Announcement

Collapse
No announcement yet.

Funtin SFF-8639: U.2 NVMe SSD To PCI-E Card Adapter

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Originally posted by Zan Lynx View Post
    That will not happen. NVMe drives will be getting faster in the future in step with PCIe. It's almost as if the manufacturers can plan ahead.
    You're certain of knowing nothing, john snow.

    NVMe main limit is controller power and NAND density/performance because the form factor is standardized and won't change, they can't just add more NAND chips to get moar speed. Both require R&D and time to improve, they aren't just adding shit with duct tape.

    If host hardware manufacturers push PCIe improvements harder, they can get the upper hand on them. It's an arms race like any other.

    For example it seems they (probably Intel given their vested interest in all this) have changed the pace on PCIe versions now, as PCIe 5.0 was announced this year and is supposed to become available in 2019.
    PCIe 4.0 was announced in 2011 and standardized only this year.
    So that's 3 years vs 6 of lead time.

    If they keep this up, in a couple PCIe versions we'll see SSD interfaces shrinking down to x2 or even x1.

    Comment


    • #12
      Originally posted by zamadatix View Post
      You missed his response by a few hundred miles, U.2 is the industry standard cabled non slot based solution torsionbar28 was looking for, no PCIe slots required. You only need to use PCIe slots/m.2 adapters if you have a board that doesn't come with native U.2 ports.
      How many consumer boards have U.2 host ports again? Socket 2011 boards? ASUS ROG and similar gold-plated "gaming" full-ATX boards?

      Really, it's not exactly "consumer" as he said.

      U.2 is mostly for servers and similar systems, it's not designed with consumer market limitations in mind so it will not penetrate consumer market much.

      If the complaint is about PCIe lanes you'll probably just want a better suited CPU but lanes are oversubscribable (in the same way having 24 SATA connectors on a consumer motherboard was done) but at that point why pay the performance premium of U.2?
      If the complaint is about the PCIe lanes the only solution is to have higher PCIe versions standardized faster, sharing them defies the point of U.2

      Comment


      • #13
        Originally posted by Zan Lynx View Post
        That will not happen. NVMe drives will be getting faster in the future in step with PCIe. It's almost as if the manufacturers can plan ahead.
        You're just blindly extrapolating from the past several years.

        What has actually started to happen, as NAND flash cells simultaneously shrink and encode ever more information (heard of QLC?), is that flash speeds are actually slowing. At the high-end, there are NVMe drives that can post up some impressive numbers, but that's not characteristic of mass market, mainstream NAND flash.

        Note that 3D XPoint is not NAND flash, so the trend doesn't apply to it. And Samsung's ZFlash is probably a low-density, high-end part that's also not destined for mainstream use.

        Comment


        • #14
          Originally posted by starshipeleven View Post
          How many consumer boards have U.2 host ports again? Socket 2011 boards? ASUS ROG and similar gold-plated "gaming" full-ATX boards?
          AMD Ryzen chipsets support SATA Express. I believe this allows a pair of SATA ports to connect as a x2 PCIe link to a U.2 drive.

          Comment


          • #15
            Originally posted by coder View Post
            You're just blindly extrapolating from the past several years.

            What has actually started to happen, as NAND flash cells simultaneously shrink and encode ever more information (heard of QLC?), is that flash speeds are actually slowing. At the high-end, there are NVMe drives that can post up some impressive numbers, but that's not characteristic of mass market, mainstream NAND flash.

            Note that 3D XPoint is not NAND flash, so the trend doesn't apply to it. And Samsung's ZFlash is probably a low-density, high-end part that's also not destined for mainstream use.
            There are several potential high speed persistent RAM technologies coming up in the next few years. And while I expect dedicated DIMM slots for these, many of them are going to build PCIe / M.2 / U.2 devices also.

            Comment


            • #16
              Originally posted by Zan Lynx View Post
              There are several potential high speed persistent RAM technologies coming up in the next few years. And while I expect dedicated DIMM slots for these, many of them are going to build PCIe / M.2 / U.2 devices also.
              There will probably always be a high-end market that can approach the max speed of the latest PCIe flavor x4 or more, but that's going to be priced at a substantial premium.

              On the desktop, Intel's 3D XPoint DIMMs will probably kill the market for such devices, leaving only the need for a larger number of lower-speed links to host one or more devices with more economical bulk storage. I think AMD made a good choice to support SATA-Expresss.

              Comment


              • #17
                Originally posted by coder View Post
                AMD Ryzen chipsets support SATA Express. I believe this allows a pair of SATA ports to connect as a x2 PCIe link to a U.2 drive.
                Yes it does, and is a more consumer-friendly type of connector (can be used to run 2 Sata devices or combined to run a U.2 one, so it isn't a huge waste of space for those that don't install a SSD).

                Its current issue is that it sends only pcie 3.0 x2 over the cable, so yeah, it's not good enough. And that's probably why it's not that common either, meanwhile even midrange and kinda low end boards have a M.2 slot.
                Last edited by starshipeleven; 18 November 2017, 07:33 PM.

                Comment


                • #18
                  Originally posted by starshipeleven View Post
                  Its current issue is that it sends only pcie 3.0 x2 over the cable, so yeah, it's not good enough.
                  Not good enough for the very fastest drives on the market, but more than fast enough for the median of NVMe devices out there.

                  IMO, it's a good tradeoff between performance, versatility, and PCIe lane allocation.

                  Comment


                  • #19
                    Originally posted by starshipeleven View Post

                    If they keep this up, in a couple PCIe versions we'll see SSD interfaces shrinking down to x2 or even x1.
                    I would like a PCIe 1x SSD today, but all stuff is either 4x or SATA. I think the market doesn't care about potential 2x or 1x users (yet), they're doing PCIe == higher end ; I think you'll see 4X 4.0 next because bar graph length is too good to pass on. (also for speed on 3.0 systems)

                    I agree with those who'd say bridge chips, muxed lanes etc. are enough, I think it's silly to pretend you need an Epyc because you might want to add more gigabytes.

                    Comment


                    • #20
                      Originally posted by coder View Post
                      AMD Ryzen chipsets support SATA Express. I believe this allows a pair of SATA ports to connect as a x2 PCIe link to a U.2 drive.
                      It might be "logical" SATA Express : motherboard manufacturer can decide on 4X M.2, or 2X M.2 and two SATA. So far I think they all go with 4x M.2 pretty much. And that's on the CPU not the chipset.
                      There can be chipset-less motherboards but none were launched (maybe with desktop Raven)

                      SATA Express connector is dead. edit : and it's not U2. it can probably be adapted.
                      No one made SATA Express drives.
                      Last edited by grok; 18 November 2017, 09:28 PM.

                      Comment

                      Working...
                      X