1. Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

GPU Load parameter not making sense

Discussion in 'GPU-Z' started by Iridar, May 11, 2013.

  1. Iridar New Member

    Joined:
    May 11, 2013
    Messages:
    2 (0.00/day)
    Thanks Received:
    0
    I am doing a research on whether GPGPU (CUDA in particular) technologies are good for video decoding.
    So far my results are promising. I've been using GPU-Z to measure GPU and Video Engine load while playing video with regular, CPU-sofware codecs (DTV-DVD Microsoft) and GPGPU-based codecs (LAV Filters, which support CUDA hardware acceleration).

    In the first case, with regular codecs I get the following picture:
    [​IMG]
    Video decoding process is carried out on CPU, GPU only participates in rendering videoframes to screen.

    In second case, with GPU hardware acceleration, I get the following picture:
    [​IMG]
    Video decoding is carried out on GPU, and yet, GPU load is lower than in the first case, I find this very confusing.

    Could someone, perhaps, shed a light on this phenomenon?
    Perhaps, GPU load is shown in correlation with power consumption?
    In example, at X power consumption GPU is operationg at Y load of possible maximum at THIS power consumption rate, but when decoding video on GPU power consumption rises, and maximum possible GPU load also rises, and so overall GPU load seems lower, yet it's actually bigger than in the first case?
    Am I making any sence here? =)

    All test are carried out with nVidia GTX 660 OC, if this matters.
    I suppose, Memory Controller load is reduced in second case because memory transfer operations are carried out via DXVA hardware accelerated circiuts of Video Engine.
     
  2. TheLaughingMan

    TheLaughingMan

    Joined:
    May 7, 2009
    Messages:
    4,998 (2.46/day)
    Thanks Received:
    1,291
    Location:
    Marietta, GA USA
    The GPU could be working in two different modes. When it is not being used it is in 2D mode (usually 150 MHz to 250 MHz) because it only needs enough power to render the screen. So you are using 40% of this stated or (12% total performance). Then you turn on GPU accel which switches the card to full power mode when you start the decoding. Then your clock is around 1.15 GHz and you are using 21% of that. That would overall be more work, but still seems low to me.
     
    Iridar says thanks.
  3. Iridar New Member

    Joined:
    May 11, 2013
    Messages:
    2 (0.00/day)
    Thanks Received:
    0
    Thanks! That's exactly how it must be! I forgot to mention that, yeah, it switches GPU Core and Memory clocks to their maximum possible values while decoding video on GPU.
    You have no idea how much your answer helped me =)
    Best of all to you, good man!
     

Currently Active Users Viewing This Thread: 1 (0 members and 1 guest)

Share This Page