LSI SandForce SF3700 Flash SSD Controller Questions Answered – LSI AIS 2013 Update


Since our discovery of the next gen LSI SandForce flash controller at Computex earlier this year, it has been the talk of the storage world.  Codenamed Griffin, ADATA displayed both enterprise and M.2 ultrabook SSD prototypes of the controller at their event display without mention or word to anyone, almost daring the storage world to take note.  It was a display most likely unsupported by LSI SandForce, and while it didn’t officially display the controller, it introduced the storage world to a speed never before seen from a single SSD controller – 1800MB/s.

LSI M.2 SATA SF-3700 Sample

Since this revelation, we have seen this controller shed its ‘ Codename Griffin’ skin and receive official validation as the new LSI SandForce SF3700 Series flash controller, indeed capable of top SSD performance speeds of 1.8GB/s.  There are a ton of questions that remain, however.  Why hasn’t it been released?  Why haven’t we seen thorough performance benchmarks?  Are there heat issues with the controller?  Is it possible, like folklore describing the Griffin with its head of an eagle and body of a lion, that the SF3700 family is myth and something that we may just never see? Let’s tackle these questions one by one, all the while showing you some early examples of what various LSI SandForce partners have to offer.


If all claims of the LSI SandForce SF3700 SSD controller are true, this will revolutionize the SSD industry and send competitors back to the drawing board, inevitably resulting in countless hours of frustration as they try to better, or maybe even just equal, what LSI SandForce is bringing to the table.  The LSI SandForce news release a few days ago speaks to such things as LSI’s enhancement of SHIELD error correction and RAISE Technology, both of which are features that competitors are going to find tough to match, but that is not all that LSI has up their sleeve.  Perhaps their second greatest step forward has to do with DuraWrite Virtual Capacity (DVC) where, depending on data entropy, DVC can increase the storage capacity of NAND by up to 3X.  In layman’s terms, a 128GB SSD can realize storage of as much as 384GB of data stored.

Toshiba SATA SF-3700 With A19 MLC 64Gb Memory

You may have noticed that we stated ‘second greatest’ step forward; the greatest of course being the incredible data tansfer speeds of the SF3700, never before seen through a single flash controller. The SF3700 is capable of parallel read and write speeds of 1.8GB/s and LSI claims that data entropy doesn’t slow this controller down, much like that SSD enthusiasts have whined about since the release of the first SF-1200 series SSD. Of course these speeds are realized only through native PCIe 2.0 x4 travel or better, but that’s the beauty of the SF3700; it is compatible with both SATA and PCIe SSD designs.  This brings us to a topic of confusion with M.2 design that we should clear up before moving on to the present state of the SF3700 and release dates.

LSI Display SF-3700 Working Display


There is a great deal of confusion with respect to M.2 standardization right now and it is causing a great deal of heartache with system purchases, as well as understanding just how we have standardized anything.  First and foremost, the word M.2 is derived from the M.2 connector that can be used for any number of connections to things such as Wifi, 3G/4G/5G, BT, SATA SSD, PCIe x2 SSD and PCIe X4 SSD.  They are all different but the same connector can be used for all.  The physical connector is standardized but the connection it makes to a system board is definitely not.  Very unhappy Sony fans have recently realized this the hard way when their new Sony VAIO Pro 13 arrived and was only capable of SATA speeds and not the typical 1.2GB/s of the Samsung XP941, that of which we have ripped apart in detail.


A SATA M.2 is only capable of SATA speeds and cannot be switched out with a native PCIe M.2, such as we see in the XP941 that we reviewed, and will see in the SF3700 M.2 drives.  Just because the SF3700 controller is capable of either SATA or native PCIe data throughput doesn’t change things, the route of travel (SATA or PCIe) is distinct and ‘pin-out’ not switchable by a simple SSD replacement.  To make things somewhat easier, the ONLY way we can ever achieve the 1.8GB/s performance is through a PCIe 2.0 X4 connection or greater, this represented on your M.2 SSD by a single notch (M key) in the connector.  SATA can be run through this be we haven’t seen it as of yet, but why would we.  It is like putting a Ford Focus engine in the shell of a Ferrari.


Now to go one step further, we will see M.2 connections with two notches in the connector (M and B Keyed). These are capable of both SATA (550MB/s) and PCIe X2 (1GB/s) connections, however again, it has to be one or the other at the connection stage at this point in time.  One cannot switch a SATA M.2 for the PCIe, even in this case, as we are told from industry professionals.  It is still one or the other.  A very curious pattern we are seeing at the LSI AIS, however, is that most M.2’s are being displayed as dual keyed (M/B keys) which would never be able to reach that precious 1.8GB/s performance that we crave so.


A quick look around the show floor will verify that EVERY active SF3700 flash controller demonstration has a very prominent heat sink over the chip itself.  Right now, it runs hot and many believe that this heat is a by-product of such performance.  Our close LSI contacts say otherwise, and actually, the controller now runs hot with performance running lower than they are happy with.  This is why the hands of every AIS participant have been tied and they are not allowed to demonstrate write performance, or even allow our quick benchmark through software on our handy-dandy flash drive.

ADATA M.2 SATA With Heatsink

The truth is that, right now, there are three firmware teams, totalling just under 100 people in the US and Asia, all working together to optimize the firmware.  Their goal is to increase performance while decreasing temperature and LSI clearly stated that we wouldn’t be standing in a room centered around the SF3700 if this couldn’t be achieved.  All are certain this is just the next step, which will then be followed by validation and that brings us to a very conservative release date estimate.

ADATA SF-3700 Notebook PC In Case W Heatsink


The company line on release of SSDs containing the SF3700 controller is Q2 2014, however, be ready for the probability of it occurring in Q1 with some great examples being demonstrated at CES.  Word on the street has always been that enterprise will see the benefits first, but we are not so sure and a manufacturer or two here at the show might have said “not if we can help it”.  Looking at the show floor alone, ADATA, Kingston and Avant all have the controller in hand and all represent he consumer space, Avant being the unheard of ‘jewel’ of SandForce tradition that has manufactured and branded SSDs for countless LSI SandForce retail, enterprise and custom solution partners.

Avant HHHL PCIe 2.0 X4 SF-3700


A bit long perhaps but we wanted to clarify some of the rumors that have been floating, as well as bring forward updates that we believed are the meat and gravy of LSI AIS 2013.  Stay tuned in coming months because you have to believe the race is on to see who will get the first SF3700 controller in hand for independent review.  Our record in the M.2 (and PCIe) review space has been pretty strong as of late with some very exclusive reports on the most viable competitor, the Samsung XP941 native M.2 PCIe SSD.  Mark my word though that, if all pans out with the SF3700, even a force as large as Samsung will be in the boardroom trying to equal, or better this new flash controller.



  1. blank

    The race it’s coming!!!

  2. blank

    I really enjoy your site. Sometimes I find little things that distract me. An example would be: ” In laments terms, a 128GB SSD can realize storage of as much as 384GB of data stored.” . I believe you mean “layman’s terms”.

  3. blank

    Thats a big controller chip!

  4. blank

    Why no PCIe 3.0 support?

Leave a Reply

Your email address will not be published. Required fields are marked *