“Half Frame Forwarding:Frame-rate Up Conversion for Tiled Rendering GPU” by Park, Kim, Ki, Seo and Shin

  • ©Jinhong Park, Minkyu Kim, Sunho Ki, Youngduke Seo, and Chulho Shin

  • ©Jinhong Park, Minkyu Kim, Sunho Ki, Youngduke Seo, and Chulho Shin

  • ©Jinhong Park, Minkyu Kim, Sunho Ki, Youngduke Seo, and Chulho Shin

Conference:


Type:


Entry Number: 36

Title:

    Half Frame Forwarding:Frame-rate Up Conversion for Tiled Rendering GPU

Presenter(s)/Author(s):



Abstract:


    Although the mobile industry has recently begun trending towards high quality graphics content, it is still difficult to satisfy this trend due to performance, power and thermal issue of GPU/CPU in mobile application processor.
    The power consumption of mobile GPU increases almost linearly with workload, depending on the graphics content. Thus, lowering the frame rate can reduce the power consumption. Some applications force the skipping of some frames during rendering in order to decrease the frame rate. Although this feature can be exploited at runtime in order to reduce the power consumption, it can make noticeable artifacts such as flickering and lagging. One common technique to help remedy these artifacts is a frame rate up conversion (FRUC) algorithm[Bowles 2012], which typically consist of two processes to determine the dynamic objects in consecutive frames: motion estimation and motion-compensated interpolation. But, those have high computational costs and, naturally, result in high power consumption. Thus, the conventional approaches are not suitable for mobile devices.
    In this work, we propose a novel FRUC algorithm based on a half frame forwarding approach with a low cost solution to detect dynamic objects for tile-based GPU rendering. The proposed algorithm makes intermediate frames using the tiles that cover a region with dynamic objects. Using an LG G3 Screen, our experimental result demonstrates that the proposed algorithm can reduce the system power consumption by up to 20.4%.


Additional Images:

©Jinhong Park, Minkyu Kim, Sunho Ki, Youngduke Seo, and Chulho Shin

PDF:



ACM Digital Library Publication:



Overview Page: