“Parallel Computing with Multiple GPUs on a Single Machine to Achieve Performance Gains” by Gulde, Weeks, Owen and Pan

  • ©Robert Gulde, Michael Weeks, G. Scott Owen, and Yi Pan

Conference:


Type(s):


Entry Number: 095

Title:

    Parallel Computing with Multiple GPUs on a Single Machine to Achieve Performance Gains

Presenter(s)/Author(s):



Abstract:


    Using a technique similar to cluster based computing, this research demonstrates the use of multiple Graphics Processing Units (GPUs) to achieve enhanced performance on a single processor workstation. Past examples of performance gains achieved by cluster based computing in order to speed up the rendering process have been demonstrated. However, these approaches have neglected parallel processing capabilities through multiple graphics cards with hardware acceleration on a single CPU based computer. In this research a technique was developed to utilize multiple threads to drive hardware accelerated graphics cards.
    Performance gains achieved by this technique demonstrated that GPUs on a single CPU system follow both Gustafson’s Law of constant time as work load scales with the number of GPUs and Amdahl’s Law of speed up achieved through work division across processing elements.


Additional Images:

©Robert Gulde, Michael Weeks, G. Scott Owen, and Yi Pan

PDF:



ACM Digital Library Publication:



Overview Page: