No announcement yet.

AMD Appears To Be Aiming For Good Radeon RDNA4 GPU Support In Linux 6.11

  • Filter
  • Time
  • Show
Clear All
new posts

  • #21
    Originally posted by Type44Q View Post

    If there's one place on the entire Web where demonstrating an awareness of syntax should matter, it's here.
    I need more information. Nvidia's drivers use a shim, what is wrong with that?


    • #22
      Originally posted by user556 View Post
      Reply is appreciated.

      Where does Mesa fit in? It doesn't need the kernel recompiled but it has driver components too, right?

      EDIT: Maybe I can partly answer this myself by reading the Mesa website ... It states it's primarily an API for OpenGL. So I guess that makes it one API but not *thee* Linux API. Linux being open to any alternatives as well. Would that be a reasonable point of view?
      Right... Mesa is an open source project which includes most of the user-space drivers. The top end of each of those drivers is generally an industry API (and a lot of the API-handling code is shared across vendors) while the bottom end is code that talks to the kernel drivers for each vendor's hardware.

      I say kernel drivers plural because open source drivers now support so many generations of hardware that there are typically a few different kernel drivers required even for a single vendor, since the hardware programming models have changed significantly over time. As an example, the kernel tree currently includes "radeon" and "amdgpu/amdkfd" kernel drivers for AMD GPUs, but there were also "mach64" and "r128" drivers for older hardware which have since been removed from the upstream tree.
      Test signature


      • #23
        Both windows and Linux have a similar driver model. A kernel mode driver which handles device init, modesetting, DMA, interrupts, etc. and a set of user mode drivers which translate higher level APIs (e.g., OpenGL, Vulkan, VAAPI, etc.) to device specific commands. The user mode drivers then interact with the kernel mode driver to submit the commands to the GPU.