Intel SSD Endurance

[Ao1]

Intel has recently posted a number of quite informative presentations on their ssd’s. One covers ssd endurance. I really didn’t appreciate the extent that endurance can be improved for a relatively small sacrifice in capacity. I was also surprised by the difference in endurance between the 80 & 160 GB drives.
http://intelstudios.edgesuite.net/id..._MEMS003/f.htm
In essence it states:
X25-M 80GB - 100% random host data 4k writes = 7.5TB
X25-M 160GB - 100% random host data 4k writes = 15TB
X25-M 160GB - 100% sequential writes = 370TB
The X25-M 160Gb has 7% spare area. If this is increased to 17% by reducing the partition size (on a secure erased drive) the 4K random write capacity extends to 42TB.
The other presentations can be found here:
http://www.intel.com/design/flash/na...ldocuments.htm

Edit: This is also worth a watch:
http://www.micronblogs.com/2009/10/h...rom-tiny-nand/

[lowfat]

So if I perform a secure erase on my drive. Then create a 70GB partition on my drive it will last that much longer?

[tiro_uspsss]

grrr.. i'm trying to wrap my head around this

[Ao1]

So if I perform a secure erase on my drive. Then create a 70GB partition on my drive it will last that much longer?

Yes.

Interestingly the added endurance of SLC on Intel drives is not just due to the better write endurance of SLC cells. The E (slc) drives come with 27% spare area. The M (mlc) versions come with 7% spare area.

I should point out that the 4k endurance quotes above are worst case scenarios. The host write endurance that Intel quote at 20GB per day for 5 years is based on random endurance writes. This is how they have calculated a write endurance of 36.5TB. These figures are conservative. (Edit; i.e. Worst case 4k writes only = 15TB. Best case sequential writes only = 370TB. Typical desk top usage pattern as a mixture of 4k writes and sequential writes = 36.5TB)

What Intel explains in that presentation is the possibilities to trade off spare area for extended endurance for more demanding enterprise applications.

To do this the blocks that are set aside must be free. (A new drive or a secure erased drive) Set a smaller partition and never write to logical block address above the partition and endurance can be tripled with a small reduction in capacity.

It is possible to monitor how effective the change in configuration is for a particular usage pattern by monitoring S.M.A.R.T data.

[Ao1]

grrr.. i'm trying to wrap my head around this

Watch the Intel presentation, it exlains it quite well.

[One_Hertz]

So an x25-m 80gb can be killed with iometer in under 40hours. cool. I wonder what this puts indinlix at.

[Razmatazz]

I have a couple of questions on this topic. I found the Intel presentation interesting.

Question:

-Is it just the writes that degrade an SSD over time? What about reads?

-Does creating a smaller partition (increasing the spare area) increase performance also?



I am using the SSD as a boot drive and for running my main applications. I have all of my media (documents, pictures, music, video, etc) on a HDD.

Question:

-Would creating a smaller partition on my SSD given how I am using it be of any significant benefit?

[Computurd]

So an x25-m 80gb can be killed with iometer in under 40hours. cool. I wonder what this puts indinlix at.

using 80gb test file?

[Spoiler]

Very interesting stuff! With the exception of the enterprise user and iometer freaks, I think most everyone else owning an Intel SSD has nothing to worry about through the life of their product.

I think adding the spare area is actually a nice trade off, but only for certain people. Personally, I think most people will have already moved on to a new computer(w/ new drives) before they would see any benefit of modifying their spare area size.

[Razmatazz]

How about the question of any changes in performance? As we can see, there is a definite increase in endurance. This it would seem, for most of us, would not be necessary. I would be much more apt to go through the hassle of increasing the spare area if there would be a performance boost.

[Ao1]

With a G2 drive and trim you will not notice a difference. Maybe it helps on degraded G1 drives or drives in raid 0 but I have not tested so I don't know. If it does make a difference it will most likely only be seen in benchmarks

[alfaunits]

I find the maximum endurance puzzling. 370TB for sequential on a 160GB drive - thats about 2000 times? What are they doing with the other 3000+ cycles of MLC??

[Computurd]

@spoiler--lol iometer freaks! i am one for sure, however if you make a test file you can use that file for as many tests as you want. as long as you arent testing writes with iometer you will be fine. i couldnt tell you how fast my drives write, i am scared of write testing!

[One_Hertz]

using 80gb test file?

It doesnt really matter what test file. I was just saying since an X25M 80gb can 4k random write at around 50MB/s, it will die in about 41-42h according to Intel.

[Levish]

It doesnt really matter what test file. I was just saying since an X25M 80gb can 4k random write at around 50MB/s, it will die in about 41-42h according to Intel.

15,000,000MB (15TB worst case scenario) of 4k writes @ 50MB/sec =

(15,000,000/50MBpsec)/60forHrs/24fordays

300,000sec or 5000hrs or nearly 208days of non stop 4k writes which doesn't even include the time spend erasing you'd have to do between each 80GB.

Maybe I put a decimal in the wrong place?

[alfaunits]

300.000 seconds = ~83 hours, not 5000 hours. You missed a division by 60 for minutes.

[One_Hertz]

300.000 seconds = ~83 hours, not 5000 hours. You missed a division by 60 for minutes.

yep 83 for the 160gb, 41-42 for the 80gb

[Biker]

yep 83 for the 160gb, 41-42 for the 80gb

I wonder if they would accept an RMA?

[Ao1]

Don’t forget that the controller will write combine the files wherever possible. Writing 15TB of single 4k files that erase a complete block each time would not be so easy.

[[XC] Oj101]

A bit off topic, but where could I find the endurance figures for a SuperTalent 16GB MLC SSD? I'm seriously considering four or them.

[Levish]

300.000 seconds = ~83 hours, not 5000 hours. You missed a division by 60 for minutes.

whoops

[F.E.A.R.]

We have 2 postville-user´s in our forum with damaged sectors
I heard that the G2 only have 3.000 erase-cycles (G1 10.000). Is it true?

[EniGmA1987]

Dont know where you heard that FEAR. The Intel G2 drives use the new Micron 34nm chips, and Micron said in some news article when the chips launched that it had the highest endurance of any MLC. Whether that is true or nbot I have no idea, and I have no idea what the official numbers are. But if regular grade nand chips for MLC are 10k write cycles, then the 34nm should be at least that?

Again, its just a news article from the manufacturer, so of cource they say big impressive things about it.

[alfaunits]

We have 2 postville-user´s in our forum with damaged sectors
I heard that the G2 only have 3.000 erase-cycles (G1 10.000). Is it true?

Did they post the SMART data? Could you, please, post the link, I am interested in seeing what the SMART data would show on such a drive.
How much did they use them though... without IOmeter writing random data on purpose, it would take ages.

[One_Hertz]

I heard that the G2 only have 3.000 erase-cycles (G1 10.000). Is it true?

I heard this as well.