Most SSDs Now Under a Dollar Per Gigabyte 187
crookedvulture writes "SSD prices continue plummeting. In just the past quarter, street prices have fallen by double-digit percentages for most models, with some slashed by 30% or more. We've reached the point where the majority of drives cost less than a dollar per gigabyte, and that's without the special coupon codes and mail-in rebates usually attached to weekly deals. Lower-capacity drives seem more resistant to deep price cuts, making 120-256GB offerings the best values right now. It's nice to see a new class of devices go from prohibitively expensive to eminently affordable in such a relatively short amount of time."
Still need a big data drive in most uses (Score:3, Insightful)
Still need a big data drive in most uses
as 120GB-256GB is small for some uses and the cloud is slower and ISP data caps suck.
Some good values. Read reviews to avoid the duds. (Score:5, Insightful)
Newegg has sales all the time. I've picked up some great drives for a steal. Non-crap 120-128GB devices for less than 80 dollars is a shock.
The price of flash is imploding and the quality of the drives/controllers/firmware is improving quite a bit. The latest crop of devices are far better in terms of speed and reliability so don't bother getting an older model to save money. Only pick up a latest gen device. You can find them cheap too, don't worry.
The latest crop of sandforce based drives are fast and cheap and seem pretty reliable. (The new firmware really makes them shine)
Intels are still more expensive, but are generally the most reliable of the bunch.
I picked up an OCZ vertex 4 128gb for less than 80 bux and it gave my laptop a whole new life.
Re:Don't let it fool you (Score:5, Insightful)
Despite all the, errm, uninformed ladies and gentlemen responding with kneejerk reactions, smaller process sizes really do reduce the program/erase endurance of NAND flash. At 50nm, MLC was at about 10,000 cycles. 34nm took us down to about 5,000 cycles. 25nm took us down to 3,000 to 5,000 cycles, which is where we're at now. So, technically, we are reducing cost at the expense of reliability.
There are mitigating factors, however. Over the same timespan, SSD controllers have improved, substantially reducing write amplification, and capacities have increased, preventing the total *writable* lifespan of drives from decreasing.
As an example, a 60GB drive good for 10,000 cycles with a write amplification factor of 2.0 has a total theoretical write lifespan of 300 TB. On the other hand, a 120GB drive good for 3,000 cycles with a write amplification factor of 1.1 has a theoretical write lifespan of 327 TB. Despite having less than a third of the "reliability" (on a cell level), the drive can actually handle slightly MORE activity overall.
It will always be a balancing act between cost and reliability when it comes to SSDs. As compared to Single Level Cells (SLC), Multi Level Cells (MLC), used by all consumer drives, has a tenth the endurance, but half the cost (storing two bits per cell rather than one). Basically, SLCs store data by trying to differentiate between two voltage levels: high, low. MLC increases that to four states (high, medium-high, medium-low, low). The reduced endurance is because it becomes harder to differentiate the levels sooner. Triple Level Cells (TLC) is starting to show up, and this stores three bits per cell using eight states. It helps density, but once again, at the cost of endurance.
This might be a good time to point out that cell-level "reliability" has no real bearing on the reliability of the entire SSD. Reduced cycle endurance means your drive will wear out faster, but it will still take years to wear them out (if ever), and when they do, they don't lose data, they just stop being able to write. If you're having your SSDs just up and die on you out of the blue, that has nothing to do with the trend towards decreasing write endurance.
Yay for SSD boot drive. (Score:5, Insightful)
Just did a re-install about a month ago: 128GB adata SX900 -- which newegg now has for $15 less than I paid (always happens) -- on a 3+ year old system.
Best. Upgrade. Ever.
12 second boot instead of 45 seconds (not that I reboot much) but the big win: lag is nonexistent. Disk intensive stuff like browsing/picking through my heavy photo catalog just flies. Most of my stuff is, of course, still on spinning drives, but key apps & data, like email and photo libraries I'm working with are on the SSD. Actions that used to take several seconds (per photo) now are nearly instantaneous. Full-text searching through email is a lot faster. Sleep/Hibernate is practically instantaneous. $100 is nothing for not having to wait a few seconds (every few seconds!) when doing photo work. I make backups of critical data onto multiple spinning disks, regardless of what kind of disk I'm using, so reliability isn't a concern. I wish I took the plunge sooner.
Re:Don't let it fool you (Score:0, Insightful)
Completely filling the disk and deleting it is a pretty easy life for an SSD. Creating lots of small files, such as repeated compilation, can easily kill them within a year.
Absolute fucking hogwash. No, strike that, not hogwash, just downright utter FUD.
As someone who has used SSDs exclusively in FreeBSD servers for OS disks (that means root filesystem, swap, /usr, /var, and /tmp -- where all world rebuilds happen), I have only seen two of our drives result in 2 reallocated NAND pages: and that's over the course of 3 years, using an OS version that lacks TRIM support. The drives are Intel X25-M 80GB drives, which are MLC-based.
When it comes to SSDs, there is absolutely nothing different about creating many small files vs. creating one big file. The filesize doesn't matter; the FTL doesn't give a shit about "files": it cares about LBAs. Specifically, 10,000 4KByte files are going to have the exact same effect on the drive as one 40,000KByte file. The FTL is going to have to map just as many LBAs, although I will admit (depending on the FTL implementation) there may be less FTL map entries on the larger file (depends on of the FTL map is implemented as start/end LBA ranges or a true 1:1 LBA map).
I wish people unfamiliar with storage subsystems and storage technology would stop spreading this fucking nonsense. Those of us who are have to go to great lengths to undo the bullshit folks like you spread. There are other reasons to be "worried" about SSDs, but NAND lifetime isn't one of them. Wear levelling algorithms work extremely well with present-generation SSDs.
Comment removed (Score:5, Insightful)