SSDs versus spinning-rust drives
Ryan Petris
ryan at petris.net
Mon Dec 2 13:03:46 MST 2019
On 2019-12-01 22:32, Jim wrote:
> I've seen the limited write cycles mentioned here. I can't help but
> wonder if it's possible to make SSDs with more write cycles. If it is
> possible, I'm guessing it becomes much more expensive to do so.
It depends on the type of SSD. SLC SSDs are the simplest and least-dense
type, though because each cell only stores a single bit, they can handle
much more degradation and therefore have many more write cycles than
others. I believe this is what most Intel Optane SSDs use, which
explains their low capacity. MLC is next, storing two bits per cell, and
is what I believe Samsung Evo drives use; they generally have a much
higher TBW specification than other SSDs on the market (but lower than
Optane) though are generally much more expensive. TLC SSDs, storing
three bits per cell, is the majority of what you're going to find on the
market right now. QLC SSDs, storing four bits per cell, are few and far
between, though there are some such as the Intel 660p which use it.
As more bits are stored per cell, write endurance, price, and write
performance generally go down, however you get a better cost/gb ratio.
Good drives will mitigate any write performance issues by having a DRAM
cache and/or have a portion of the SSD act as an SLC ssd, moving it to
the TLC/QLC portion in the background after being committed to disk.
I've even seen a few disks apply compression to the data prior to
committing to disk, reducing the amount of space actually used and
therefore prolonging the life of the SSD (though this won't help if the
disk is encrypted).
That said, the TBW ratings are what the manufacturer is comfortable
warrantying the drive at; it isn't necessarily going to burst into
flames after writing 1 bit over the TBW rating. It will probably last
many many more write cycles after that. Even so, most will never run
into the TBW rating anyway, even for the minuscule TBW ratings of QLC
drives.
More information about the PLUG-discuss
mailing list