Comments Locked

16 Comments

Back to Article

  • Xex360 - Wednesday, June 21, 2023 - link

    Doesn't the Galaxy S23 Ultra uses UFS 4.0?
  • Gustavoar - Wednesday, June 21, 2023 - link

    Not by Micron
  • shabby - Wednesday, June 21, 2023 - link

    My s23 only gets 1000MB read and 230MB write.
  • dontlistentome - Wednesday, June 21, 2023 - link

    2007 called, it wants it's 3MB/sec back.
  • shabby - Wednesday, June 21, 2023 - link

    Wat?
  • dontlistentome - Wednesday, June 21, 2023 - link

    Describing tiny storage buried deep inside a small device that fits in your pocket as 'only' when it does a gigaBYTE a second.
  • shabby - Wednesday, June 21, 2023 - link

    I'm paying premium for these devices, I expect premium components. And 230MB/sec write speeds is laughable.
  • dontlistentome - Thursday, June 22, 2023 - link

    I'll bite. What data do you need to get onto a smartphone at that rate? And what mechanism do you have able to feed it that quickly? How do you know it's storage contrained rather than CPU or network?
  • shabby - Thursday, June 22, 2023 - link

    If the write speeds of storage are only 230MB/sec how slow are the random writes? Few MB/sec? This is now the slowest part of the phone, slowing down app installs and general usage of the phone.
  • iphonebestgamephone - Monday, June 26, 2023 - link

    Username cheks out
  • gagegfg - Monday, June 26, 2023 - link

    Galaxy Note 10 plus UFS 2.1 512 GB. 1500/1200 MBps read/write.
  • ceisserer - Wednesday, June 21, 2023 - link

    wouldn't these UFS chips a way to create super-fast, yet tiny USB drivers?
  • shabby - Wednesday, June 21, 2023 - link

    Not really, at least on Samsung phones the USB port is really slow, almost usb2 speeds.
  • iphonebestgamephone - Monday, June 26, 2023 - link

    If it were usb 2 speeds how is it able to 4k 60 output via hdmi??
  • shabby - Wednesday, June 21, 2023 - link

    Oh you mean actual usb drives, unlikely since they're not cheap, current crop of usb drives use the cheapest nand available.
  • twotwotwo - Wednesday, June 21, 2023 - link

    The hot/cold thing is cool--you can reduce GC/write amplification if you mostly have long-lived data in some Flash erase blocks and shorter-lived data in others, because you don't rewrite long-lived blocks a lot (or, from the other end, because the short-lived are mostly garbage by GC time).

    I wonder if this is relying on the OS for hints, and if there's potential cross-application in NVMe land. Slow writes you expect to be long-lived could skip an SLC cache and leave it for bursty traffic, for instance.

    You could probably get a bit of the benefit just from structuring background GC right so that the oldest blocks rewritten in a GC all end up together. Wouldn't surprise me if this is already done.

Log in

Don't have an account? Sign up now