Loading...

Improving the Utilization Rate of the GPU Memory Resources by Exploiting Application's Heterogeneity

Darabi Moghaddam, Sina | 2022

43 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 56242 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Sarbazi Azad, Hamid
  7. Abstract:
  8. In recent years, GPUs have become a popular choice for high-performance general-purpose systems. However, since general-purpose applications do not always utilize processing and memory resources efficiently, energy consumption in these systems has not been managed effectively. As the variety of resources within GPUs continues to increase, the problem of low utilization has become more critical. Past solutions have focused on integrating resources and using multi-tasking, but these methods have limitations in terms of performance and security. Therefore, this research proposes new methods for improving the resource utilization and energy efficiency of GPUs. This dissertation first examines the heterogeneity in GPU resource usage under general-purpose workloads. Next, a dynamic method for partitioning unified memory between shared memory and cache memory is presented. This method divides the memory space according to the program's needs during program execution, resulting in a 21% and 18% improvement over the base processor and the latest work in this direction. The proposal also introduces a new way to run multiple programs simultaneously on GPUs, utilizing space without using shared memories and the registry file of one program for another program. The results show a 26% improvement compared to spatial multi-tasking while maintaining service quality. Additionally, the proposal extends the second method to make unusable resources usable for the same program, leading to a 39% improvement in average performance compared to the base processor. Overall, the solutions presented in this research proposal aim to improve the energy consumption of graphics processors with general-purpose applications. Based on simulation results, the proposal achieves a 15%, 17%, and 58% reduction in energy consumption for the three tasks, respectively. These findings have important implications for the design and implementation of energy-efficient GPUs for a wide range of general-purpose applications
  9. Keywords:
  10. Graphic Processing ; Inhomogeneity ; Graphics Procssing Unit (GPU) ; Energy Efficiency ; Multiprogramming ; Computed Unified Device Architecture (CUDA)Framework ; Memory Space Integration

 Digital Object List

 Bookmark

No TOC