Samsung starts mass-producing PCIe 3.0 SSD

Remember the mini NVM Express drive Samsung teased during its SSD Global Summit last summer? Well, the SM951 is now in mass production—and it's very fast indeed. The four-lane PCIe 3.0 drive is rated to hit sequential speeds of 2150MB/s with reads and 1550MB/s with writes. When the drive is paired with a Gen2 interface, those top speeds fall to 1600MB/s and 1350MB/s, respectively, which is still pretty quick for an M.2 2280 gumstick.

Source: Samsung

Gen3 connectivity conveys energy efficiency benefits, Samsung says, and the SM951 also supports an ultra-low-power mode called L1.2. In this state, "all high-speed circuits" are turned off, and power consumption drops to just two milliwatts.

The SM951 is available in 128GB, 256GB, and 512GB capacities, but only to big-name system builders. Expect to see the drive tucked into high-end notebooks and workstations before too long. If its predecessor is any indication, the SM915 could eventually become available at online retailers like Newegg, as well. Fingers crossed.

Comments closed
    • anotherengineer
    • 5 years ago

    Samsung starts mass-producing PCIe 3.0 SSD

    “but only to big-name system builders.”

    That sounds like an oxymoron to me…………………..sigh

      • lycium
      • 5 years ago

      Yeah it’s all the rage these days to keep the best products out of consumers’ hands. Intel flashing around their fancy 128MB Iris Pro: “Nya nya you can’t buy it!”

    • Anovoca
    • 5 years ago

    I finally bit the bullet and drop bills on a 500gb samsung ssd. That is exactly why Samsung has decided to mass produce something faster. They might not know that was the reason, but that was the reason.

    • willmore
    • 5 years ago

    I’m still having trouble correcting my preconception that these little stick devices are not meant for netbooks and such and therefore have crappy performance.

    Smaller==faster only used to apply to transistors on chips, not storage devices.

      • UnfriendlyFire
      • 5 years ago

      Think of it as going from 5¼-inch floppy disks to 3½-inch floppy disks.

        • willmore
        • 5 years ago

        You’re not helping and you’re just making me feel old. 🙂

      • Kougar
      • 5 years ago

      Yeah, took me awhile to break that thinking as well. The first mSATA and M.2 SSDs were just worse performing copies of the 2.5″ originals. Yet suddenly at some point SSD manufacturers have fixed the dozen major issues and can now cram everything together without compromising performance. Go figure!

      Once again we’re bottlenecked by the system, I’m not sure if any motherboard manufacturers utilize PCIe 3.0 M.2 slots let alone x4 ones. Most Z97 boards are still 2nd Gen PCIe x2.

      This drive is already only 1,000MB/s away from maxing out its PCIe 3.0 x4 bandwidth, I wonder if a larger PCIe model could hit it with that controller?

        • Dezeer
        • 5 years ago

        Yeah the first mSata SSDs were slower and more expensive than other 2,5″ SSDs, but something like 840 Evo mSata is really close in price and performance to the Evo 2,5 model. I think the first, and current, M.2 SSDs had problems with cooling the controller.

        The drive is closer to 1,500MB/s away from maxing the interface.

        Is there any change we are going to see M.2 22110 drives?

          • Kougar
          • 5 years ago

          According to [url<]http://www.pugetsystems.com/labs/articles/Overview-of-M-2-SSDs-586/[/url<] actual performance for the 3rd Gen x4 tops out around 3.2GB/s alas. Good thing PCIe 4 was already announced... I did a considerable amount of finger-work. The ONLY Z97 maker to offer 3rd Gen x4 M.2 is ASRock and only on a couple boards (Extreme6 and Extreme9). Nobody else that I could find. Only ASRock, ASUS, and MSI even offer full speed 3rd Gen x4 M.2 slots on X99 boards. More than I realized, but not many...

            • Dezeer
            • 5 years ago

            I think the writer is mixing up the Sata, PCI-e 1.0 and 2.0 8b/10b encoding and PCI-e 3.0 and 4.0 128b/130b encoding.

            Only Z97 boards with 4 wide M.2 are Asrock Extreme6, Asrock Extreme9 and Asus ROG Maximus VII Impact, while X99 has about 20 boards.

            • UnfriendlyFire
            • 5 years ago

            I would assume majority of M.2 ports would be PCI-E 3.0 by the time Broadwell (desktop) or Skylake shows up.

            • willmore
            • 5 years ago

            We’re going to need to see a new socket as the current non-workstation sockets don’t have more than 16 PCI-E high speed lines.

            Or we need to go back to the idea of a switch on the motherboard so that there can be more flexability in the I/O deployment on the board.

            We used to do things that way back in the FSB era. But we went away from that when the memory controllers were integrated into the processor.

            It should make high speed implementations easier as the PCI-E traces wouldn’t need to run as far–meaning that the impediance of the lines would be easier control and compensate for as well as receiving less interference. But, it would mean another complex reasonably high power device on the MB.

            • Kougar
            • 5 years ago

            I completely agree, switches are where it’s at. Most recent exotic motherboard solutions have been using them.

            Take the problem of the ASUS Z97-A motherboard. To stick the SM951 into a riser card one still needs a PCIe 3.0 x4 slot. This means the ONLY slot with enough lanes available is the secondary GPU slot, but to install a PCIe based SSD riser means the GPU slot must be dropped to run x8 as well. Which is crazy considering Z97 was supposed to help mitigate this exact problem with peripherals.

            Since the only PCIe 3 slots available are via the CPU they either need a switch, or Intel needs to update its chipset (and still probably need a switch to allocate the lanes properly), or Intel needs to add more PCIe lanes to its mainstream chips. I thought the number of lanes off Haswell-E was absurd but now I know better.

            • Kougar
            • 5 years ago

            When it comes to Intel, I would make no such assumptions anymore. But yes, Intel clearly needs to do something.

            Also consider that currently PCIe 3.0 ports are only available directly from the CPU. To offer PCIe 3.0 ports on its chipset it would also have to boost the interconnect between the CPU and chipset to handle the extra traffic.

            My understanding is that just a single SM951 by itself would be capable of nearly maxing out the DMI 2.0 interface on Z97 chipsets (it runs at 20Gb/s)

      • WaltC
      • 5 years ago

      My preconception has always been that anything planned for “netbooks and such” will be cheap, cheap, cheap and just generally not very good at all…;)

        • willmore
        • 5 years ago

        Yep, and somehow these SSD on a stick devices snuck out of the netbook ghetto.

    • UnfriendlyFire
    • 5 years ago

    “sequential speeds of 2150MB/s with reads and 1550MB/s with write”

    How many 7,200 RPM HDDs would you need to match that?

      • willmore
      • 5 years ago

      10 in a stripe and a good headwind.

        • Srsly_Bro
        • 5 years ago

        tailwind?

      • ptsant
      • 5 years ago

      With the price/GB of these SSDs, you could probably pay for as many HDDs. Not that you’d want that, of course. But I’d really like to get bigger, not necessarily faster, SSDs with lower prices. Even with a SATA SSD, performance for a home user is becoming a non-issue.

      • Klimax
      • 5 years ago

      About 10-15, bit less when using 10k+ Sequential is easy, random is impossible for HDDs.

        • UnfriendlyFire
        • 5 years ago

        EDIT: NVM

          • Beelzebubba9
          • 5 years ago

          Since IOPS scale with spindle speed, even a theoretical 30K drive would only do about 400 random 4K operations a second, which is a tiny fraction of what that SSD can do.

        • willmore
        • 5 years ago

        The question was about sequential operations.

        10 Seagate 2TB 7200 RPM drives would meet the sequential read and write specs. If you stripe them, you’d also have 20TB of storage, which might be nice. Though, the chance of data loss would be pretty high.

          • Klimax
          • 5 years ago

          I was accounting for different speeds over entire hard drive. (They are usually quite large) It can vary between 100-200MB/s. (IIRC I observed for WD Black) That is, I went for average observable speed and not maximum.

          And I know what question was, I just included afterwards note on random operations.

          ETA: Wonder for what dovnvote(s) were…

    • chuckula
    • 5 years ago

    NVMe helps too! Do you know if these drives are still AHCI or do they use NVMe?

      • smilingcrow
      • 5 years ago

      NVMe – [url<]http://www.anandtech.com/show/8236/samsung-ssd-global-summit-2014-845-dc-pro-with-vnand-sm951-with-nvme-support[/url<]

Pin It on Pinterest

Share This