SandForce intros next-gen SF-2000 series SSD controllers

SandForce has quickly become one of the premiere controller providers in the solid-state storage market. Today, the company takes the wraps off of its next-gen SF-2000 series controllers destined for client (rather than enterprise) SSDs. Like the existing SF-1000 series, the new family is infused with a black box of encryption and compression technologies dubbed DuraWrite. This unique approach yields a low write amplification factor that should make more efficient use of the limited number of write-erase cycles offered by flash memory.

Speaking of flash, the new controllers work with MLC and SLC chips built using 2x-nm and 3x-nm fabrication technology. They also support the latest flash technologies: asynchronous Toggle DDR NAND and synchronous flash from the second-generation ONFI camp. The controllers can access this flash at up to 166 MT/s.

The move to finer flash fabrication tech is said to produce higher error rates and lower the write-erase endurance of individual memory cells. When we discussed this issue with SandForce at CES earlier this year, the company was quick to point out DuraWrite’s low amplification factor. It also touted RAISE, which has been around since the SF-1000 series and is best thought of as a RAID-like array of flash dies. According to SandForce, RAISE is well equipped to handle higher error rates at the flash level.

There are two members of the SF-2000 family. The SF-2200 is the big daddy, offering a 6Gbps SATA interface and eight memory channels, each of which has two data lanes. Users can expect sustained read speeds to hit 550MB/s, while writes to top out at an even 500MB/s. The SF-2200 is being targeted at enthusiasts, alongside an SF-2100 meant for the mainstream and entry-level markets. Although the SF-2100 offers the same features, expect less performance. The SF-2100 is limited to a 3Gbps SATA interface and only four memory channels (you still get two data lanes per channel). Interestingly, the SF-2100’s performance with 4KB random reads and writes appears to be on par with the SF-2200: 60k IOps with reads and 20k IOps with writes. Those are sustained rates; the controllers are purportedly capable of hitting 60k random write IOps in short bursts.

Stronger security is on the menu for these new controllers, which now offer 256-bit AES encrpytion. The presentation slides also tease an “extensive silicon roadmap” that includes products built for SAS, PCI Express, and USB 3.0 interfaces. Stay tuned for a preview of the SF-2200’s performance.

Comments closed
    • Anomymous Gerbil
    • 9 years ago

    “Speaking of flash, the new controllers work with MLC and SLC chips built using 2x-nm and 3x-nm fabrication technology.”

    Why does a controller care about whether the chips are built with 2x, 3x or XXXX fab tech?

      • cygnus1
      • 9 years ago

      It might mean that it can handle the increased error rate generated by the smaller node flash chips.

      • Firestarter
      • 9 years ago

      Operating/communication voltages and the like might differ, maybe timing as well. I guess it’s like RAM, but less standardized.

    • DPete27
    • 9 years ago

    I agree with those concerned about lifespan on the new fab size. I’ve seen some charts saying the new memory chips are somewhere in the 100,000 TB of writes as far as lifespan which sounds like a boat load, but nobody converts that into actual years for the average or enthusiast user. Are we talking 5 years, 10 years, 50 years? (Obviously usage varies, but a general number would be nice)

    Maybe the guys at techreport could help us out on this and clarify?…..hint hint.

      • Buzzard44
      • 9 years ago

      While many point out that the lifespan of SSDs is fine given wear-leveling algorithms and other tricks, I believe it should still be a serious concern.

      SSDs are still viewed by many, and probably most, as a technology of the future, which will become prevalent as price goes down. We’re still a good bit above $1/GB, which is well over an order of magnitude above the price of HDDs.

      As the process shrinks smaller and smaller, what will be drive life expectancy then? How long of a run can we expect from SSD technology? Seems like it might start to lose steam once it finally becomes mainstream.

        • Anomymous Gerbil
        • 9 years ago

        “Seems like it might start to lose steam once it finally becomes mainstream.”

        Well, there’s an unfounded claim if ever Ive seen one.

        Every time someone raises concerns about SSD lifetime, someone else replies with estimates of average write patterns, amplification factors, cell lifetime etc, that shows that for all-but-extraordinary use, your SSD will last tens of years before you lose enough cells for the drive to become useless. The estimated/average lifetimes are so long that even assuming constant reduction in cell life and process shrinks, you’re not going to have to worry too much about early drive death from cell falures.

        SSDs are here to stay, until the next you-beaut technology comes along…

    • NeelyCam
    • 9 years ago

    Did you guys see the AnandTech review of Vertex 3? Good god they are fast!!!

    • ClickClick5
    • 9 years ago

    Just throw two Raptors in RAID and be happy on the speed, space per dollar and no worries about write cycles.

      • UberGerbil
      • 9 years ago

      And you’re going to do that in your typical laptop exactly how?

        • ClickClick5
        • 9 years ago

        Mmmmm, point.

        Though I had this in mind.
        [url<]http://www.originpc.com/[/url<] I'm just giving SSDs about another 2-3 years before I buy one. There are to many little things that turn me away from them.

      • travbrad
      • 9 years ago

      You may be “happy on the speed”, but they are still 100x slower in random write/reads than an SSD. No amount of RAID will overcome the limitations of it being a mechanical device. As for the write cycles, with any realistic workloads they will still last longer than an average mechanical hard drive.

      On top of all that SSDs also use less power, produce less noise and heat, and use way less space.

      The one advantage Raptors have is price/GB, but I doubt most people have a 300-600GB system partition, so it’s often about the same price going with a combination of a regular mechanical+SSD vs RAIDed Raptors. Most people don’t need to read their mp3s at 300MBps, but WILL notice a 100x increase in random access speeds.

    • albundy
    • 9 years ago

    does this mean that the release on the SF-2500/2600 controller is on hold?

    • SuperSpy
    • 9 years ago

    EDIT: Dupelicate post =/

    • SuperSpy
    • 9 years ago

    Am I the only one slightly concerned over how they gloss over the lower write cycle of these newer chips?

      • flip-mode
      • 9 years ago

      Hopefully, you are. Even with the lowered write cycles the drives should theoretically decades or even a couple of centuries, according to Anand.

      [url<]http://www.anandtech.com/show/4159/ocz-vertex-3-pro-preview-the-first-sf2500-ssd/2[/url<]

        • hiro_pro
        • 9 years ago

        i read the article too but it seams to be a basic exercise in math and does little to address what really goes on in a hard drive. I have files on my boot drive that are barely written to. i also have temp files, page file, .ost and a hibernation file that are written to multiple times each day. how long do i have to run windows before one of those four files writes 10,000 times?

          • crazybus
          • 9 years ago

          When those files are modified the same NAND cells aren’t overwritten repeatedly. The drive’s firmware works to spread the writes over all the available free space on the drive, hence the large over-provisioning you see on enterprise-type drives that see a lot of writes.

          • flip-mode
          • 9 years ago

          There’s a part of the article that explains it and the feature is called “wear leveling”. I’m not well versed on it, but the simple explanation is that writes are not performed over and over again to the same memory cell. Rather, writes are performed to one cell, then the next, then the next, and so on. The drive’s controller – as I understand it – moves around you static files so it can write as evenly as possible to all the cells on the drive.

          Then there’s “write amplification”, which I don’t fully understand, but apparently the Sandforce controller has a write amplification factor of 0.6 or something, so that somehow, less data has to actually be permanently written to the drive than is actually sent to the drive – I don’t understand how that works at all, and I don’t really need to, I guess.

          At the end of the day, the point is that these drives should easily last for at least a decade, and in theory they should last 10 or 20 times longer than that before the drive exhausts the MLC’s write tolerance (although, there’s probably a list of other reasons that it will probably fail before that).

          • UberGerbil
          • 9 years ago

          But that’s not the issue. Wear-levelling by the controller — which happens below the level visible by software — ensures that those writes get distributed across all the flash. Each block has a write count attached to it, and as that value increases the contents of the block get swapped with a block that has a lower count. So, simplistically speaking, if Windows attempts to write 10,000 times to the exact same 4K block in the page file on your boot SSD, the actual wear works out to 1 write to each of 10,000 different blocks on the drive. In reality, of course, it won’t be quite that even (but then again there are other subtleties too, like the reservoir blocks that aren’t visible to software that get pulled into use when the most-written to blocks start to fail, further extending lifetime). But the very existence of those files that are barely-written-to helps extend the lifespan of the drive in spite of the files that are frequently written-to. (Likewise, free space on the drive — and TRIM, which enables the controller to know which written-to blocks it can now treat as free space — also helps. In fact, it helps even more because the drive can just write to those blocks rather than juggle the swapping of data in existing blocks when a rewrite occurs.)

      • FuturePastNow
      • 9 years ago

      That’s a fact of life for all SSDs, though. It’s not Sandforce’s fault. As the manufacturing process of transistors gets smaller, they get less durable, so future NAND won’t last as long as its predecessors.

      • UberGerbil
      • 9 years ago

      I wouldn’t say they “gloss over” it — they address it (see the para right after the diagram) and claim they have features that mitigate it. It’s reasonable to be skeptical until independent reviews test those claims, but I don’t think it’s reasonable to assert they didn’t address it at all.

      We’ve seen drops in write endurance with each new process node, and also in the shift from SLC to MLC. The mfrs have found ways to compensate each time; moreover, as SSDs get larger they’re more likely to be less than 100% full, which helps a lot for endurance (keeping an extra pool of blocks with 0 cumulative writes increases the average block lifespan significantly). As long as they don’t do questionable things like reserving extra blocks so that your usable capacity is less than advertised, or making unrealistic assumptions about usage patterns, there shouldn’t be a problem. But, as always, wait for the reviews.

    • 5150
    • 9 years ago

    Sandforce no doubt has speed, however, I don’t like how they’ve hidden information from consumers in their firmware update and OCZ has really screwed up with their 25nm chip fiasco currently going on. I don’t know if I trust either company anymore.

      • PrincipalSkinner
      • 9 years ago

      You should trust them as much as you should trust Intel, AMD and nVidia. In other words, don’t trust them at all.

    • PrincipalSkinner
    • 9 years ago

    I would very much like to see round up of all Sand Force based drives, once they all come out. Over here I can buy AData and OCZ but OCZ comes with 2 year warranty.

Pin It on Pinterest

Share This