Computex 2017: Intel unleashes NVMe RAID for X299

The unveiling of Intel's X299 chipset has made a big splash here at Computex 2017. The major motherboard manufacturers are all aflutter showing off all the RGB LEDs—ahem, compelling features that set their X299 boards apart. Buried in all the noise are hints of new RAID functionality that Intel developed to accompany X299. The company doesn't seem to have directly disseminated any information about what it is or how it works, but we've managed to cobble together a rough idea by talking to Intel's motherboard partners.

For starters, the new RAID functionality is called Virtual Raid On CPU, or VROC. Users dropping high-speed NVMe storage into M.2 and spare PCIe slots can select arbitrary subsets of those drives to create a blazing-fast RAID 0 array. Here's where the waters get muddied a bit. Some Asus guy with pretty solid storage credentials who goes by the Gasior surname asserts that those arrays are bootable. However, Gigabyte told us that VROC depends on Intel's RSTe driver, which is software you run on top of Windows. These two claims would seem to be at odds, at least at first sight.

VROC header on EVGA's X299 FTW K

At the end of the day, it would appear that Intel's out to make a quick buck (my words, not the manufacturers'). X299 boards may optionally include a VROC header on the PCB. Support for RAID 0 NVMe arrays is free, but you have to shell out $99 for a physical VROC key to plug into the header to unlock RAID 1 and RAID 10. For RAID 5, there's a more expensive key (we heard both $199 and $299 are possible). These keys aren't being channeled through motherboard manufacturers, so as far as we know Intel will be selling them directly.

Since the implementation details of VROC are nowhere to be found, we don't know whether the key contains any real mojo or not. All signs point to it merely being a sort of physical DRM. We'll keep you posted as we learn more.

Comments closed
    • psuedonymous
    • 2 years ago

    The big question with X299 (as almost every board connecting the m.2 slots to the CPU rather than PCH) is whether Optane cache drives will work. Normally the transparent caching is managed by the PCH in 200-series chipsets.

    • davidbowser
    • 2 years ago

    See now THIS is the stuff that I like to see on high end mobos: Next gen storage with advanced features! I was really liking the ROG Zenith with the 10GB Ethernet and 802.11ad wireless, so now I am going to be thinking of real feature tradeoffs instead of just AMD/Intel bang for buck.

    Now the ball is back with AMD (or 3rd parties) to see what they can do to beat the RAID-y goodness!

    • albundy
    • 2 years ago

    because optane was DOA?

      • Waco
      • 2 years ago

      Yeah, no. Optane is far better than any Flash device for low queue depth reads or any kind of write.

    • wof
    • 2 years ago

    Shouldn’t the title really be “Computex 2017: Intel puts NVMe RAID for X299 on a leash”?

    • ermo
    • 2 years ago

    Dick move, intel. >:-[

      • curtisb
      • 2 years ago

      You spelled “disk love” wrong. And your emoji is off…

    • not@home
    • 2 years ago

    If this is acually a hardware thing, I wonder how long it will take for someone to come up with a work around hack.

    • jihadjoe
    • 2 years ago

    Won’t it work if the disks are JBOD to the BIOS and soft RAID-1 in the OS?

      • willmore
      • 2 years ago

      Unless Intel chose to lock out certain functionality without the presence of a hardware key. At least as far as their driver goes. If Windows has a native driver to RAID a bunch of drives, I don’t know how Intel would justify disabling that.

      Also, I don’t see how they’re going to stop Linux users from RAIDing these devices. There’s no Intel drivers involved in that case.

        • curtisb
        • 2 years ago

        [quote<]Also, I don't see how they're going to stop Linux users from RAIDing these devices. There's no Intel drivers involved in that case.[/quote<] Hardware RAIDing drives is different from software RAIDing them in the OS. Hardware RAIDing happens underneath the OS. The "Virtual RAID [i<]on CPU[/i<]" implies that it is a "software" RAID within the chipset, and the RAID calculations are being handled by the system CPU rather than having a dedicated RAID controller ASIC. That's still done at a hardware level rather than an OS level, though.

          • willmore
          • 2 years ago

          How does “Virtual RAID on CPU” imply that the chipset is doing it?

          Also, there are Linux drivers for such situations.

            • curtisb
            • 2 years ago

            How do you think RST RAID has been done all these years? All of those cheap Highpoint and Promise controllers were also a software RAID solution that used firmware in a chip on those cards, but used the system CPU to do the RAID calculations. That’s why it’s called “software RAID.” Intel is just using a different term now and calling it “virtual RAID.”

            You’ll pay big money for hardware-based RAID controllers with dedicated RAID ASICs. Anywhere from hundreds to multiple thousands, depending on the number of channels, drives, and RAID levels a given controller supports.

            • willmore
            • 2 years ago

            Those cards and the Intel raid are collectively called called ‘fake raid’ for a reason. They’re all done on the processor.

            Intel isn’t doinging anything here but charging people for being gullible.

            • Waco
            • 2 years ago

            Assuming they aren’t doing something fancy in the CPU since these are NVMe drives…

            I could easily see this be a somewhat-accelerated stack on-chip to keep latencies down.

            • curtisb
            • 2 years ago

            There’s also that, too. I imagine it is somewhat accelerated, especially since it requires a RAID enabler key to go beyond RAID0. You usually only see that on controllers that have some sort of ASIC or accelerator chip. Some controllers now even require a license for certain features to be enabled (i.e. SanDisk DAS Cache).

            • curtisb
            • 2 years ago

            There’s nothing fake about them. It’s still RAID. They just don’t have a dedicated ASIC so they use the system CPU. The bad stigma about it being done on the processor is from over a decade ago when CPU’s weren’t nearly as fast as they are now, so those controllers could affect overall system performance. These days you won’t even notice it in a benchmark, let alone real world usage.

            The original question also mentioned doing software RAID1 in the OS. That also uses the system CPU and puts the OS between the operations.

            • willmore
            • 2 years ago

            That is actually the name of it: [url<]http://skrypuch.com/raid/[/url<] You could just google for "linux fake raid". This web site was just the funniest one I could quickly find. This 'chipset RAID' or 'BIOS RAID' came about in the 486 era when there wasn't enough bandwidth on the bus for the CPU to do the parity calculations. They were despised for a reason. You had to pay extra to get a MB or add-in card that supported it but it did nothing that the OS couldn't have already done for itself--and probably done a better job of. Now that we have tons of BW and the CPUs are great at doing the parity calculation, it makes a lot of sense to implement RAID this way. After all, it lets us remove the RAID controller which was a source of failures and made the system less realiable. But, we still have Intel trying to act like this is some kind of special 'value add' feature. It wasn't a few decades ago and it isn't now. Just let the OS do the job and stop trying to tie it to CPUs and chipsets--and multi-hundred dollar add-in dongles.

            • curtisb
            • 2 years ago

            I personally don’t want the OS doing it. Ever had a RAID array fail? They are recoverable, (unless it’s RAID0 or JBOD), but if the OS is on the failed array it makes it difficult to recover…because then you have to get your OS working first. Adds more time to the recovery. No thanks.

            We also don’t know for sure that this is a software RAID on these X299 boards. The more I think about it, the more I don’t believe it is a full on software RAID. As I’ve mentioned in other comments, the RAID enabler key isn’t anything new. RAID controllers have had varying levels of RAID support for years, and the more RAID types and drives they support, the more expensive the controller.

          • Monstieur
          • 2 years ago

          This looks like rebadged RSTe. RSTe has this weird thing where the physical drives are still visible in Device Manager, but they do not register as logical disks in Disk Management. I assume the RSTe driver is some kind of filter driver that abstracts the logical disk layer alone and performs the RAID calculations in the kernel, even though the physical disks are visible to the OS.

          Even fake-RAID controllers come in two types. The first type has a driver + basic controller (like RSTe) where the OS can see the physical disks, but the driver presents the logical disks. The second type has a driver + advanced controller (like entry level HighPoint controllers) where the controller hides the physical disks and only presents logical disks to the OS. In both cases the driver performs the calculations in the kernel.

    • curtisb
    • 2 years ago

    [quote<]VROC depends on Intel's RSTe driver, which is software you run on top of Windows. These two claims would seem to be at odds, at least at first sight.[/quote<] They're not really at odds. All RAID Controllers require a driver, it's just a matter of whether or not that driver is included with the OS to be bootable during the install. Otherwise, you just supply the driver during the install process so the controller and drive become visible to the OS. That's not really anything new, it just seems so because Windows 8 and onwards has included basic support for booting from RST arrays for some time so we've gotten used to it. This is probably a new implementation that doesn't work with what's included with the current version of Windows 10 (or Server 2016).

      • Waco
      • 2 years ago

      Yeah. I remember slipstreaming RAID drivers into installs, “F4 for floppy driver” crap.

      It’s nice not doing that, but it means that people are even more unlikely to know what to do when the boot driver isn’t available. 😛

        • curtisb
        • 2 years ago

        Yeah, but at least these days you can put the drivers in a folder on your install flash drive. 🙂

        For personal installs I tend to still load the latest RST driver so it’s there during the install. I know I’m going to load it afterwards anyway. At work we just use SCCM where the drivers are loaded in a package.

        • Forge
        • 2 years ago

        F6, you mean. We can see who has done it more/more recently. I remember F5 to select HAL and F7 to force the non-ACPI PIC HAL, too.

          • curtisb
          • 2 years ago

          Intel’s zip file for the RST drivers is actually still called f6flpy-x64.zip (or -x86.zip for 32-bit).

          • Waco
          • 2 years ago

          Ha, yeah. It’s been a while since I’ve built a Windows 98/2000/XP machine that booted from RAID or anything like it.

    • DPete27
    • 2 years ago

    I’ll keep it on the key ring with my Hellcat’s red key.

    • chuckula
    • 2 years ago

    [quote<]Here's where the waters get muddied a bit. Some Asus guy with pretty solid storage credentials who goes by the Gasior surname asserts that those arrays are bootable. However, Gigabyte told us that VROC depends on Intel's RSTe driver, which is software you run on top of Windows. These two claims would seem to be at odds, at least at first sight.[/quote<] That guy talks like he's messed up in the head. I'm afraid he's suffering from cognitive dissonance.

      • Redocbew
      • 2 years ago

      No doubt. I wouldn’t be surprised if he set these systems up and let them run just to watch them die.

      • cynan
      • 2 years ago

      How quickly we forget about raid drivers that were required for Windows 7.

      • Thrashdog
      • 2 years ago

      So let me get this straight: this Gasior guy claims that [i<]purportedly,[/i<] you can boot from VROC arrays? Sounds sketchy to me.

        • Waco
        • 2 years ago

        Booting from a RAID array isn’t magic, assuming there’s a driver available.

          • morphine
          • 2 years ago

          Whoosh.

            • Waco
            • 2 years ago

            I clearly missed something. Anyone willing to pity the fool and explain the joke? 🙂

            EDIT: Or is this just poking fun at the usual TR [i<]purportedly[/i<] tomfoolery?

            • morphine
            • 2 years ago

            Geoff Gasior (who that paragraph in the article refers to) was a TR editor for many a year before joining Asus 🙂

            • Waco
            • 2 years ago

            Ha, understood. I’m bad with names. 🙂 Thanks for the explanation!

    • drfish
    • 2 years ago

    A dongle? Really?

      • curtisb
      • 2 years ago

      That’s really not anything new in the high-end workstation and server world. RAID enabler keys have been around for ages.

        • chuckula
        • 2 years ago

        It’s almost like this “new” feature might be a repackaged RAID solution that was already out on the market.

      • Forge
      • 2 years ago

      Yeah, tell that to the little phone-jack looking mofo on my SuperO mobos that enables iSCSI initiation.

Pin It on Pinterest

Share This