Intel has replaced its mainstream X25-M solid state drive with the 320, which is more than two times faster doing sequential writes. The 320 uses third-generation 25nm process technology, unlike the 510, which Intel announced earlier this month and which relied on older, second generation 34nm technology. The smaller process …
Point of order
Smaller process technology also means less available write cycles per cell. With sufficient shuffling all that under the carpen ("wear-leveling") you won't notice until the warranty runs out, but traditional disk might just outlast MLC. SLC will be a tad harder, even in smaller process technologies.
Re: Point of order
Without figures, surely speculation is a bit meaningless? I do hear what you're saying and so figures become important, but if I say "It could be more reliable than a spinning drive, it could be less reliable", it really doesn't say very much at all.
I'd like to see MTBFs speak for themselves.
It's not speculation.
You're right that the remark is low on actual numbers--the article didn't touch on it at all, which is a bit of an oversight IMO. From memory (and from anandtech, actually), you can have about 100k write cycles per cell in the previous generation's SLC (what was that, 45nm? 60?) and 10k in MLC. That's dropping to 5k or even 3k for MLC, going down to 25nm. (As an aside, anand's pooh-poohing takes his write volume average and counts years before every single last single cell can no longer reliably hold a charge, rendering a number that's both staggering and utterly meaningless.) The problem with comparing this to hard drives is that the failure modes are different, and that MT[BT]F for SSDs hinges on write volume, not hours of spin on the axles. The point is that not only are the failure modes changing, but for sufficiently small process technology MLC you might end up with a usability window that's far more like the warranty period than we're used to. I don't think anybody has really realised yet what the implications are, so it's still mostly an unknown. As in, something to watch for and deal with as it arises.
The 25nm writes are about 3,000. The older 34nm was about 5,000. Wear leveling and write amplification, being as good as it is (in SandForce controllers at least), means that you could optimally write a total of 360TB to the flash drive before you see cells give out. That's 3.8MB/s of writes, every second, for 3 years. Since OCZ employs 23% or 13% (depending on the "E" series or not) of extra non-user-accessible storage, and from what I understand, the Intel drives have 7%, that 360TB figure is quite obtainable. Even if you assume you'll only pull 2000 writes per cell on average, you're still looking at 2.5MB/s for 3 years.
Needless to say, the MTBF is likely going to be due to failure of the controller chip(s) or the like rather than cell wearing. Even an OCZ drive can withstand an entire (single) flash chip death.
That's 3.8MB/s of writes, every second, for 3 years
So... a vista machine is a no-no then...hang on a sec I need another restore point... and of course an obligatory search index.. back in a mo, maybe..
Another difference between the 510 and the 320 is that the 320 carries a supercap (or some sort of energy storage, haven't taken one apart yet) to protect volatile memory in the case of power loss. Some may not think this is important, but it is. And the 320 is also by far the cheapest SSD on the market with any sort of power loss protection.
It's also interesting that you're comparing the 320 to year old drives (And the OWC, which is not exactly the most stellar example of a Sand Force 1200 controller-based drive.) Good comparisons would be the Micron C400 and something based on the Sand Force 25xx controller, both of which are faster in just about every respect... but somewhat more expensive as well.
Will there be a 1.8" version?
Different usage model
Everyone seems to assume that SSD's are being used for storage when in fact most of the gains of SSD (Most importantly access times) are for OS and programs which change less frequently. The amount of write cycles to SSD's will likely be *FAR* less than a 2TB spinning disk.
I'd say the opposite
My storage contents don't tend to change much. Most files are created and then stay there - for example, MP3s, videos etc.
The OS disk on the other hand, well settings change, Windows hammers swap files etc..
Guess it depends on usage though.
Windows does more writes to the disk than I ever could: Swap File, Search Indexes, Restore Images. (and we must not forget program scratch files!)
Afterall Vista was not happy unless it was thrashing away in the corner.
Most of my data is write once read many, Being a keen photo/video-grapher and owning an HD camera means I have 2TB of Video data. once video is captured and editied that is it. There are occasional files that I regularly edit but they are usually small spreadsheet type files.
Hey this all looks great, but my burning question is - can you put several of these into a say RAID 5 array & is there issues with doing so? I have the M version and the Intel documentation talked about them not being suitable for RAID.
I believe the main downside of RAIDing SSDs is that no RAID controller I am aware of supports the TRIM command (with which the OS tells the drive when data has actually been deleted). Lack of this can impact performance as the drive fills up as it has to shuffle more data around when writing.
If you raid them, TRIM doesnt work
The functionality that allows an MLC SSD drive to clear up unused cells relies on getting information back from the driver about where data is written during the cleanup process. With a RAID setup the information returned is about the virtual RAID disk, not actual physical disks. This means the clean up doesnt happen, and the disks in the RAID will develop slow write performance over time as every write becomes a read, calculate and write action.
But does it work?
Yes, fantastic much improved spec.
But does it work?
I'll let some other poor bastard find the bugs though.
The capacity is almost enough to get one...
If what the above people are saying about lifespan is true, I think I'll be sticking with my old style disks a while longer.
I think the lack of RAID functionality is the killer for business side apps though.
Excusing for a second the Horrid Henry reference, I think MS should provide a list, or even a utility to map all the temporary stuff to another disk. e.g. Temp files, Paging, Logs, etc.
I know you can find it all, but it takes hours.
I recently got Superspeed's ramdisk on one of my 16GB 8943Gs, and I've put everything temporary I can find, on there on a 4Gig Ramdisk. My machine just f*cking flies. Everything happens literally instantly. 3 Gig for OS and stuff, 9 Gig for Cached disk files, and 4 Gig for TempDB etc...
Putting tempdb on a ramdisk is the most amazing thing you've ever seen. What it does to complex select intos and the like has to be seen to be believed.
- Facebook offshores HUGE WAD OF CASH to Caymans - via Ireland
- Justin Bieber BEGGED for a $200k RIM JOB – and got REJECTED
- Microsoft teams up with Feds, Europol in ZeroAccess botnet zombie hunt
- Review Bigger on the inside: WD’s Tardis-like Black² Dual Drive laptop disk
- Mexican Cobalt-60 robbers are DEAD MEN, say authorities