Help with a AMD GPU in truenas scale - eviltoast

Hello c/Selfhosted!

Although I’m still new with truenas, I’ve been a happy truenas scale hoster for a year more or less and I’ve been increasing the reach of my self hosted server little by little.

The problem came when I decided to add jellyfin and a GPU for encoding. My server is mostly made of old parts and the GPU is not different. The GPU is recognized by truenas scale as a “Advanced Micro Devices, Inc. [AMD/ATI] Cape Verde PRO [Radeon HD 7750/8740 R7 250E”, which AFAIK has hardware encoding/decoding as per Jellyfin wiki.

But the only place I can see the GPU is in lspci and in System Settings/Isolated GPU PCI Ids (and it’s not isolated). Whenever I try to change the configuration of an app to allocate the GPU I can only select “Allocate 0 amd.com/gpu GPU”, there are no more options.

I’ve searched for this a lot but I found very little info about AMD GPUs and how to debug this issue.

I’am missing something? Could anybody point me in the right direction? Any commands I can run to diagnose?

Thanks for reading!

  • Fenixin@sh.itjust.worksOPEnglish
    2·
    4 months ago

    Thanks for the answer.

    Just checked with lspci -v and it says radeon:

    03:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Cape Verde PRO [Radeon HD 7750/8740 / R7 250E] (prog-if 00 [VGA controller])
        Subsystem: PC Partner Limited / Sapphire Technology Cape Verde PRO [Radeon HD 7750/8740 / R7 250E]
        Flags: bus master, fast devsel, latency 0, IRQ 28, NUMA node 0
        Memory at d0000000 (64-bit, prefetchable) [size=256M]
        Memory at fbf80000 (64-bit, non-prefetchable) [size=256K]
        I/O ports at e000 [size=256]
        Expansion ROM at 000c0000 [disabled] [size=128K]
        Capabilities: [48] Vendor Specific Information: Len=08 <?>
        Capabilities: [50] Power Management version 3
        Capabilities: [58] Express Legacy Endpoint, MSI 00
        Capabilities: [a0] MSI: Enable+ Count=1/1 Maskable- 64bit+
        Capabilities: [100] Vendor Specific Information: ID=0001 Rev=1 Len=010 <?>
        Capabilities: [150] Advanced Error Reporting
        Capabilities: [270] Secondary PCI Express
        Kernel driver in use: radeon
        Kernel modules: radeon, amdgpu

    Is that the one it’s supposed to be active?

    About the allocate thing, I think that is a number and not an index in an array. I think that you can allocate as many gpus you want:

    I’ve seen that same menu in screenshots and I’m supposed to be able to choose something different from zero.

    • Vik@lemmy.worldEnglish
      2·
      4 months ago

      Since this is GCN based, you may be able to use the newer AMDGPU kernel driver? I’m not sure about that specific SKU, however. I remember that using AMDGPU on, for example, Hawaii (like the R9 290) was particularly finicky

      • Fenixin@sh.itjust.worksOPEnglish
        1·
        4 months ago

        Not sure how to change the driver to AMDGPU, I blacklisted the Radeon driver but the kernel didn’t load the other one. I read somewhere that I have to do a initframs update but the command doens’t exist in truenas scale. How do I force to load the other driver?

        • chameleon@fedia.io
          3·
          4 months ago

          For that card, you probably have to set the radeon.si_support=0 amdgpu.si_support=1 kernel options to allow amdgpu to work. I don’t have a TrueNAS system laying around so I don’t know what the idiomatic way to change them is.

          Using amdgpu on that card has been considered experimental ever since it was added like 6 years ago, and nobody has invested any real efforts to stabilize it. It’s entirely possible that amdgpu on that card is simply never gonna work. But yeah I think the radeon driver isn’t really fully functional anymore either, so I guess it’s worth a shot…

          • vividspecter@lemm.eeEnglish
            3·
            4 months ago

            Using amdgpu on that card has been considered experimental ever since it was added like 6 years ago

            If I recall right, it hasn’t been enabled by default simply because it is missing some features like analog TV out support (which most people don’t want or need in 2024).