0 EA records processed.
  0 reparse records processed.

CHKDSK is verifying indexes (stage 2 of 5)...
  9764 index entries processed.
  Index verification completed.


CHKDSK is verifying security descriptors (stage 3 of 5)...
  9728 file SDs/SIDs processed
Cleaning up 18 unused index entries from index $SII of file 0x9.
Cleaning up 18 unused index entries from index $SDH of file 0x9.
Cleaning up 18 unused security descriptors.
Security descriptor verification completed.
  19 data files processed.
CHKDSK is verifying Usn Journal...
  66195040 USN bytes processed.
Usn Journal verification completed.
CHKDSK is verifying file data (stage 4 of 5)...
Read failure with status 0xc00000b5 at offset 0x32aea1000 for 0x10000 bytes.
Read failure with status 0xc00000b5 at offset 0x32aeb0000 for 0x1000 bytes.
Read failure with status 0xc00000b5 at offset 0x32aeb1000 for 0x10000 bytes.
Read failure with status 0xc00000b5 at offset 0x32aeb1000 for 0x1000 bytes.
Read failure with status 0xc00000b5 at offset 0x5cf582000 for 0x10000 bytes.
Read failure with status 0xc00000b5 at offset 0x5cf58c000 for 0x1000 bytes.
Read failure with status 0xc00000b5 at offset 0x5cf58d000 for 0x10000 bytes.
Read failure with status 0xc00000b5 at offset 0x5cf58d000 for 0x1000 bytes.
Read failure with status 0xc00000b5 at offset 0x8d2dd0000 for 0x10000 bytes.
Read failure with status 0xc00000b5 at offset 0x8d2dd0000 for 0x1000 bytes.
Read failure with status 0xc00000b5 at offset 0x8d2dd1000 for 0x10000 bytes.
Read failure with status 0xc00000b5 at offset 0x8d2dd1000 for 0x1000 bytes.
Windows replaced bad clusters in file 55
of name \Static\Static.zip.
  9712 files processed.
File data verification completed.
CHKDSK is verifying free space (stage 5 of 5)...
  7014172 free clusters processed.
Free space verification is complete.
Adding 6 bad clusters to the Bad Clusters File.
Correcting errors in the Volume Bitmap.
Windows has made corrections to the file system.

  62517247 KB total disk space.
  34317456 KB in 29 files.
       464 KB in 20 indexes.
    142611 KB in use by the system.
     65536 KB occupied by the log file.
  28056692 KB available on disk.

      4096 bytes in each allocation unit.
  15629311 total allocation units on disk.
   7014173 allocation units available on disk.

EDIT1: C5 "Read Failure Block Count" (uncorrectable bit errors) went from 1 to 3 as a result of CHKDSK...



EDIT2: Ran 7zip extraction test on my static.zip file and only 3 files threw CRC fails due to "broken file" out of 4522 files

[Christopher]

Perhaps D flashing would help, but I think its a long shot. Might as well.

[bluestang]

Well, we had a power outage here for about 45-50 minutes. My UPS only lasted 2 minutes, battery crapped out, better get one one order!

Anything else gonna go wrong...oh yeah wait! When power came back on, only 2 of the 3 servers here fired back up after I shut them down since it looked like power would be out for hours. Damn power supply and memory was bad.

So, my endurance/work PC was down without power for over an hour....and M225 is still here! Didn't suffer from power off state like the others

[Christopher]

Well, we had a power outage here for about 45-50 minutes. My UPS only lasted 2 minutes, battery crapped out, better get one one order!

Anything else gonna go wrong...oh yeah wait! When power came back on, only 2 of the 3 servers here fired back up after I shut them down since it looked like power would be out for hours. Damn power supply and memory was bad.

So, my endurance/work PC was down without power for over an hour....and M225 is still here! Didn't suffer from power off state like the others

Bravo


EDIT
Incidentally, I got my VertexLE in the mail today. The few benches I've run on it seem pretty good, but for some reason the SMART attributes that track reads/host/raw writes don't seem to be working. Not sure if that's normal or not, but overall it's pretty nice. It might be nice to run a SF1000 without Hynix nand too in the test.