Been using AMD gpus for years due to their superb Linux support. Would like to play around with AI image generation completely offline so a Nvidia gpu is kinda must. Besides that I do some gaming as well.

Currently I have AMD RX 6750 XT and am considering replacing it with a RTX 5060 TI with 16 GB vram. How well does Nvidia gpus work in modern Linux systems. I’m running Tumbleweed with KDE plasma on X11.

Other hardware is Ryzen 7 7700X and an Asrock B650 PG Lightning motherboard.

Edit: Been using InvokeAI so far but cpu only is super slow.

  • FrostyPolicy@suppo.fiOP
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    1 day ago

    Wish it was that simple. If I select AMD device to use it won’t work. It crashes with: RuntimeError: HIP error: invalid device function.

    Edit: Have a feeling it might be trying to use the igpu for some reason.

    • verdi@tarte.nuage-libre.fr
      link
      fedilink
      Français
      arrow-up
      2
      ·
      1 day ago

      Just checked RDNA2 may not be fully supported HIP. Check this REDDIT guide for llama based adventures.

      Did u run

      rocm-smi

      to check if you have rocm installed?

      Check this issue at invoke’s hub for your same problem.

      critically, don’t forget:

      #!/bin/bash

      export HSA_OVERRIDE_GFX_VERSION=10.3.0

      bash invoke.sh

      Regarding upgrades

      The 9060XT would be a much better choice than the 5060 if one uses Linux and is upgrading, although, seeing price trends lately, nothing looks like a good choice at all at the moment. Maybe a second hand 7800xt, that would be a 20-30% upgrade.

      • FrostyPolicy@suppo.fiOP
        link
        fedilink
        English
        arrow-up
        1
        ·
        23 hours ago

        Did u run

        rocm-smi
        

        It gives:

        ========================================== ROCm System Management Interface ==========================================
        ==================================================== Concise Info ====================================================
        Device  Node  IDs              Temp    Power    Partitions          SCLK    MCLK     Fan  Perf  PwrCap  VRAM%  GPU%
        (DID,     GUID)  (Edge)  (Avg)    (Mem, Compute, ID)
        ======================================================================================================================
        0       1     0x73df,   60335  40.0°C  8.0W     N/A, N/A, 0         500Mhz  96Mhz    0%   auto  220.0W  19%    0%
        1       2     0x164e,   6110   39.0°C  29.047W  N/A, N/A, 0         N/A     3000Mhz  0%   auto  N/A     4%     0%
        ======================================================================================================================
        ================================================ End of ROCm SMI Log =================================================
        
      • FrostyPolicy@suppo.fiOP
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        23 hours ago

        You’re a life saver! Did the force reinstall mentioned in that github issue and ran Invoke with the command you provided. Now it runs with gpu rendering 🥳. Though I ran out of vram 🤣.

        • verdi@tarte.nuage-libre.fr
          link
          fedilink
          Français
          arrow-up
          1
          ·
          21 hours ago

          I’m happy it worked, props to the people helping on github, they’re the real heroes here, I’m just the messenger! Have fun m8!