Strange limit in AMD OpenCL driver

Discussion in 'Tools and Software' started by CarstenS, Mar 20, 2021.

  1. CarstenS

    Legend Veteran Subscriber

    May 31, 2002
    Likes Received:
    TIL that on a Tahiti board with 6 instead of the usual 3 GiByte of VRAM, the AMD OpenCL driver seems to think 3 GiByte should suffice for everyone.

    It's CL_DEVICE_GLOBAL_MEMORY is set to 3072 MiByte, leaving half of the memory effectively useless. Is there any known workaround, an environment variable or such or is this a hard, articifical and stupid limitation?

    In OpenGL and D3D, I can use the memory in games alright and don't get perf degradation in situations between 3 and 6 GiByte, so I am inclined to assume it's a real board with 6 GiBytes of memory and not just someone having manipulated the BIOS.
  2. JoeJ

    Veteran Newcomer

    Apr 1, 2018
    Likes Received:
    Just tried with my Vega56, where i get the full 8GB. Guess it's a driver bug and they might fix it if you report. Assuming there's still someone left to work on OpenCL, in his sparetime, maybe... Ryzen CPU is no longer listed as CL device :(

    Available Platform Vendor (0): Advanced Micro Devices, Inc.
    Device 0 gfx900 Device ID is 697402416

    Selected Platform Vendor: Advanced Micro Devices, Inc.
    Profile: FULL_PROFILE
    Version: OpenCL 2.1 AMD-APP (3075.13)
    Device: gfx900
    Version: OpenCL 2.0 AMD-APP (3075.13)
    Adress Bits: 64
    Extensions: cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_gl_sharing cl_khr_gl_depth_images cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_media_ops2 cl_amd_popcnt cl_khr_d3d10_sharing cl_khr_d3d11_sharing cl_khr_dx9_media_sharing cl_khr_image2d_from_buffer cl_khr_subgroups cl_khr_gl_event cl_khr_depth_images cl_khr_mipmap_image cl_khr_mipmap_image_writes cl_amd_liquid_flash cl_amd_copy_buffer_p2p cl_amd_planar_yuv
    Global Memory Size: 8573157376
    Global Cache Size: 16384
    Cache Line: 64
    Local Memory Size: 32768
    Max Work Group Size: 256
    Max Work Item Sizes: [1024,1024,1024]
    Max Image Width: 16384

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.