6 ms·

Looks like a no-brainer if you're on the Ubuntu treadmill. Key differences versus the previous LTS version: * Distribution packages built with additional secu

• ryanisnan 3y ago
  Asking genuinely, what causes Ubuntu to be more of a treadmill from say Debian?
  • cs702 3y ago
    Sorry, I don't understand your question. By "treadmill" I mean sticking to a particular distro year after year, upgrading regularly -- it could be Ubuntu, Debian, Red Hat, CentOS, etc. Unless you're constantly switching distros, you're on a treadmill. I wasn't comparing Ubuntu to other distros.
    • FuriouslyAdrift 3y ago
      "Treadmill" is usually used pejoratively, e.g. hedonistic treadmill.
      • cs702 3y ago
        Thanks. Yes, you're right, sometimes "treadmill" is used pejoratively to describe forced upgrades that otherwise would be unnecessary. I was referring to the fact that we are all voluntarily installing security updates and/or improved versions of the software we use, year after year.
        • halJordan 2y ago
          Fwiw i did not understand your original comment as pejorative, and thought it was a well done turn of phrase
          • ahartmetz 2y ago
            FWIW, I did pick up the pejorative connotation at first, but then realized it probably wasn't intended that way due to the complete lack of other negativity in the post.
      • gloryjulio 2y ago
        It's hackernews. I would recognize that this is in a joking manner
    • ryanisnan 3y ago
      Oh I see. I interpreted your comment as if it were more of a treadmill than other distributions.
  • jsheard 3y ago
    If you use the non-LTS versions of Ubuntu then you are expected to upgrade to each new version in short order, the previous version (23.10) is only supported until July.
    • ryanisnan 3y ago
      Django has the same problem. You either have to upgrade all the time, or you have to jump LTS to LTS, which is also not advised from Django's perspective.
      • cqqxo4zV46cp 2y ago
        There are so many ways in which the Django upgrade process could be worse, and IMO very few ways in which it could be better. Not really sure how this is a “problem”.
  • actionfromafar 3y ago
    I wish Debian was a treadmill with fixed release dates like Ubuntu has.
    • mort96 3y ago
      I mean it's not the same month every time, but it seems to be on a pretty stable 2 year cycle?
    • NewJazz 3y ago
      What benefit would that have? Even if new versions are not released on a fixed schedule, you don't have to upgrade immediately.
      • actionfromafar 3y ago
        Migration scheduling and planning for total software lifetime. Everything is just smoother on a fixed cadence.
        • NewJazz 3y ago
          You can do a fixed cadence even if Debian doesn't though. Just every year, upgrade to the new stable if it is available.
          • actionfromafar 3y ago
            Yeah, but now you have your own problems instead of pain shared by millions of other users. A lot of orgs on the same cadence is a great and under-estimated thing. I'm just sayin'
            • NewJazz 2y ago
              I don't get what you mean, actually. A lot of orgs upgrading simultaneously means that none of them have prior experience to look to. A staggered rollout of new software versions is often viewed as a less risky deployment method.
• echelon 3y ago
  > Looks like a no-brainer if you're on the Ubuntu treadmill. It'll take forever for ML packages to upgrade. The Python/ML/PyTorch ecosystem is slow to move.
  • cs702 3y ago
    Good point! I'll edit my comment.
  • Symmetry 3y ago
    I'll be trying to get PyTorch working on Docker with 24.04 and an AMD card. Lets see how that works out.
  • gpjt 2y ago
    I've been running my ML stack on Arch for the last six months and haven't had any issues so far. So perhaps not a big issue on a more stable distro like Ubuntu.
• meunal 3y ago
  "Users who train AI models on Ubuntu machines should wait until their ML stack (typically, Nvidia + PyTorch) is known to run without issues on 24.04." NVIDIA + PyTorch user here. No issues at all (w/ 550.54.15).
  • paulmd 2y ago
    I'm here for the upstreamed Arc GPU support, lol. I could probably have made it work with 22.04 with enough poking but I don't generally like upgrading major versions anyway. I was actually checking earlier but it hadn't released yet, lol. Probably jumping the gun anyway - it doesn't look like Arc has a release of the dev tooling for 24.04 yet anyway, no `noble` repo yet either. I figured at least I could get passthrough working to a 22.04 container, couldn't get the intel_iommu to behave, looks like it was still getting sucked up by the parent OS despite my GRUB kernel options. Might need to try HWE on the VM host too, or maybe it's just a fact of life based on how the outputs are wired up on my serpent canyon NUC (headless might work better, perhaps?). I'm actually fine if the ML stuff takes a bit, I kinda just want to play with GPGPU and I definitely want to play with the AV1 encoder. I ended up bumping into the same issue that stalled me out before, it happens in both 22.04 HWE and 24.04, the step about installing the out-of-kernel modules is busted. Installing the i915-dkms can't be done at the same time as the platform-vsec-dkms and platform-cse-dkms, apparently? But I pushed on it a little harder, and it seems like it works if you run two separate apt commands. I've got a couple of these weird environments, I need to set up a way to virtualize and passthrough them so I don't have to maintain actual machines for it. Not sure exactly what that looks like, maybe proxmox and some qemu machines in a pseudo-ESXi configuration or something. When I get around to getting the new disks in my NAS, I can also do iscsi/iPXE I guess (and I have optane for write cache).