• DocumentCode
    528450
  • Title

    Graphics performance analysis using Amdahl´s law

  • Author

    Issa, Joseph ; Figueira, Silvia

  • Author_Institution
    Dept. of Comput. Eng., Santa Clara Univ., Santa Clara, CA, USA
  • fYear
    2010
  • fDate
    11-14 July 2010
  • Firstpage
    127
  • Lastpage
    132
  • Abstract
    The performance analysis of graphics processors is important for designing future graphics processors. The analysis can help graphics hardware vendors to compare and analyze performance on different benchmarks, as well as help fine tune current architectures to achieve higher performance. In this paper, we discuss a performance analysis model for graphics processors using Amdahl´s law regression method. We present a performance analysis method for different graphics vendors using industry standard benchmarks such as 3DMarkVantage and Crysis game. In particular, we analyze performance from a measured baseline to different core and memory frequencies, as well as different number of shaders, for different discrete graphics cards on a given benchmark. The goal is to achieve performance analysis results with an acceptable error difference (<;;10%) between predicted and measured performance data.
  • Keywords
    computer graphic equipment; microprocessor chips; regression analysis; 3DMarkVantage; Amdahl law regression method; Crysis game; discrete graphics card; graphics hardware; graphics performance analysis; graphics processor; graphics vendor; industry standard benchmark; memory frequency; Benchmark testing; Equations; Frequency measurement; Graphics; Mathematical model; Program processors; Time domain analysis; Amdahl´s law; graphics performance; graphics performance prediction; performance analysis; performance measurement; performance modeling;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Performance Evaluation of Computer and Telecommunication Systems (SPECTS), 2010 International Symposium on
  • Conference_Location
    Ottawa, ON
  • Print_ISBN
    978-1-56555-340-8
  • Type

    conf

  • Filename
    5588709