A survey of architectural techniques for DRAM power management

Abstract

Recent trends of CMOS technology scaling and widespread use of multicore processors have dramatically increased the power consumption of main memory. It has been estimated that modern data-centers spend more than 30% of their total power consumption in main memory alone. This excessive power dissipation has created the problem of "memory power wall" which has emerged as a major design constraint inhibiting further performance scaling. Recently, several techniques have been proposed to address this issue. The focus of this paper is to survey several architectural techniques designed for improving power efficiency of main memory systems, specifically DRAM systems. To help the reader in gaining insights into the similarities and differences between the techniques, this paper also presents a classification of the techniques on the basis of their characteristics. The aim of the paper is to equip the engineers and architects with knowledge of the state of the art DRAM power saving techniques and motivate them to design novel solutions for addressing the challenges presented by the memory power wall problem.

Experiences in autotuning matrix multiplication for energy minimization on GPUs

“…The center of the third row in Figure 9 indicates that DRAM-stressing kernels provide lower energy efficiency. This confirms research results identifying the memory access as being among the most energy-consuming operations [41][42][43]. At the same time, the bandwidth rates for all kernels exceeding 820 GFLOPs are above the mean (see the right plot).…”

A Survey of Techniques for Approximate Computing

“…By exploiting this margin, AC can aggressively improve performance and energy efficiency. For example, by intelligently reducing the eDRAM/DRAM refresh rate or SRAM supply voltage, the energy consumed in storage and memory access can be reduced with minor loss in precision [Mittal 2012[Mittal , 2014b. Similarly, the AC approach can alleviate the scalability bottleneck , improving performance by early loop termination [Sidiroglou et al 2011;, skipping memory accesses [Yazdanbakhsh et al 2015b], offloading computations to an accelerator , improving yield , and much more.…”

A Survey of Methods for Analyzing and Improving GPU Energy Efficiency

“…For every watt of power dissipated in the computing equipment, an additional 0.5 to 1W of power is consumed by the cooling system itself [Patel et al 2003], and with increasing ratio of cooling power to computing power, compaction of devices is inhibited, which results in increased operation costs. Due to these trends, in recent years, the energy cost of high-performance computing clusters has been estimated to contribute more than the hardware acquisition cost of IT equipment itself [Bianchini and Rajamony 2004;Mittal 2012].…”