Efficient Design Space Exploration of GPGPU Architectures

Abstract

The goal of this work is to revisit GPU design and introduce a fast, low-cost and effective approach to optimize resource allocation in future GPUs. We have achieved this goal by using the Plackett-Burman methodology to explore the design space efficiently. We further formulate the design exploration problem as that of a constraint optimization. Our approach produces the optimum configuration in 84% of the cases, and in case that it does not, it produces the second optimal case with a performance penalty of less than 3.5%. Also, our method reduces the number of explorations one needs to perform by as much as 78%.

Multi-objective, Energy-Aware GPGPU Design Space Exploration for Medical or Industrial Applications

“…Mirsoleimani et al [6] also employ a linear regression model, which was used to find the most important GPU parameters affecting the run time of different GPGPU programs, where GPGPU is the acronym for General Purpose Computation on Graphics Processing Units. Jooya et al [7] presented an approach to find the GPU configuration which delivers the best run time under a given transistor budget. It uses the Plackett-Burman scheme to set up the experiments and solves a knapsack problem.…”

Full-System Simulation of Mobile CPU/GPU Platforms

“…3a gives an overview of the Bifrost architecture. 2 The architecture features up to 32 unified Shader Core (SC)s, and a single logical L2 GPU cache that is split into several fully coherent physical cache segments. Full system coherency support and shared main memory tightly couples the GPU and CPU memory systems.…”