TR Forums

So, I was reading Anandtech's review of the new Crucial/Micron M500 SSDs, and the writer was lamenting the fact that since people are using higher-density NAND for various reasons, there are various caveats, usually either reduced endurance (as in the Samsung 840's TLC NAND) or reduced performance (as in the Crucial M500).

So, then, if that's they case, why hasn't anyone made a desktop-specific 3.5", 5.25", or even external (via 1394, UASP3, eSATA, or Thunderbolt) SSD using lots of smaller (in terms of capacity), lower-density NAND? I know that there are limits to how many channels the SSD controllers have, but at least some of the controllers available support being used in a modular fashion; OCZ does something similar in a sort of "internal RAID" on their Revodrive PCI-Express SSDs.

In theory, then, it seems like this would allow the manufacturer to make a high-performance, high-capacity SSD at a relatively low materials cost, a savings which they could then pass on to the consumers as a better price-per-gigabyte. They'd sell a bundle of them, and it'd enable people to move away from rusty spinny drives with their awful random performance.

Is there some technical hurdle I'm just overlooking in my ignorance, or is there some other reason this isn't done?

Newer, higher density chips are cheaper than lots of small chips.

JustAnEngineer wrote:

Newer, higher density chips are cheaper than lots of small chips.

Well, I suppose if there's no backstock of lower-density chips, which was sort of my assumption that didn't get stated.

In any case, even if there isn't, wouldn't there be an advantage in terms of both performance and reliability to using more lower-density chips? It just seems like there's always a trade-off, and that's fine, but since there ARE advantages to using lower-density NAND, it seems strange to me that nobody is making use of those advantages.

I guess my whole question is, why do we continue to restrict SSDs to the 2.5" form factor when they could easily be produced in larger sizes? Surely a larger form factor would simply designs, no?

As far as I can recall the product you want does exist, but in Enterprise drives using either SLC or MLC drives with enhanced reliability. In terms of formfactor: 2.5" vs 3.5" there is a limit concerning how many chips each controller can manage, and the length of the traces from the chips matter in terms of latency, so to make a 3.5" SSD you're adding more controllers, a tonne more latency and A LOT more complexity. The result would be an SSD with the performance nearing a spinning disk for several times the comparable cost OR a large SSD that's not affordable to any client due to cost-to-manufacture.

zzz wrote:

As far as I can recall the product you want does exist, but in Enterprise drives using either SLC or MLC drives with enhanced reliability. In terms of formfactor: 2.5" vs 3.5" there is a limit concerning how many chips each controller can manage, and the length of the traces from the chips matter in terms of latency, so to make a 3.5" SSD you're adding more controllers, a tonne more latency and A LOT more complexity. The result would be an SSD with the performance nearing a spinning disk for several times the comparable cost OR a large SSD that's not affordable to any client due to cost-to-manufacture.

Hmmh. Well, maybe so.

Top-end 2.5" SSDs are already limited by SATA anyway, and plenty reliable, so I don't really know what I'm looking for. I guess I'm just frustrated with the $/GB of current SSDs.

auxy wrote:

zzz wrote:

As far as I can recall the product you want does exist, but in Enterprise drives using either SLC or MLC drives with enhanced reliability. In terms of formfactor: 2.5" vs 3.5" there is a limit concerning how many chips each controller can manage, and the length of the traces from the chips matter in terms of latency, so to make a 3.5" SSD you're adding more controllers, a tonne more latency and A LOT more complexity. The result would be an SSD with the performance nearing a spinning disk for several times the comparable cost OR a large SSD that's not affordable to any client due to cost-to-manufacture.

Hmmh. Well, maybe so.

Top-end 2.5" SSDs are already limited by SATA anyway, and plenty reliable, so I don't really know what I'm looking for. I guess I'm just frustrated with the $/GB of current SSDs.

With respect to 'reliability' of higher density flash, the controller and firmware engineers regularly come up with ever more clever techniques to minimize write amplification and keep the drive healthy. Sandforce's controllers are a good example of this, but everyone does it.

With respect to size and speed- there's nothing directly stopping SSD producers from fielding a product that has more than the eight or ten channels to address more NAND packages, except the disproportionate increase in development and manufacturing costs versus the likely demand. It's far more economical to stripe together multiple existing controllers on a PCIe card to achieve the same basic performance.

However, we can expect that to begin to change with SATA Express along with Moore's law as it continues to apply to transistor density over time. Getting 1GB/s out of non-volatile storage will soon become trivial, and with the major controller manufacturers really starting to focus on maintaining transfer rate consistency, like in Intel's 3700 and OCZ's Vector, local mass storage will cease to be a major performance limiter for all but the most storage-focused of workloads.

Airmantharp wrote:

With respect to 'reliability' of higher density flash, the controller and firmware engineers regularly come up with ever more clever techniques to minimize write amplification and keep the drive healthy. Sandforce's controllers are a good example of this, but everyone does it.

With respect to size and speed- there's nothing directly stopping SSD producers from fielding a product that has more than the eight or ten channels to address more NAND packages, except the disproportionate increase in development and manufacturing costs versus the likely demand. It's far more economical to stripe together multiple existing controllers on a PCIe card to achieve the same basic performance.

However, we can expect that to begin to change with SATA Express along with Moore's law as it continues to apply to transistor density over time. Getting 1GB/s out of non-volatile storage will soon become trivial, and with the major controller manufacturers really starting to focus on maintaining transfer rate consistency, like in Intel's 3700 and OCZ's Vector, local mass storage will cease to be a major performance limiter for all but the most storage-focused of workloads.

Thanks for your reply to my post!

I'm pretty excited about the future. Someone smarter than me once posited that we'd eventually abandon the idea of RAM and simply have the system work from nonvolatile storage. I definitely think that's in the future, but some ways off.

Still, I'm excited about the future of storage. I just wish SATA Express would go ahead and come on.

auxy wrote:

I guess I'm just frustrated with the $/GB of current SSDs.

The TLC 840 is pretty cheap in terms of $/GB and it goes regularly on sale. What is the frustration?

Flying Fox wrote:

The TLC 840 is pretty cheap in terms of $/GB and it goes regularly on sale. What is the frustration? ;)

Mostly trying to fit 2.5" drives into cases with no mounts for them. o(-`д´- ｡)

Or instead of trying to get an entire industry to put R&D into solving your annoyance, you could just get a 2.5" to 3.5" adaptor like everyone else does. Hell, some SSDs come with those for free.

I like these: http://www.newegg.com/Product/Product.a ... 6817996040

...but you can get them cheaper, too. Now ain't that an easier and quicker solution than waiting for a new SSD in a 3.5" form factor which wouldn't solve your immediate problem anyway?

Scrotos wrote:

...but you can get them cheaper, too. Now ain't that an easier and quicker solution than waiting for a new SSD in a 3.5" form factor which wouldn't solve your immediate problem anyway? :D

:lol:

I run a home business building, upgrading, maintaining, repairing, and networking computers; I'm rather familiar with these adapters (and in fact the specific ones you linked!) (´Д⊂ヽ

I was just being facetious. (╹◡╹)