Tzupy wrote:Micron claims at least 72TB of writes until failure for this new C400 drive, the same as the C300, which they claim was conservatively specced.
I have doubts about the drive lifespan on the new 25nm process, considering the 40% lower (than the 34nm) NAND cell lifespan.
FWIW my Intel G2 has less than 1TB of writes on it and claims to have used only 1% of its lifespan.
I hope the NAND manufacturers won't eventually reveal that the published lifespans aren't true...
Given that uneven writes (imperfect wear leveling) would cause problems with a SSD well before the entire drive is worn out I would expect them to over-provision and conservatively spec a SSD.
If your R&D team, marketing, lawyers, etcetera didn't cover their but on this you'd be hearing about it all over the place.
Newegg reviews for C300 still remain overwhelmingly positive 10 months after release. I suppose we'll have a another year or two to go before we'll have in the wild failure numbers but did you see this quote?
Micron also mentioned to me that it writes more than 72TB of data to its drives during development, so this is a verified rating.
Notice they didn't say how much over 72TB.
Also remember that the more free space you leave for the controller to use for background operations the longer the drive will last see
http://intelstudios.edgesuite.net/idf/2 ... S003/f.htm around the 16 minute mark (slide 19) they took 160GB X-25M Gen 2 drives and formatted them differently and ran them to first failure in an accelerated write scenario single user wouldn't ever see
Code: Select all
Formatted TB Written
Capacity before failure
160GB 29
144GB 68
128GB 104
_96GB 150
The thing is write amplification means more writes internally than what is reported by the OS so extra free space significantly increases the life expectancy of the flash. The only thing Micron or any other SSD manufacturer has to do to offset 25nm flash lifetime issues is put more in (it offsets the cost advantage but it's worth it because the benefit isn't linear). And that's before you consider:
Improving the controller
Improving firmware
Increasing cache
Any Driver/OS changes to reduce writes
from 19 minutes page 21
* Write workload impacts WAF (Write Amplification Factor) and thus endurance
* Increasing transfer sizes of random write workloads generally decreases WAF and increases endurance
* (I'll paraphrase the 3rd bullet) Endurance varies by use and no one can definitively say how long a drive will last. Actual quote "Only way to really forecast what an SSDs endurance will be is to measure it under the targeted workload"
They made it clearer in the 2009 IDF but the presentation is no longer available on the Intel website so I had to pick from data in the 2010 IDF.
knowing all the ins and outs of SSDs and warranty policies I'm willing to take Marvell/Micron/Crucial at their word. If they say endurance won't be an issue on the C400 I expect that means for 95+% of us it won't be. I'm thinking they hedged their bets enough that you and I won't notice the negative difference between 25nm and 34nm flash.
Now to summarise the known data: