SSD Write Endurance 25nm Vs 34nm

[Ao1]

Nice This will be interesting.

[Vapor]

Very nice johnw

Wish we could get a pair of drives with identical controllers but with 34nm vs. 25nm NAND....like a Force 60GB vs. Force 55GB, but lifetime throttle basically precludes Sandforce from this and no one else has been so...loose...with NAND selection Maybe some Indilinx drives?

Anvil, I'm curious about the various compressibility levels, are they a combination of 0-fill and incompressible or are they more 'advanced' than that (for lack of a better term)?

[Dragoonseal]

Wish we could get a pair of drives with identical controllers but with 34nm vs. 25nm NAND....like a Force 60GB vs. Force 55GB, but lifetime throttle basically precludes Sandforce from this and no one else has been so...loose...with NAND selection Maybe some Indilinx drives?

While it would still be an interesting test don't we already know that, all other things being equal, 34nm NAND lasts longer than 25nm NAND?

I'm more interested in seeing how they compensate for that, things like improved controllers, improved redundancy, extra over provisioning, etc.

[Vapor]

While it would still be an interesting test don't we already know that, all other things being equal, 34nm NAND lasts longer than 25nm NAND?

We know what we're told, but I'd still like to see it Sample size of 1 each is too small to be significant, but I'd still like to see something like 1.6x the lifespan.

I'm more interested in seeing how they compensate for that, things like improved controllers, improved redundancy, extra over provisioning, etc.

Agreed that the overall product is the most important aspect of testing

[Anvil]

@johnw

Exceptional sequential write rate on that little thing
Looking forward to the next few days, wondering how TRIM is handling the constant writes.

@Vapor
The levels are fine-tuned "functions" that return random data, so there is no combination of 0-Fill and incompressible.

The levels are
0-Fill
8% (Database, a typical compression rate for database files)
25% (doesn't really correspond to anything in particular)
46% (applications, dlls, ...)
67% (same as 25%)
100% Incompressible

The % is the compression rate that 7Zip returns for a that particular datastream, SF compression rates might not correspond at all to 7Zip.
(might be possible to find out using the SF2 controller as 233, 241 reports every GB written and not just every 64GB as is the case with SF1)

A C300 64GB would have been a perfect match for the m4 that is about to start later this week.

[Vapor]

Ah, so it is more advanced than just a mixture of 0-fill and incompressible, very nice

I still want to get 'in' on the testing myself...was going to do a 2nd SF1200 drive using partially compressible to juxtapose Ao1's with incompressible, but the lifetime throttle showed itself just a couple days before I was going to order.

Am pondering switching to a single 120GB drive and using one my 50GB V2s (with 7000 hours and just 1.7TB of writes) vs. a C300 64GB vs. a SF1200 with no lifetime throttle. I do want to contribute other than being in the peanut gallery

[bulanula]

If only we had a C300 now !

We could easily see which 2010 SSD is the best in endurance etc. Keep the expensive ( M4 / SF2 ) stuff for later on!

[johnw]

If only we had a C300 now !

We could easily see which 2010 SSD is the best in endurance etc. Keep the expensive ( M4 / SF2 ) stuff for later on!

You know, you could buy a C300 and run the test yourself. I expect Anvil would be glad to send you a copy of his app.

[Anvil]

Now that we are closing in on 0 MWI we should consider other means of creating the chart, maybe start using P/E?

Windows Update rebooted the PC so it's been idling for 9 hours.

116,86TB Host writes
MWI 36

--

@johnw
How is the 470's 177 Wear leveling count evolving, could it be that it works like the Crucials, each count = 1 P/E cycle. (meaning 1= 64GB and so on)

[One_Hertz]

Yeah when we get to 0 we will need a new way of charting this. 155TB. 20%.

[bulanula]

You know, you could buy a C300 and run the test yourself. I expect Anvil would be glad to send you a copy of his app.

If I wasn't broke I would have done this a long time ago etc. ! Stuff ( normal commodities ) is very expensive here in the UK and electronics are even more expensive.

Maybe somebody in a better financial situation from the US etc. ???

Hate to be in the peanut gallery but that is the way it's going to be unless I inherit some cash overnight

And also, I don't have a Windows machine in the house anyway ( yes, I'm THAT cheap LOL ) so maybe Anvil could make a Linux version ?

I agree, we need a new chart indicator.

[Vapor]

I would think a simple bar chart for writes (either in write-hours or bytes) would be good Writes and writes-after-zeroMWI would probably be the most pertinent categories.

Looks like I'm going to step into the testing with a C300 64GB....and maybe try to rekindle the SF-1200 testing. I'll probably get a 34nm drive SF-1200 and would like to pair my testing up with someone interested/capable testing a 25nm drive. Something like a F60 and F60A...yes, LTT is in the way, we'll figure out a way around it. If you're interested in testing a SF-1200 and coordinating with my testing, please let me know in the next couple of days--it would be easier to do this at the same time.

[johnw]

@johnw
How is the 470's 177 Wear leveling count evolving, could it be that it works like the Crucials, each count = 1 P/E cycle. (meaning 1= 64GB and so on)

177 went from 1 to 485 with about 8.6 TiB (TB?) written. I'm not sure what that means. Smartctl says it is the difference between the most and least worn block in percent, but I think it is wrong about that (it is not percent anyway, at best it would be percent x 1000 or something)

1 day elapsed:


original smart:


1 day smart:

[Vapor]

Disclaimer: this post is all just a guess

Looks kind of like Raw is P/E cycles and Normalized value is MWI. 10% wear (100-90) does correspond to 485 P/E cycles. I'll guess we'll know more when it dies

8800GB / 485 = 18.14GB per count. With a write amp of ~3.5x, 18.14GB works out to 64GB (one P/E cycle).

[johnw]

Disclaimer: this post is all just a guess

Looks kind of like Raw is P/E cycles and Normalized value is MWI. 10% wear (100-90) does correspond to 485 P/E cycles. I'll guess we'll know more when it dies

Could be! If normalized 177 is MWI, then it looks like I can run it to zero in about 10 days, and 86TiB of writes. The Intel MWI looks like it is going to 0 around 200TiB, IIRC. Samsung is more paranoid than Intel? Or 177 is not MWI. Hmmm.

[Vapor]

It's the only thing acting like a MWI...so I guess just roll with it

If it is MWI, Samsung could be more conservative than Intel, write amplification could be very different (~90TiB on 64GB vs. 200TiB on 40GB is disturbingly different though), a combination of the two, or something else entirely (different NAND?).

Good news to add to the thread: got a C300 64G and it will be entered into testing around the same time as the M4 64GB

[deathman20]

Subbed to see results. I got 2x 128Gig M4's coming by the weeks end here and after that I'll be taking my 2x Intel G2 80Gigs out of Raid 0. Curious to how much they have been writen to since I put them in service roughly 21 months ago.

[Vapor]

Subbed to see results. I got 2x 128Gig M4's coming by the weeks end here and after that I'll be taking my 2x Intel G2 80Gigs out of Raid 0. Curious to how much they have been writen to since I put them in service roughly 21 months ago.

HD Sentinal can read SMART data through a RAID0 array

[Dragoonseal]

HD Sentinal can read SMART data through a RAID0 array

So can the Intel SSD Toolbox, or at least some of them? I'm not sure how many there actually are, seems to have all the important ones though.



Jesus, 15 months of use and a MWI of only 98? These things are going to last longer than I will. And I have all my games, user folders, and page file on them, everything but my media. Makes me curious what will start killing SSDs after the next 10 years or so, surely some other components of theirs will start to fail before the NAND ever wears out. Power surges or dirty power sources maybe?

[Vapor]

Ah, didn't know Intel's tool could do it, I don't have Intel SSDs in RAID. Good info to have.

You only had ~4.1TB of host writes (1.36 * 3) in the past 15mo across three drives, makes sense MWI didn't go down much

[deathman20]

OOoh guess im going to check that out tonight!

[B.A.T]

Hello everyone!

I will be joining you with my M4 like Anvil announced earlier this week.

It's brand new out of the box and below follows some smart info before any benchmarking and testing has started.

CDI


GsmartControl


I will be testing the M4 from my main rig (look in my signature) and to make sure trim and smart info is working I've installed my os to a single drive (C300 256GB) for the duration of the test. The M4 has also been flashed with the latest firmware(0002).

Some performance numbers (The m4 has been quickformated and trimmed between each benchmark):

AS SSD


CDM


ATTO


I talked to Anvil earlier today and he wants me to wait until the C300 is ready so we can start at the same time.

Is there anything else I should think of?

B.A.T

[sergiu]

Makes me curious what will start killing SSDs after the next 10 years or so, surely some other components of theirs will start to fail before the NAND ever wears out. Power surges or dirty power sources maybe?

I had the same question for some time and I also believe power sources will be the major death cause for SSDs. I was even thinking to buy an oscilloscope to check periodically the power supply for signs of failures.

[Ao1]

Hello everyone!

I will be joining you with my M4 like Anvil announced earlier this week.

[bulanula]

Thank you so much for providing the C300 and the M4 for the testing. I am 90% sure that the C300 will perform much better etc.

This thread is really starting to heat up now