Paper Info
Title
An energy-efficient bandwidth allocation method for single-chip heterogeneous processor
Abstract
In recent years, many studies on optimization of energy consumption have focused on heterogeneous processor architectures. Heterogeneous computing model composed of CPU and GPU has developed from co-processing between the chips to CPU and GPU integration on the same die. On the one hand, because GPU can work under low voltage and ultra-low frequency, heterogeneous multi-core processors exhibit better energy efficiency compared to traditional CPU; On the other hand, because GPU usually has little or no shared cache, its memory bandwidth demand will be several times more than that of CPU for the same performance. Considering the constraints of power consumption and memory bandwidth, the contention problems of power and bandwidth among CPU and GPU in heterogeneous processor will inevitably occur. In this paper, energy-efficient bandwidth allocation method for heterogeneous processors allocates bandwidth to CPU and GPU and increases the overall throughput of CPU and GPU. Meanwhile the efficiency of energy consumption of heterogeneous processors is improved. The queuing delay is introduced to build a performance-energy model for heterogeneous processors. According to experience parameters, we analyze how the memory bandwidth allocation of off-chip affects the total throughput and energy efficiency. This paper also explores the change of performance-energy gains with the cache miss ratio of processors and the degree of scarce of available peak bandwidth. Based on the above research, the analytical solution of energy-efficient bandwidth allocation for optimal performance is obtained, and the solution can get more performance-energy gains compared with natural bandwidth allocation.
Year
DOI
Venue
2016
10.1109/IGCC.2016.7892617
2016 Seventh International Green and Sustainable Computing Conference (IGSC)
Keywords
Field
DocType
energy-efficient bandwidth allocation,Single-chip Heterogeneous Processor,multi-core processors,queuing delay,performance-energy gains
Memory bandwidth,Shared memory,Bandwidth allocation,Computer science,Parallel computing,Symmetric multiprocessor system,Bandwidth (signal processing),Dynamic bandwidth allocation,Multi-core processor,CPU shielding
Conference
ISBN
Citations 
PageRank 
978-1-5090-5118-2
0
0.34
References 
Authors
15
8
Name
Order
Citations
PageRank
Yatao Zhu100.68
Shuai Zhang221.38
Xiaochun Ye312528.41
Da Wang4448.79
Xu Tan5253.93
FAN Dong-Rui622238.18
Zhimin Zhang75411.10
Hongliang Li81833101.92