Thesis
Exploiting power-performance tradeoffs for GPU-NVM systems
Master of Science (M.S.), Drexel University
Jun 2021
DOI:
https://doi.org/10.17918/00000436
Abstract
The increase in popularity of compute-intensive applications such as deep learning and graph analytics brings about a surge in the usage of Graphics Processing Units due to the GPU architecture's ability to concurrently process large amounts of data. However, GPU is a power hungry hardware, which creates a scalability issue for large systems. This asks for better methods to reduce GPU power consumption, and researchers need tools to help them carefully analyze power metrics. Current GPU power simulators only provide models for outdated GPU generations, causing challenges for research and analysis of systems using modern generations of GPU. This thesis extends the power model of the NVIDIA GTX 480 provided by the GPGPU-Sim simulator to support modern architectures of NVIDIA GPUs using a counter-based methodology. We also explore the use of Non-Volatile Memory technologies as replacement of DRAM on GPUs to reduce power consumption while maintaining high performance, as emerging NVM technologies stand out as energy-efficient alternatives for memory storage.
Metrics
50 File views/ downloads
36 Record Views
Details
- Title
- Exploiting power-performance tradeoffs for GPU-NVM systems
- Creators
- Hieu Quang Mai
- Contributors
- Anup Das (Advisor)
- Awarding Institution
- Drexel University
- Degree Awarded
- Master of Science (M.S.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Number of pages
- vii, 31 pages
- Resource Type
- Thesis
- Language
- English
- Academic Unit
- College of Engineering (1970-2026); Electrical (and Computer) Engineering (1970-2026); Drexel University
- Other Identifier
- 991015242080304721