SSD Write Endurance 25nm Vs 34nm

[Vapor]

Great news

How will it be tested? 50% compressible for static data and uncompressible data for test? (seems fair )

Probably 46% for static data and 67% for test data. With the E9/233 SMART value, I can see how NAND writes directly relate to wear, so no absolute need for 101% incompressible data for the test. Many pages back I found the 67% data had a compression curve similar to the types of data I would put on an SSD (apps, documents). The 46% is most similar to the OS and apps compression curve, but 67% is a more conservative version (less compressible than OS/Apps, but more compressible than documents), so I might do 67/67 or 46/67, not entirely sure yet.

Attachment 0
I'm not sure but I don't think it is MWI according to what I found and posted in my original Snip from the Intel pdf.
OneHertz statement it is now at 120 after starting at 0 is too fast of an increase for MWI paramaters to be used in reverse.
What it is based on is not explained though.

Percentage Used Endurance Indicator- % used over what? MWI?

That exact quote is what made me think it was just MWI counting upward and past 100

If anything, a 120 is too low a value for it to be MWI based...to maintain the 1.015x WA, it needs a value of 125 at 241TiB.

[Anvil]

@Anvil...How's the app looking for release?

I'm waiting for some feed-back, if all is OK then it should be made available today/tonight.

[EniGmA1987]

So Vapor, is the Samsung 470 good or bad drive? Some of the charts make it look amazing to me while some make it look like the worst drive ever.

[Vapor]

So Vapor, is the Samsung 470 good or bad drive? Some of the charts make it look amazing to me while some make it look like the worst drive ever.

It burned through its MWI SMART attribute very quickly through an apparent write amplification of ~5x (which is why it looks 'bad' in many charts), but it's still going strong having written its formatted size (64GB) over 4000 times (enough for writing 64GB/59.6GiB per day for ~11 years)

[Anvil]

@Enigma

I'm using the Samsung 470 128GB as the boot drive right now and it's performing just like any other top shelf SSD, as for Endurance I'd say that it's doing exceptionally well. (based on the 64GB drive in this test)

The one thing I do miss is some more SMART attributes, not a big issue though.
I'll post some benchmarks later tonight or tomorrow.

[B.A.T]

Todays update:

156.5918 TiB
525 hours
Avg speed 89.04 MiB/s.
AD gone from 15 to 10.
P/E 2731.
MD5 OK.

Only thing changed is that CDI is reporting BAD in healt status, why I don't know but the AD is marked red and CA markes yellow.


Here is also a screenshot of Gsmartcontroll

[Vapor]

Pretty cool that CDI and Gsmartcontrol turn red when you pass the manufacturer's threshold

[B.A.T]

And the most exciting has yet to come

2-3 more days and I'll cross mwi=1

[Vapor]

I wonder if CA will scale past 100 and still act like MWI? Although MWI is easy enough to calculate, it has stuck to steps of 30raw.

[B.A.T]

We'll soon enough know. And how far it will endure

[johnw]

Only thing changed is that CDI is reporting BAD in healt status, why I don't know but the AD is marked red and CA markes yellow.

That is the way SMART works. The attributes start at some number, like 100, then decrease until they fall below a threshold. In this case, you have fallen below the 10 threshold, which indicates a SMART failure for that attribute.

[johnw]

There is no rest, learn to understand the power of good programming languages

How can you not realize that there are people reading your posts who have actually written linux kernel code and kernel modules? It is quite obvious that you have no clue how to write a kernel module.

[johnw]

237.913 TiB, 652 hours, sa177: 1/1/19564

247.432 TiB, 678 hours, sa177: 1/1/20345

Average speed reported by Anvil's app has been steady at about 112MB/s.

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.

So, attribute 177 just passed 20,000. If it is counting the average number of erases each flash block has undergone, that is impressive, since most 2X or 3Xnm flash is only rated for 5000 cyles. Even 34nm eMLC (enterprise MLC) is only rated for 30,000. Either this Samsung flash is very good, or attribute 177 is not what we thought it was. I guess we will have to wait and see if it passes 30,000 next.

[Vapor]

Prototype chart, thoughts?



Realized almost most of the SSDs were already at MWI = 1, so took inspiration from subway maps to reinclude moving parts on the main charts. All lines should be visible at all lengths, even if there isn't much conflict right now

(m4 already 'owns' the vertical slot between the Samsung and the X25-V)

[Computurd]

nice chart Vapor! that does a very good job of illustrating both the expected lifetime, and the *after* lifetime

[mgoldshteyn]

You guys do realize that the Intel 320 series probably has over a thousand blocks it can use for reallocation. So, at the seemingly linear rate this test is currently moving, we could be well into multiple petabytes before the 40 GB 320 runs out of blocks it can use for reallocation. I mean, what's a page on a modern SSD, 8KB? The page size actually is exactly 8KB on the 320, IIRC? If I remember correctly, the block size on the 320 is 2MB, and we all know erases are done in blocks and not pages. How much spare area is there? 8-12 %? So, worst case, we're talking 40GB * 8% / 2MB = 1,525 blocks. Since, we're really most likely reallocating blocks and not sectors, I am not sure how to read the graphs to see where we're at in terms of reallocated blocks.

[nn_step]

Second update:


OCZ Agility 3 AGT3-25SAT3-120G - Both drives are throwing failures but drives are still functional
Corsair Force CSSD-F120GB2-BRKT - a couple minor sector failures on one of the drives.
OCZ Vertex 2 OCZSSD2-2VTXE60G - Nothing yet to report
Corsair Performance 3 Series CSSD-P3128GB2-BRKT - One drive failed completely July 26 @ 11:43am and second drive is throwing failures but has yet to fail.
Crucial RealSSD C300 CTFDDAC064MAG-1G1 - a couple minor sector failures on one of the drives.
SAMSUNG 470 Series MZ-5PA128/US - a couple minor sector failures on one of the drives.
Intel 510 Series (Elm Crest) SSDSC2MH120A2K5 - One drive failed completely July 24 @ 9:14am and second drive failed July 26 @ 3:37pm
Intel X25-M SSDSA2MH160G2K5 - a couple minor sector failures on one of the drives.
Kingston SSDNow V+ Series SNVP325-S2B/128GB - One drive failed completely July 26 @ 2:14pm and second drive is throwing failures but has yet to fail.


-------- End of line ----------

How can you not realize that there are people reading your posts who have actually written linux kernel code and kernel modules? It is quite obvious that you have no clue how to write a kernel module.

thank you for your words, be what they may.

From what I've seen, it seems that the rest of the drives will all die this month.

[Computurd]

From what I've seen, it seems that the rest of the drives will all die this month.

trying to stay out of this, but cant.

Your "testing" follows no standard that has been followed in this thread. You arent participating in any form with the testing in this thread.

this is a distraction from what is being done here. Your results arent matching the carefully observed and recorded results of others in this thread.
The fact that there is *supposedly* no throttling on your SF devices points to something being wrong/misrepresented with your results.

You are muddying the waters for those reading, with little to NO hard data.

2) Intel drives received 3% less sunlight than the other drives

statements such as this are inane. i think it is in jest? dont know your motivation there. I dont feel it is being said in a cooperative manner.

I encourage you to continue your testing, but I also feel that you should start your own thread, and i also *feel* that other active members of this subsection feel the same.

[nn_step]

trying to stay out of this, but cant.

Your "testing" follows no standard that has been followed in this thread. You arent participating in any form with the testing in this thread.

this is a distraction from what is being done here. Your results arent matching the carefully observed and recorded results of others in this thread.
The fact that there is *supposedly* no throttling on your SF devices points to something being wrong/misrepresented with your results.

You are muddying the waters for those reading, with little to NO hard data.



statements such as this are inane. i think it is in jest? dont know your motivation there. I dont feel it is being said in a cooperative manner.

I encourage you to continue your testing, but I also feel that you should start your own thread, and i also *feel* that other active members of this subsection feel the same.

Thank you for your calm perspective and I agree, this thread and my experiment despite similar goals are not related enough to share the same thread.

[Ao1]

Prototype chart, thoughts?

Nice to see this, especially as writes are far exceeding MWI. Would something like this not be more easy to visualise?

[Anvil]

@Anvil...How's the app looking for release?

It's still in beta but can be downloaded from the following link (TheSSDReview)

--

190.33TB Host writes
Reallocated sectors 6

MD5 OK, no change in avg speed.

[B.A.T]

M4 update:

162.3427 TiB
544 hours
Avg speed 88.85 MiB/s.
AD gone from 10 to 6.
P/E 2833.
MD5 OK.

[johnw]

I encourage you to continue your testing, but I also feel that you should start your own thread, and i also *feel* that other active members of this subsection feel the same.

If he is actually doing any testing. There are at least three anomalies in what he has posted that suggest he is making it up:

1) He has not posted any SMART attributes despite repeated requests to do so

2) He claims to be writing continuously to the SSDs at 50MB/s, despite the fact that we know that Sandforce drives throttle much slower than that, and he seems unaware of Sandforce warranty throttling.

3) He has posted code that he claims is everything for a linux kernel module that obviously cannot be a kernel module without additional code

[bulanula]

2) Intel drives received 3% less sunlight than the other drives

That really made my day and the fact that he never encountered the SF throttling and the code is incomplete ( who writes a kernel module for this kind of stuff ??? ) just proves that he either is taking us as fools or really suffers from a condition that makes him overcomplicate stuff.

Besides, you never post any SMART data and just say "drive is near failing or has failed" which is obviously not good enough in this situation.

Please take this farce of a test somewhere else or consider solving some of the issues above and posting all of your data before saying "drive has failed" or similar statements.

This really looks like a company sponsored "independent" testing that makes their product look best. They never show facts they just tell us "our product is gazillion times better" and no proof / data whatsoever. Thanks but no thanks.

[Hopalong X]

Should have used UV block.

Suffer the consequences without proper protection!