Some Thoughts on SandForce's 3rd Generation SSD Controller
by Anand Lal Shimpi on January 12, 2012 9:47 PM EST
I spent a few minutes talking to SandForce in between meetings today and got a general idea of what a third generation SF controller might look like. The SandForce folks are proudly displaying their new LSI business cards at the show, but they are also hard at work on a 3rd generation controller.
The SF-2000 series controllers are already limited on the sequential side by 6Gbps SATA as well as the ONFI 2.x interface. Both need to be addressed to improve sequential performance, which we likely won't see until 2013. In other words, I wouldn't expect to see improvements in highly compressible sequential transfers with the 3rd generation SF controller.
There is tons of room for improvement in small file, random read/write performance. Plan on seeing a significant improvement there from the third generation SF controller. This is mostly a function of adding extra processing power in the controller itself. I'm hearing numbers as high as 2x current random IO performance.
SandForce is also planning on improving write speeds when dealing with incompressible data. Larger internal data structures, a faster processor and some other firmware architecture tweaks can enable better performance here. SandForce tells me that improving performance when dealing with incompressible data is its top priority at this point.
Finally there's talk about looking at other interfaces in addition to SATA. It's possible that we may see a PCIe version of SandForce's 3rd generation controller. This makes a lot of sense, especially as we march towards an eventual SATA Express (PCIe based) standard. Companies with native PCIe based controllers today will have a better chance of influencing the spec and working through the kinks in their own products as the spec is finalized. SandForce also indicated that being able to efficiently aggregate multiple PCIe based controllers would be a very important point to focus on.
As always, SandForce will launch its third generation SSD controller for the enterprise first but we'll probably see client drives show up before that. Enterprise validation takes a lot longer than client, hence the lag in availability despite an earlier introduction.
SandForce isn't officially talking about timeframe for availability, but my guess is we'll know more around the middle of the year.
Facebook
Twitter
RSS
14 Comments
View All Comments
jjj - Thursday, January 12, 2012 - link
Any chance you asked LSI/SF about hybrid HDDs SoCs? Something native/inegrated not like what Seagate is using now.neotiger - Thursday, January 12, 2012 - link
There was no performance improvement in IOPS from the 1st gen to 2nd gen, which was very disappointing.Hopefully they can address that this time.
Also they need to make their SSD crash safe. Right now a crash will lead to data loss in a SandForce SSD because it doesn't come with capacitors. I know their "enterprise-class" products offer capacitors, but charging 3X more expensive just to make their product carsh-safe is beyond ridiculous.
If Intel can make their consumer SSD crash-safe, then so should SandForce.
MrSpadge - Friday, January 13, 2012 - link
That's a matter of building the SSD, not the controller. Anyone could add some capacitors to provide a failure safeguard.iwod - Thursday, January 12, 2012 - link
We need standardizing on bootble PCI Express Storage ASAP. SSD Speed is contantly being bottleneck by its interface SATA which was designed with HDD in mind.FaaR - Friday, January 13, 2012 - link
You're not bottlenecked FOR REAL even by 3gbit/s SATA, and much less 6, because the vast vast majority of file transactions in computers use very small buffer sizes that don't run anywhere near the transfer rate limit of the interface.iwod - Saturday, January 14, 2012 - link
Last time Anand revise his benchmarks it was shown Seq is still up to 60% of our daily work loads, with Random Read being the 2nd most important metric. Random Write has MUCH less impact then we anticipated.Boogaloo - Friday, January 13, 2012 - link
Is the disk even the bottleneck in real world applications anymore? I wish anandtech still did tests like boot or application load times, but from what I've seen elsewhere on the net all recent generation SSDs perform about the same in non synthetic benchmarks, despite ostensibly varying widely in performance.Even reviews for the incredibly expensive PCIe based disks like the revodrive and fusion io only showed impressive results in server type tests, they fell into the same pile as the rest for general use.
Maybe it's the random IO performance holding everyone back, and this drive will really break away from the pack, but I'm not hopeful for anything special here.
piroroadkill - Friday, January 13, 2012 - link
This reason is why I bought an Intel 320 series 120GB after having my Vertex 2E fail on me.Reliability over all out speed, and you really don't notice the difference anyway...
LtGoonRush - Friday, January 13, 2012 - link
Back in March (when ONFI 3.0 was finalized) Sandforce said they were "proud supporters of the ONFI specification" and intended to have a controller supporting ONFI 3.0 in 2012. Do you really think we won't see ONFI 3.0 support on the SF-3000? It seems like a pretty big risk to wait until SF-4000, as by then there will probably be a lot of NAND around that Sandforce controllers can't use.Beenthere - Friday, January 13, 2012 - link
...OCZ and all the rest of the SSD makers invest enough time qualifying their consumer grade SSDs so that they are reliable and free of compatibility issues. Until then they are just marketing half-baked crap to the naive and gullible.