1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Will it be possible to mod DX12 Support for HD 5000 and 6000?

Discussion in 'Videocards - AMD Radeon Drivers Section' started by mcfart, Jun 26, 2015.

  1. Atmosphere__

    Atmosphere__ New Member

    Messages:
    8
    Likes Received:
    1
    GPU:
    RX480/8G
    The beta 17.12 beta didnt have vulcan included, because vulcan is based on both OpenGL and Mantle. Thus drivers were only created for GCN and not TerraScale, which is a VLIW (Very Long Instruction Word).

    VLIW: Essentially 1 instruction can do multiple actions on different pieces of data, but the actions so a software compiler actually creates a flow/wave of instructions, based on the driver input. This stream of instructions is then executed on the chip. TerraScale is best compared with a sort of Fixed function cluster, where GCN has a more General Purpose approach.
    So your ‘Mod’, is actually a per-app tunable instruction compiler which breaks up all of the instructions for a given 3D scene (or part of) into groups in basic operations, ordered by dependancy on a computation-result.

    It is even more complicated than that, but as you might known the GCN-setup is the main thing why parallel processing is possible in hardware at all, but the work-pooling is quite a PiA if you want the last drop of computational power.

    Still on my HD6870-2GB in cf... and 17.11 drivers... i enjoyed that combo for a long time.. pitcairn, at Tahiti level performance and beyond in some titles..
    Now enjoying polaris :cool:

    Lol, didnt just noticed this topic started almost 4 years ago
     
  2. stormy

    stormy Ancient Guru

    Messages:
    2,578
    Likes Received:
    14
    GPU:
    Gigabyte RX 570
    Maybe four years old, but does anyone else have déjà vu about the whole DX10 on XP when Vista came out?
     
  3. Alessio1989

    Alessio1989 Maha Guru

    Messages:
    1,299
    Likes Received:
    196
    GPU:
    .
    Nothing do to with OS version.
     
  4. user1

    user1 Maha Guru

    Messages:
    1,335
    Likes Received:
    416
    GPU:
    hd 6870
    Native Dx12 support is not possible, pre-gcn cards lack the ability to support wddm2.0's unified memory model, gcn and anything fermi or newer can access system memory directly , where as vliw(hd2000-6000) cannot , you must first copy the memory to its onboard memory then read, this is very slow , this "memory swaping" is why pregcn cards perform exceptionally poorly when using more video memory than vram is available.

    You could emulate this feature inorder to support wddm 2.0, but it would be very slow, rendering dx 12 support basically useless.

    Now vulkan support is another matter, I think it could be done, since it doesnt have the same requirements. It is however understandable why amd did not attempt to support vulkan, as it was not ready by the time of eol.
     
    Cave Waverider likes this.

  5. dezmand07

    dezmand07 Member

    Messages:
    47
    Likes Received:
    1
    GPU:
    Intel IrisProGraphics 580
    Where does emulation on the CPU? Read again what I wrote.
    I successfully initialized Dx12 from FL = D3D_FEATURE_LEVEL_11_0 on old video cards.
    Initialization and full draw of course different things, but no emulation is needed when initializing.
     
  6. user1

    user1 Maha Guru

    Messages:
    1,335
    Likes Received:
    416
    GPU:
    hd 6870
    Has nothing to do with the directx featurelevel, its the memory model, the MMU on pre-gcn cards do not support the unified memory model that wddm2.0 requires , gcn and fermi have a feature called "unified virtual memory" where virtual address space is shared between the cpu and gpu,
    terascale and earlier cards do not support this , they have separate address pools, when i mean emulate, i mean emulating the "unified virtual memory".
     
    Cave Waverider likes this.

Share This Page