Industry heavyweights collaborate on PCIe SSDs

A group including Intel, Dell, IBM, and EMC is collaborating on a standard for PCI Express-based solid-state disks. Ars Technica has the skinny on the group’s plans, which aim to define a standard connector and form factor for PCIe SSDs. It looks like the consortium is leaning toward sticking with the 2.5″ form factor commonly used by modern SSDs, which strikes me as a little odd considering that PCIe-based models generally reach their jaw-dropping performance levels by combining multiple SSD controllers and flash chips in what are effectively multi-drive RAID arrays. Good luck cramming all that into a 2.5″ drive.

According to Ars, the group is actually hoping to migrate all solid-state disks from Serial ATA to PCI Express. That might seem unnecessary with 6Gbps SATA delivering more than enough bandwidth for today’s fastest flash arrays. However, SATA controllers are usually found on I/O hubs one hop away from the CPU, where they must share interconnect bandwidth with other peripherals. The Serial ATA protocol wasn’t designed with solid-state storage in mind, either; a custom-tailored standard riding atop a PCI Express interface could offer better performance even if the PCIe link doesn’t have any more bandwidth than its SATA counterpart.

The prospect of an ultra-fast SSD plugging directly into PCIe lanes on the CPU is certainly tantalizing. As Ars astutely points out, we could see even greater benefits if operating systems more intelligently used the solid-state storage they have available. SSDs seem best suited to their own tier in a system’s memory hierarchy—as fast scratch disks and repositories for one’s most commonly accessed files.

Comments closed
    • JustAnEngineer
    • 9 years ago

    A cynic might assume some sort of relationship between Intel’s lack of chipset support for SATA3 and their sudden backing of a different way of connecting high performance drives to your PC.

    • GreatGooglyMoogly
    • 9 years ago

    So, can you boot from PCI-E disks nowadays?

    • Bensam123
    • 9 years ago

    …why? Just buy a PCI-E based SAS/SATA controller and them SSD’s.

    I am curious as to why they haven’t taken this concept a bit further then they already have. What they’re doing now is easy. What they should do is give each chip it’s own controller and then form a raid 0 array out of them. That isn’t something consumers can do as it’s on such a small level. You’d need a ridiculous amount of SSDs to accomplish the same thing. Similar to how dual channel memory works, only it would be something like sixteen channel memory (depending on the number of memory chips)

    • Voldenuit
    • 9 years ago

    We need more mobos with PCIE 4x ports, and ones that aren’t obscured by double wide GPU coolers.

    • Decelerate
    • 9 years ago

    Damn, I’d want my hard drives to connect to smaller ports than a pcie connection…

    • OneArmedScissor
    • 9 years ago

    This would make kind of a weird trade off for future low power PCs. If SATA becomes just another tacked on legacy thing, they could eventually remove any purpose for southbridges, without having to bother integrating anything beyond USB. That would allow for a simple equivalent of a giant smartphone SoC.

    …except then we’re stuck with PCI express for good, and that seems to be a power hog that’s standing in the way of chopping something like Bobcat down to the x86 equivalent of the up and coming multi-core ARM chips.

    • flip-mode
    • 9 years ago

    When will motherboards come with a full-fledged on-board SSD?

      • OneArmedScissor
      • 9 years ago

      Probably when large SSDs are cheap and useful enough to the average person that they won’t feel like they’re going to lose sales of SSDs for giving you a few GB of flash memory.

      So…never. 🙁

      • tfp
      • 9 years ago

      If they do it right you’ll plug in cards like RAM. That way you can upgrade just like with ram.

      • eitje
      • 9 years ago

      when the cost-per-gb hits the right spot to make that a non-upgradable component.

      as it turns out, traditional HDDs still haven’t hit that point, and probably never will.

    • Rurouni
    • 9 years ago

    Maybe… just maybe…. you don’t really need to directly plug the SSD into the PCIe slot. You know, maybe they can use something like a cable… You know, notebook doesn’t have a lot of space…
    The article also said a standardized drive connector, hot swap, etc.. Maybe they just want to make a drive connector that are standardized so that you can use the same cable for different brand? Maybe it will works like current SATA based SSD, but the cable plug directly to the PCIe?
    Argh, I just can’t think why they settled on 2.5″ form factor.. do they must settle for the SSD to be connected directly to the PCIe slot? hm..

    • CheetoPet
    • 9 years ago

    Is SATA 3 really enough? The SF-2000 is supposed to come close to max’ing that out, unfortunate considering there’s no SATA 4 even on the horizon yet & flash memory is only gonna get faster. A direct PCIe link to storage seems like a logical way forward.

      • OneArmedScissor
      • 9 years ago

      It’s enough for PCs. Servers have always used all sorts of whacky things, depending on the application. I don’t think it’s going to hurt anyone even a few years down the road.

    • ew
    • 9 years ago

    This just makes sense. There’s nothing SATA does that PCIe doesn’t except have a cable standard. SATA is just a latency/bandwidth sucking middle man at this point.

    PRO TIP: If you’re designing an external PCIe disk cable standard then frick’in make sure the cable can do power as well!!!

    • passive
    • 9 years ago

    I’ve got a lot of quibbles with your summary of this development.

    First off, form factor. The fastest single-controller SSDs available today got along just fine with 2.5″. So will the next generation models. But more than that, because SSDs don’t have the same shape requirements as magnetic storage, it’s pretty easy to cut down on capacity and increase parallelism. There are limits, of course, but practically speaking the 2.5″ form factor won’t hinder any but the most extreme of designs.

    Second is your comment that it might seem unnecessary because of the bandwidth of 6Gbps SATA. I don’t believe this is accurate either. The maximum bandwidth provided by this spec is 750MB/s, and real world controllers will always be a bit off of that. The Revodrive OCZ announced yesterday is at 740MB/s. Next generation Sandforce is supposed to hit 500MB/s per controller. PCIe 3.0 provides 1GB/s per channel, and I don’t see any reason multiple channels couldn’t be used.

    That is all.

    • cegras
    • 9 years ago

    And OCZ isn’t in on this? Ouch.

      • 5150
      • 9 years ago

      That’s because they beat all these others to the punch.

        • indeego
        • 9 years ago

        Yeah but it’s ghetto OCZ.

        /[

          • 5150
          • 9 years ago

          Totally fair. They easily have the fastest products and a quick response forum, but they do smack of unorganized.

    • IntelMole
    • 9 years ago

    Does this enable the possibility of discrete controller logic as a separate module on future multi-core CPUs?

    Putting the disks closer to the CPUs like that might be quite exciting.

      • Meadows
      • 9 years ago

      Indeed, destroy your OS with overclocking today!

    • joneljay
    • 9 years ago

    The performance gain of SSD on PCIe would be negligible over SATA. However this opens up everything inside the SSD, including flash chips and controller, to direct control under CPU. It is possible to have pure host based SSD where the CPU doing the bulk of computation, just like fake raid cards. Even more appealing, it would be possible to “repair” defective SSD since CPU can access the flash chips directly to recover data.

    • Krogoth
    • 9 years ago

    SSDs are already shifting the traditional memory hierarchy in modern OS. They just need to reach critical mass for OS-guys to justify recoding the hierarchy to optimize for their usage.

    In the foreseeable future (5-10 years) I easily see this setup in a desktop/workstation. HDD being used for bulk, permanent archival data storage. SSDs being use for faster, somewhat temporary data storage (scratch disk, swap, OS etc). Good, old main memory being used for its traditional role as ultra-fast, temporary data storage. Servers themselves use SSDs as local storage with on-site and/or off-site permanent archival data via SAN/NAS utilizing HDDs/tape media.

    The days of fast-cold-booting are coming back again.

      • ShadowEyez
      • 9 years ago

      This looks interesting.

      In the next few years, I see the focus of SSD’s will be coming down in price, and moving up in performance. Then 3-5 years out they will likley start focusing on capacity, at which point SSD’s can replace HDD’s for most uses.

Pin It on Pinterest

Share This