extensions/NVX/NVX_gpu_memory_info.txt

Name

    NVX_gpu_memory_info

Name Strings

    GL_NVX_gpu_memory_info

Contact

    Howard Stroyan, NVIDIA Corporation (hstroyan 'at' nvidia.com)

Contributors

    Eric Werness
    Evan Hart
    Mark Kilgard

Status

    Complete.  Shipping in NVIDIA 195.XX drivers

    NVIDIA's Tegra drivers will not expose this extensions.

Version

    Last Modified Date: October 3, 2013
    Author Revision: 1.4

Number

    438

Dependencies

    OpenGL 2.0 is required.
    The extension is written against the OpenGL 2.0 Specification.

Overview

    Most graphics systems offer a limited amount of onboard
    high-performance memory for storing textures, geometric
    primitives, and other data used for rendering.

    OpenGL implementations are expected to manage the residence of
    objects (that is, the memory pools in which objects are placed)
    automatically.  This is simple for applications to use, and the
    high level of abstraction allows many different underlying
    hardware implementations.  However performance sensitive 
    applications that are willing to adjust their usage of these memory 
    resources in order to maintain their desired performance are unable 
    to determine when the limited onboard memory resources are 
    approaching full utilization and swapping (with its related 
    performance impact) is imminent.

    GL_NVX_gpu_memory_info provides applications visibility into GPU 
    hardware memory utilization in order to allow the application to 
    effectively manage its resource allocations in the scope of the 
    current available GPU memory.  This information is made available 
    to the applications in the form of the total available resource 
    size (after initial system allocations) and the current available 
    resource (e.g. free memory) as well as a count of the number and
    total size of evictions of data from GPU memory since the last time
    this information was queried from this context using this extension.

New Procedures and Functions

        none

New Tokens

    Accepted by the <pname> parameter of GetIntegerv:
 
        GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX          0x9047
        GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX    0x9048
        GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX  0x9049
        GPU_MEMORY_INFO_EVICTION_COUNT_NVX            0x904A
        GPU_MEMORY_INFO_EVICTED_MEMORY_NVX            0x904B

Additions to Chapter 2 of the OpenGL 2.0 Specification (OpenGL Operation)

        none

Additions to Chapter 3 of the OpenGL 2.0 Specification (Rasterization)

        none

Additions to Chapter 4 of the OpenGL 2.0 Specification (Per-Fragment
Operations and the Frame Buffer)

        none

Additions to Chapter 5 of the OpenGL 2.0 Specification (Special Functions)

        none

Additions to Chapter 6 of the OpenGL 2.0 Specification (State and
State Requests)

    Section 6.1.x, GPU Memory Information Query

    Inserted between sections 6.1.14 and 6.1.15 

    Information on GPU memory resources available can be queried
    by specifying one of: 

        GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX 
        GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX 
        GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX 
        GPU_MEMORY_INFO_EVICTION_COUNT_NVX 
        GPU_MEMORY_INFO_EVICTED_MEMORY_NVX 

    as the value parameter to GetIntegerv.

    These will return the memory status of the context's associated GPU memory.  
    The status returned is not intended as an exact measurement of the 
    system's current status.  It will provide a approximate indicator of 
    the overall GPU memory utilization so that an application can determine 
    when the resource usage is approaching the full capacity of the GPU memory 
    and it may need to adjust its usage pattern to avoid performance limiting 
    swapping.  Each query returns a integer where the values have the 
    following meanings:

    GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX 
      - dedicated video memory, total size (in kb) of the GPU memory

    GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX 
      - total available memory, total size (in Kb) of the memory 
        available for allocations

    GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX 
      - current available dedicated video memory (in kb), 
        currently unused GPU memory

    GPU_MEMORY_INFO_EVICTION_COUNT_NVX 
      - count of total evictions seen by system

    GPU_MEMORY_INFO_EVICTED_MEMORY_NVX 
      - size of total video memory evicted (in kb)

    The values reported are subject to change and cannot be used as a
    guarantee of successful future allocations.  The current available 
    dedicated memory will be less than the total GPU memory size since 
    system allocations for shared resources (i.e. desktop surfaces) will 
    be accounted for.  The reported current available GPU memory will be 
    influenced by the current binding state of resources such as 
    textures and object buffers that can potentially migrate between 
    system and dedicated GPU memory based on their usage.  The eviction 
    count indicates that number of eviction events that caused 
    an allocation to be removed from dedicated video memory to free GPU 
    video memory to make room for other allocations.  The eviction size 
    is the total size of memory released as a result of these evictions.
    These eviction values are running counters and may not start at zero
    on the first query.  This extension does not attempt to account for 
    any potential fragmentation issues that may limit full utilization 
    of the GPU memory resource depending on implementation.

Additions to Appendix A of the OpenGL 2.0 Specification (Invariance)

        none

Additions to the AGL/GLX/WGL Specifications

        None

Errors

        None

Issues

1)    Implementing the eviction information is OS dependent.  
      For XP and Linux the eviction information is specific to the current process/state
      since eviction is determined in the individual client.
      For Vista it is system wide since eviction is determined by the OS.

2)    Should Tegra advertise and support this extension?

      RESOLVED:  No.  Tegra's unified memory architecture doesn't sensibly
      map to the queries of this extensions.

      A future extension is needed to address this.

Revision History

   1.2 Added eviction count and size information.
   1.3 Changed to separate enums / GetIntegerv calls for each value.
       Removed reset of eviction data on query. 
   1.4 Document Tegra will not support this extension.