• LouNeko@lemmy.world
        link
        fedilink
        English
        arrow-up
        8
        ·
        edit-2
        1 month ago

        A lot of people buy Nshitia (me included) because of Cuda and DLSS which are years ahead of AMD and decades ahead of Intel. Even with worse driver support on Linux, Nvidia still outpreforms AMD cards of the same class on every front, except net power consumtion.

        • 3bygone3@lemmy.world
          link
          fedilink
          English
          arrow-up
          1
          ·
          1 month ago

          From what I’ve seen they are leading in power consumption as well. Yes, 5090 is more power hungry than any AMD consumer GPU right now but it also comes with adequately higher performance.

          • Septimaeus@infosec.pub
            link
            fedilink
            English
            arrow-up
            1
            ·
            1 month ago

            With respect to per-watt metrics, 4090 was such an efficiency leap that 5090 struggles to match it. AFAIK 4090 still holds the efficiency crown in discrete graphics. (Maybe not for other acceleration tasks. Also driver/FW updates sometimes shift benchmark standings over time.)

        • WereCat@lemmy.world
          link
          fedilink
          English
          arrow-up
          1
          ·
          1 month ago

          It’s fine if you buy it because you know what you need and know that you’re paying more because of X. I’m talking about people that buy it just because it’s NVIDIA with no thought for anything else.

    • rumschlumpel@feddit.org
      link
      fedilink
      English
      arrow-up
      3
      ·
      edit-2
      1 month ago

      Probably pretty much exclusively for machine learning applications, though I wonder if it’s even still a good choice to use consumer gaming GPUs at these prices. But IDK what a developer would buy instead for home use, I can’t quite imagine that something like a Raspberry Pi AI hat has a better performance/cost ratio.

      • Septimaeus@infosec.pub
        link
        fedilink
        English
        arrow-up
        2
        ·
        1 month ago

        For local inference these days it’s basically all AS since it’s the only consumer hardware that can replace $20-80k server builds, but if you mean ML/AI developers specifically, they still prefer discrete GPUs for local training (on personal budgets: either recent consumer gaming/professional cards or older AI cards).

          • Septimaeus@infosec.pub
            link
            fedilink
            English
            arrow-up
            2
            ·
            1 month ago

            My bad, Apple Silicon, the private local ML “meta” as of ~Nov’25 (and last I saw) was RDMA clusters of M3 Ultras that could scale up to 3.6TB VRAM (a lot)

    • stoy@lemmy.zip
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 month ago

      I have simply kept buying nVidia as I understand what the model numbers mean, I have no idea of what AMDs GPUs model numbers mean…

      I have a 3070 at home right now, what would be an upgrade from AMD?

      • Chais@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        5
        ·
        1 month ago

        Seems to be roughly on par with the RX 6800, according to user benchmarks.com, so everything after should be an improvement. The -XT models are generally stronger. Also, in an attempt to adapt their naming schema to nvidia’s amd’s 7900 series was succeeded by the 90xx series.
        Generally speaking nvidia seems to have the edge in ray-tracing, while amd offers faster rasterisation.