SSD Write Endurance 25nm Vs 34nm

[bulanula]

Any news on the C300 vapor?

[Vapor]

Any news on the C300 vapor?

UPS attempts redelivery tomorrow and I work from home tomorrow so testing should start tomorrow

[johnw]

53.426 TiB, 153 hours, sa177: 11/11/4491

57.962 TiB, 166 hours, sa177: 3/3/4869

[Computurd]

It seems people have reported this to Intel in May, perhaps earlier. So Intel knows about it. But they haven't issued any statements or warnings about it, and have not recalled the drives nor have they issued updated firmware.

you would figure that anyone with this issue would be granted an RMA real quick with the 5 year warranty and whatnot. sounds like an extreme corner case.

Power On/Off Cycles 50,000 cycles
Power On/Off Cycles is defined as power being removed from the
SSD, and then restored. Most host systems remove power from the
SSD when entering suspend and hibernate as well as on a system
shutdown.

that is a damn big number. they didnt pull that out of thin air, either. im sure there was some severe testing of this. i haven't ever heard of this issue, and im all over the place. the fact that Anvil mentioned it with the G2 even surprised me. very surprised ive never heard of it actually. (must be very rare)

[B.A.T]

93 hours, 26,5944 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 87 to 85.
Avg speed for all 93 hours is roughly 83,2 MiB

[Anvil]

My drive is still slowing down, it's down from 32.8MB/s to 32.6MB/s (on avg) since last report, so, I'll definitely continue doing the manual TRIM evert 24hours.

133.29TB Host writes
MWI 27

No other changes.


@Vapor

Looking forward to that C300, you might catch up on my drive

[One_Hertz]

Good point. Do you know what NTFS does if a program requests a write of, say, 1KiB?

The rest of the cluster is padded with zeroes.

[bulanula]

Two things I am really looking forward to :

-johnw's Samsung and going below MWI of 0
-vapor's C300 testing start and screenshots of SMART attributes so we can compare them to the M4 eg things like factory bad block counts which will matter too in the 25nm vs 34nm debate!

Thank you all for making this possible !

[johnw]

57.962 TiB, 166 hours, sa177: 3/3/4869

62.535 TiB, 178 hours, sa177: 1/1/5230

The normalized value of attribute 177 has been 1 for a while, I don't think it will decrease further. The raw value has been continuing to increase (it was 5017 about 8 hours before this update) even after the normalized value pegged to 1 (which happened about 8 hours before this update).

Anvil's app was writing merrily along without any obvious problems, but I decided to stop the app for a while. I'll wait until Anvil's updated app with MD5 copy/check is ready until starting the writes again. In the meantime, I am computing the MD5 sum for the static data file, and running AS-SSD.

[johnw]

The md5 checksum for the 42GB static data file on the Samsung 470 is still the same.

The speed in AS-SSD still looks okay:



So now I will just wait for the MD5-enabled version of Anvil's app before starting the writing again.

By the way, I got the 64GB Samsung 470 for $99. For just under $400, you could stick four of these on Intel fakeRAID, and have a 256GB volume with great performance.

[devsk]

62.535 TiB, 178 hours, sa177: 1/1/5230

The normalized value of attribute 177 has been 1 for a while, I don't think it will decrease further. The raw value has been continuing to increase (it was 5017 about 8 hours before this update) even after the normalized value pegged to 1 (which happened about 8 hours before this update).

Anvil's app was writing merrily along without any obvious problems, but I decided to stop the app for a while. I'll wait until Anvil's updated app with MD5 copy/check is ready until starting the writes again. In the meantime, I am computing the MD5 sum for the static data file, and running AS-SSD.

I will be very happy if my Vertex 1 120GB survives till 62 TiB... Currently, it looks like it will die at about 4TiB mark (1000 erase count on 5000, and written about 1TiB in one year).

[sergiu]

I will be very happy if my Vertex 1 120GB survives till 62 TiB... Currently, it looks like it will die at about 4TiB mark (1000 erase count on 5000, and written about 1TiB in one year).

I just computed the WA for Samsung model and I got 5.22 which seems for me a very high value. Your Vertex 1 would have a WA of around 120 which is huge. Somebody should bring some older models to test them and compare how much wear algorithms evolved and which drive has the best ones. A few days ago I was convinced that drives with lower speed might be software limited. Now I believe that the final result is the tradeoff between WA and speed and it would make sense. A 40GB model has around 9.76 million 4KiB pages or around 76.2 thousand blocks from which it must choose wisely so that time spend in this operation is not higher (or significantly higher) than page write/ block erase time.

[Vapor]

Got the C300 and getting it running now...here's the intake SMART/performance data. Had to flash from 0006 to 0007 and then also transport the 40GB file over the network and onto the C300, so that took some time, but here's what I have:

SMART right after flash 0007:


CDM:


AS-SSD:


SMART after benchmarks and dropping files onto the SSD....I suddenly have more SMART values?

[Anvil]

Anvil's app was writing merrily along without any obvious problems, but I decided to stop the app for a while. I'll wait until Anvil's updated app with MD5 copy/check is ready until starting the writes again. In the meantime, I am computing the MD5 sum for the static data file, and running AS-SSD.

I'm giving it a few loops on my setup, shouldn't be long now.

The file should be no more than a few GB, copying the file still counts as writes but some time is spent on creating the MD5 checksum.

edit:

@Vapor

A lot of new attributes?
That looks really strange, will try on one of my C300's, they are all at 0007.

[johnw]

I just computed the WA for Samsung model and I got 5.22 which seems for me a very high value.

Of course, you are assuming SMART attribute 177 raw value is the average erase count of the blocks, and that there is not a large variance in erase counts among the blocks. That may be a correct assumption, but I have not been able to find a single bit of documentation from Samsung to confirm this interpretation.

[johnw]

SMART after benchmarks and dropping files onto the SSD....I suddenly have more SMART values?

Getting smarter?

[B.A.T]

105 hours, 30,2923 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 85 to 83.
Avg speed for all 105 hours is roughly 83,6 MiB

[bulanula]

WOW really interesting C300 behaviour right there !

That "Getting smarter ?" reply was epic johnw ! LOLed so hard.

[Vapor]

4.83hrs, 1.0822TiB, 100MWI, 17 raw wear

[Vapor]

And those last four values are changing and changing often.

[Anvil]

Those attributes are not found on the C300's I've looked at.

Did you try using smartmontools?

The MD5 version will be available tomorrow, performing tests until then.

[Vapor]

smartmontools/GSmartControl can only recognize F2/242 (the one with 95 as the value) as "Total_LBAs_Read", the rest are shown but unknowns.

I was afraid I messed something up during the flash to 0007...I'll explain what I did.

Booted the system with the C300 plugged in and see it's 0006. Burn the 0007 firmware CD and boot off of it and flash it, all goes well.

Boot back into Windows, take the initial Crystal Disk Info screenshot and go to run AS-SSD...4k writes are ~1MB/s so I figure something is wrong. Double check alignment [ok], double check TRIM [ok], double check write cache [ok], shut down to check AHCI/IDE. I figure the system must be in IDE mode because the flash was successful and because performance is wretched.

In BIOS, it says it's AHCI, so I flip it to IDE and put the 0007 CD back in to reflash. Boot off the disk and it says 0007 is already flashed, can't do anything else. Go back into BIOS, switch it back to AHCI, boot into Windows and performance is fine and suddenly I have 7 more SMART values showing up

[B.A.T]

114 hours, 32,8314 TiB, Wear Leveling Count and Percentage of the rated lifetime used has gone from 83 to 82.
Avg speed for all 114 hours is roughly 83,8 MiB

[Anvil]

...
I was afraid I messed something up during the flash to 0007...I'll explain what I did.
...

I don't think you did, the flash upgrade wouldn't have worked at all if there was something off with your setup.

The result however was really something , more questions to be answered.
Could this be a new revision of the hardware. (controller)

Will double check on my other drives as well, the ones I checked last night were all 64GB.

--

The MD5 test looks to be working just fine, I've let it run through the night and it's been running every 10 loops, no errors.

135.82TB Host writes
MWI 25 (it just turned 25)

Reallocated sectors still at 6.

[Vapor]

I don't think you did, the flash upgrade wouldn't have worked at all if there was something off with your setup.

Yeah...that's where I'm particularly confused though since A) apparently I flashed in AHCI, which Crucial's guide says isn't possible, B) the mysterious ~1MB/s 4k writes are still unexplained, and C) I didn't actually do anything (just toggled between AHCI/IDE/AHCI and booted from the 0007 CD, which immediately said nothing could be done) between the initial flash to 0007 and getting the 7 extra SMART values. Definitely more questions now, even aside from "how did those SMART attributes show up?"

Anyway, C300 update this post, new charts next post.

3.918TiB, 17.58hrs, 99 MWI, 63 raw wear (looks like steps of 50).



TiB went from 1.0822 -> 3.918 (3.62x)
Hours went from 4.83 -> 17.58 (3.64x)
Power-on hours went from 6 -> 18 (3x...with low resolution measurement, it was a high 18, it's already 19 a few minutes later)
F5 went from 311891737 -> 1080080908 (3.46x)
F6 went from 2457063059 -> 8454389999 (3.44x)
F7 went from 307445319 -> 1057855304 (3.44x)
F8 went from 4446418 -> 22225604 (5x)

Seems F5-F7 could be timers of some sort? They do continue to move even with turning Anvil's app off. I think I would have to leave Anvil's app turned off for an extended period of time to know more, since I wasn't tracking how much it was moving.

Anyway, further look at it: F6 is 8x F7, which makes me think bits vs. bytes. F5 is 1.021x F7, which I can't find any significance to. And the 5x increase (between screenshots) in F8 is kind of making me think it's NAND writes.