SSD Write Endurance 25nm Vs 34nm

[bulanula]

This thread kind of confirms what I have been saying all along :

-avoid Samsung and Sandforce if you can get Intel or Crucial at the same price
-avoid 25nm if you can get 34nm at same price

Really eager to see a dead SSD folks !

[Anvil]

None of those SSD's are limiting normal users, in any way

I'm not saying that just because I've got a few of both of those two you mentioned

[johnw]

This thread does not confirm anything of the kind. Where do you come up with this crazy stuff?

[bulanula]

Well that is my personal opinion on this kind of stuff so I though I would share it here.

So far, the Samsung is seemingly less endurant than the Intel and the Sandforce is getting warranty throttled.
Also with the same controller and same firmware 25nm NAND will never be better than 34nm NAND ( it's just physics ).

[Eggcake]

I can understand the first statement - that SF and Samsung is not on the same level as Intel or Crucial. Either because the endurance is pretty low (Samsung) or for other reasons (Sandforce). However, the endurance of the Samsung drive is still very high and should not be a problem for most of the users. Although I will never support SandForce while they throttle their devices and advertise unachievable performance, it's still not a problem for most of the users as you really have to stress your drive for a long period of time.

And the second statement may be true in theory, but how on earth do you conclude such a thing from this thread? The tests (so far) clearly show that the 25nm Intel 320 has roughly the same endurance as the 34nm G2. Yes, it's not the same drive, but I still don't quite understand how you come to this conclusion...

[johnw]

Well that is my personal opinion on this kind of stuff so I though I would share it here.

So far, the Samsung is seemingly less endurant than the Intel and the Sandforce is getting warranty throttled.
Also with the same controller and same firmware 25nm NAND will never be better than 34nm NAND ( it's just physics ).

If it is your personal opinion, then why not state it as your personal opinion rather than claiming the data in this thread confirms it?

Since the Samsung has not failed yet, there is no data in this thread to confirm that the Samsung is "less endurant".

And for the same controller and firmware, 25nm flash COULD be better than 34nm flash. It is not just physics. There is engineering and manufacturing involved, too. The 25nm chips could simply be higher quality (for various reasons) than the 34nm chips. We don't know. Once the SSDs start failing, then we will have some data.

[Vapor]

Updated charts

C300 isn't in the MWI Exhaustion graphs because it's just too soon...MWI is down to just 97, very hard to extrapolate with accuracy off of that.

Host Writes So Far


(bars with a border = testing stopped/completed)


Raw data graphs

Writes vs. Wear:


MWI Exhaustion:



Normalized data graphs
The SSDs are not all the same size, these charts normalize for 25GiB of onboard NAND.

Writes vs. Wear:


MWI Exhaustion:



Write-days data graphs
Not all SSDs write at the same speed, these charts factor out write speeds and look at endurance as a function of time.

Writes vs. Wear:


MWI Exhaustion:

[johnw]

vapor:

One graph I'd like to see would be average block erase count vs. TiB written. For the Samsung, I think we agree that attribute 177 is probably the average block erase count. For Micron, 0xAD is the "average erase count of all good blocks". On the Intel 320, I think 0xE9 attribute raw value may be the average block erase count ("the number of cycles the NAND media has undergone").

[Vapor]

I can do that

Will be tough to back track on the Intel 320, unless One_Hertz has that info handy. I have been logging it for your Samsung and the two Crucial SSDs, which is good.

One caution with the graph, however....it will just be an upside-down MWI vs. TiB graph. The C300 and your Samsung move the MWI in increments of 50raw (though the Samsung doesn't really stick to it perfectly) and the m4's moves in increments of 30raw. Still, it will be useful for post-MWI Exhaustion wear tracking

[johnw]

Still, it will be useful for post-MWI Exhaustion wear tracking

Exactly. That's why I requested it!

Although I'm not sure it is even available on the Intels. I just checked some of my Intel SSDs (very low writes) and the raw value of the wear parameter is 0 on all of them. The writes are not THAT low!

[Vapor]

Samsung seems capable of nearly 10x the cycle usage speed of C300...which had me thinking. Considering one of the values will always be dwarfed on the chart (at least until the Samsung dies), logarithmic or linear axis scale?





EDIT: theoretically, these values are available on SandForce via SMART attribute 233, just divide by NAND size. At least one new SandForce will be entering in the coming weeks

EDIT2: a lack of early-life Samsung data is why it's 'better' than the Crucials early in life on the logarithmic chart...Excel isent to brite If logarithmic scale axis is the choice--should I add interpolated data?

[johnw]

Samsung seems capable of nearly 10x the cycle usage speed of C300...which had me thinking. Considering one of the values will always be dwarfed on the chart (at least until the Samsung dies), logarithmic or linear axis scale?

I prefer the linear scale graph, since the relationship is expected to be linear. Also, we would be able to see, visually, if the ~WA changes, by observing a change in the slope of the graph.

[bulanula]

At least one new SandForce will be entering in the coming weeks

Can somebody say "unthrottled SF2 drive" ???

[Computurd]

-avoid 25nm if you can get 34nm at same price

i hate to but i have to jump in here as well. there is NO correlation between this statement and any results contained in this thread thus far. simply not true, or if it is, that result hasnt been borne out yet.



CMON guys, first to 1/2 PB is the winner

[johnw]

65.521 TiB, 189 hours, sa177: 1/1/5483

74.641 TiB, 214 hours, sa177: 1/1/6237

The other two unknown SMART attributes, 178 and 235, are still at 72/72/276 and 99/99/2, just as they were when the SSD was fresh out of the box.

[Ao1]

Hi John, when you were waiting for the MDF5 update did you leave the drive on idle? Maybe that period allowed static data rotation to revitalise the NAND reserve? Seems a bit strange that it is still going strong 9TB after getting to MWI 1. Even if the NAND specs are based on minimum PE cycles, that still seems quite a lot of data

[dr_sharp]

I don't think the OCZ data should be extrapolated the way it is presented in the graphs. The testing stopped because it was completely throttled. In the eyes of the users on this forum, that would be considered a failure.

I think you should remove the dotted trendline from OCZ data.

[B.A.T]

And there was the first week of testing over. To sum it up so far:

• The M4 have written 49,0105 TiB.

• (AD)Wear leveling count has gone from 100 to 72.

• Avg speed 84.97 MiB/s

• 168 hours finished

Otherwise everything looks normal, the speed is good and there has not been any reallocated sectors yet.

[One_Hertz]

186.5TiB. 3%. Reallocated sectors up to 19.

[Anvil]

I don't think the OCZ data should be extrapolated the way it is presented in the graphs. The testing stopped because it was completely throttled. In the eyes of the users on this forum, that would be considered a failure.

I think you should remove the dotted trendline from OCZ data.

It managed to write more than most people are likely to write during 3 years of rather heavy usage, (8TiB per year is way above normal usage)
The dotted line shows interesting data as it shows what could have been, if it wasn't for the throttling.
So, yes, it failed to complete the Endurance test, it did not fail as a drive though.

Hi John, when you were waiting for the MDF5 update did you leave the drive on idle? Maybe that period allowed static data rotation to revitalise the NAND reserve? Seems a bit strange that it is still going strong 9TB after getting to MWI 1. Even if the NAND specs are based on minimum PE cycles, that still seems quite a lot of data

I'd say it might go on for a few more weeks depending on how wear leveling is implemented.
3000 P/E cycles is about 190TB. (best case)

--

141.55TB Host writes
MWI 22 (early)
Reallocated sectors, still at 6.

During the last 24.5 hours it managed an average of 32.13MiB/s writes + 9 rounds of MD5 testing a ~1.7GB file.

[bulanula]

So far what SSD would you guys say is the "best" and what would you recommend people buy ???

[Vapor]

I don't think the OCZ data should be extrapolated the way it is presented in the graphs. The testing stopped because it was completely throttled. In the eyes of the users on this forum, that would be considered a failure.

I think you should remove the dotted trendline from OCZ data.

Wear vs. TiB might still stick to that line, even with LTT (and my guess is that it would stick pretty closely to that line). The tester chose to stop testing but the drive was still functioning as intended, even if it were not functioning as desired.

Wear vs. Write-days extrapolation would change dramatically (for the 'better') as the drive would now be forced to last 1000+ days rather than the 55-60 it could wear itself down in without LTT. If anything, I feel I should add another line indicating actual life expectancy in days...but that would mess up the visibility of the charted data from every other drive

That said, I expect to remove the data from the Vertex 2 40GB when a SF-1200 drive without LTT enters testing (partially because testing on the V2-40GB is permanently incomplete and partially because of the initial 0-fill numbers complicating things), but until then it'll be on the chart because there is no alternative/better SF-1200 data.


C300 Update:

15.0234TiB, 95MWI, 251raw, 0 reallocated, 62.12MiB/s, 42/0 with MD5 checks (1.65GiB file)

[Vapor]

Updated charts

First time the C300 is in the MWIE charts...MWI reached 95 with a raw of 251, so it was early in the cycle and is, hopefully, good for extrapolating.

Host Writes So Far


(bars with a border = testing stopped/completed)


Raw data graphs

Writes vs. Wear:


MWI Exhaustion:


Writes vs. NAND Cycles:



Normalized data graphs
The SSDs are not all the same size, these charts normalize for 25GiB of onboard NAND.

Writes vs. Wear:


MWI Exhaustion:



Write-days data graphs
Not all SSDs write at the same speed, these charts factor out write speeds and look at endurance as a function of time.

Writes vs. Wear:


MWI Exhaustion:

[Anvil]

Finally some figures for the C300

So, 300TiB, should be great fun!

[bulanula]

On another note, according to the "formula", a 128GB C300 with 34nm NAND should last ( 5000 * 128 ) or 640 000 GB of writes or 640 TB of writes ! The 64GB version of the C300 should last about half of that so about 320 TB etc. ???

Also you guys might wanna include secure erase endurance testing here as well. How many times can a SSD be secure erased ??? Might be a good point of investigation.