Samsung’s first NF1 SSD puts 8 TB on a wide gumstick

Samsung just announced an apparently-unnamed 8-TB NF1 SSD. In case you're confused, NF1 isn't the name of the SSD, but instead the name of the form factor. NF1 is a derivative of M.2 designed for enterprise environments and to better suit the shape of Samsung's NAND packages. The idea is that the wider drives can fit flash chips in two-wide rows.

The drive itself is built using sixteen 512-GB NAND packages connected to an all-new controller that supports NVMe 1.3 and PCIe 4.0. Despite the freshness of the hardware, Samsung's quoted sequential performance numbers—3100 MB/s for reads and 2000 MB/s for writes—aren't exactly eyebrow-raising. The rated 500K IOPS on 4K random read operations is more inspiring, although the company says that these drives are only good for 50K IOPS on random writes.

If you feel like you've seen NF1 drives before, that's possible. The M.2-derived form factor was called Next-Generation Small Form Factor, or NGSFF. That's a call-back to the original name of M.2, NGFF. Ironically, NGSFF or NF1 drives are actually larger than M.2 drives. NF1 drives make use of a couple of the “no contact” pins in the M.2 spec to enable hot-swapping. That's a critical feature in enterprise environments, as is maximized storage density. Samsung says that by using 72 of the new 8-TB SSDs, you can stuff 576 TB of solid-state storage into a 2U rackmount chassis.

According to Samsung, these new SSDs are already available now, although you won't find them on the virtual shelves at Newegg. If 12 petabytes of solid-state storage in a 42U rack doesn't do it for you, don't fret. Samsung says a new version of this SSD based on 1-TB NAND packages should be available in the second half of this year.

Comments closed
    • Waco
    • 1 year ago

    You don’t need “eyebrow raising” performance when you’re cramming 72 of them in a box. Very few servers can sustain 144 GB/s of throughput in any meaningful way for writes anyway…and it’s nigh impossible to do that while you’re moving data on/off the node as well.

    • freebird
    • 1 year ago

    “Samsung’s first NF1 SSD puts 8 TB on a wide gumstick”

    Is that Wintergreen or Fruity flavor?
    and
    That is one EXPENSIVE Stick of Gum!!!

      • Goty
      • 1 year ago

      There’s a joke about write endurance and Fruit Stripe gum in there somewhere, but I’m not smart enough to find it.

    • albundy
    • 1 year ago

    the m.2 form factor isnt a good one, or at least not implemented efficiently. ram slots are efficient. and better cooled.

      • Bauxite
      • 1 year ago

      Yup, laptop design ended up being gobbled up for enterprise and desktops because they needed *something* small besides x4 cards and they needed it yesteryear.

      U.2 and custom”ish” like this has displaced it mostly in enterprise, desktops still stuck because staying on the laptop commodity chain is far cheaper.

      • Goty
      • 1 year ago

      Cooling is actually a somewhat tricky thing with SSDs, as you don’t actually want to cool down the flash too much when programming it since doing so will actually shorten its lifespan (https://www.eeweb.com/profile/eli-tiomkin/articles/industrial-temperature-and-nand-flash-in-ssd-products). Obviously, this is at odds with the need to cool the controller to prevent thermal throttling, which is the reason for solutions like the inclusion of the copper-backed label used on Samsung’s 960 and later NVMe SSDs; it acts as a heat spreader, distributing the heat from the controller to the NAND chips.

      So, while the M.2 form factor may make it a bit harder to cool, this isn’t necessarily a bad thing.

        • not_a_gerbil
        • 1 year ago

        I don’t think you can really qualify the lifespan from that data because it shows two opposite effects:

        “The data shows that NAND programmed to its endurance limit at -40oC will have a higher RBER than NAND programmed at 25oC, and higher than that of NAND programmed at 85oC”

        And in opposition to writing the data at a higher temp is the data stored at a higher temp decreases reliability.

        “The acceleration factor for a NAND relative to a temperature of 55oC is shown in the table below. For example, for NAND devices specified with 1 year of data retention, storing at 85oC will accelerate the charge de-trapping mechanism by 26 times when compared to storing at 55oC.”

        So while writing at high temp could increase reliability, reading that data back a year later is less reliable if your SSD is running at high temperature.

          • psuedonymous
          • 1 year ago

          But an SSD in active operation will be checking and refreshing cells. That controller is not just there to look pretty!

          The whole “NVME SSDs need cooling!” meme has a similar origin to the ‘ramsink’ meme. FBDIMMs needed active cooling, but on consumer DDR/DDR2/DDR3/DDR4 (ECC or otherwise, or even RDIMMs) they are utterly pointless. The only NVME drive that had any real ‘issue’ with thermal throttling was the PM951, and that’s because it was an OEM drive delivered with an unmodified baseline firmware intended to be installed in a chassis with direct heatsinking and have the firmware customised and reflashed to throttle at a temperature/time to patch the thermal solution in place, rather than a drive intended for end users to install out of the box. When installed in a consumer machine with the unmodified firmware it would just go ALL POWER ALL THE TIME until it hit the thermal trip.

          • Goty
          • 1 year ago

          There’s really no conflict there, actually. The controller will heat up when it is under load, thereby heating the NAND above its idle temperature and reducing the wear incurred by the flash cells in the process, and ambient temps aren’t likely to be high enough in consumer applications to severely degrade the ability of the flash to retain data.

            • not_a_gerbil
            • 1 year ago

            Read the article you linked. The conclusion says the opposite of your claim.

            “NAND is subject to two competing factors relative to temperature.”

            • Goty
            • 1 year ago

            Read the post I posted. I’m not saying there aren’t two competing factors, I’m saying the data retention factor isn’t typically an issue in consumer environments where ambient temperatures tend to be comparably lower and the NAND will only be heated during read/write operations when the controller also heats up.

            • not_a_gerbil
            • 1 year ago

            “I’m saying the data retention factor isn’t typically an issue in consumer environments where ambient temperatures tend to be comparably lower.”

            This is wrong because consumer environments don’t have lower temps than server rooms. Professional server rooms have separate environmental controls to keep the server room cool and that temp is lower than comsumer’s homes.

            • Goty
            • 1 year ago

            Even conceding that, the heating of the NAND during programming provides a significant benefit to its durability. Which, y’know, was the whole point of my post in the first place.

            Thanks for playing.

    • tsk
    • 1 year ago

    Wow PCIe 4.0, didnt think I’d live to see the day.

      • psuedonymous
      • 1 year ago

      Turns out it was just a press-release mistake, it just has x4 PCIe 3.0 lanes.

    • JosiahBradley
    • 1 year ago

    You mean 1TB NAND packages later this year right? Also 24PB in a rack is insane at these speeds.

      • morphine
      • 1 year ago

      Yep. That was just one letter off, so… fixed 🙂

Pin It on Pinterest

Share This