当前位置: X-MOL 学术arXiv.cs.AR › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Prefetcher-based DRAM Architecture
arXiv - CS - Hardware Architecture Pub Date : 2021-05-21 , DOI: arxiv-2105.10427
Saurabh Jaiswal, Shailendra Kumar Gupta, Soumya Soubhagya Dandapat

Advancement in Processor technology has made it easy to handle data-intensive workloads, but limiting main memory advances has created performance bottlenecks. In DRAM, there have been improvements in DRAM access latency as well as reduction in cost-per-bit with the increase in cell density. But still DRAM data transfer rate lags behind the processing speed of the current generation processors. As Memory advancements based on hardware have been progressing at a slower pace, to cope up with High-end Processors, Architectural level advancements such as Prediction techniques, Replacement policies, etc are the major subject. In the recent field of research, Data prediction is a sought out topic as correct prediction can boost performance by decreasing the amount of excess memory access by predicting data beforehand using data access trends and behaviors. Though prediction techniques have been implemented at most of the Computer Architecture, We propose implementing data prediction in DRAM level architectures like TL-DRAM and CROW. Both of these method distributes the DRAM into different parts which contain a smaller section which is faster and larger section which contains the bulk of data but is comparatively slower. We wish to use data prediction in between these sections of memory to have predicted data transferred to the faster sections to improve the overall performance by reducing the memory access time.

中文翻译:

基于预取器的DRAM架构

处理器技术的进步使处理数据密集型工作负载变得容易,但是限制主内存的进步却造成了性能瓶颈。在DRAM中,随着单元密度的增加,DRAM的访问延迟得到了改善,每位成本也有所降低。但是,DRAM数据传输速率仍落后于当前处理器的处理速度。随着基于硬件的内存进步一直在以较慢的速度发展,以应付高端处理器,诸如预测技术,替换策略等架构级别的进步成为主要课题。在最近的研究领域,数据预测是一个热门话题,因为正确的预测可以通过使用数据访问趋势和行为预先预测数据来减少过多的内存访问量,从而提高性能。尽管大多数计算机体系结构都已实现了预测技术,但我们建议在TL-DRAM和CROW等DRAM级别的体系结构中实现数据预测。这两种方法都将DRAM分配到不同的部分,这些部分包含一个较小的部分,该部分较快,而一个较大的部分,其包含大量数据,但相对较慢。我们希望在存储器的这些部分之间使用数据预测,以将预测的数据传输到更快的部分,从而通过减少存储器访问时间来提高整体性能。我们建议在TL-DRAM和CROW等DRAM级别的体系结构中实现数据预测。这两种方法都将DRAM分配到不同的部分,这些部分包含一个较小的部分,该部分较快,而一个较大的部分,其包含大量数据,但相对较慢。我们希望在存储器的这些部分之间使用数据预测,以将预测的数据传输到更快的部分,从而通过减少存储器访问时间来提高整体性能。我们建议在TL-DRAM和CROW等DRAM级别的体系结构中实现数据预测。这两种方法都将DRAM分配到不同的部分,这些部分包含一个较小的部分,该部分较快,而一个较大的部分,其包含大量数据,但相对较慢。我们希望在存储器的这些部分之间使用数据预测,以将预测的数据传输到更快的部分,从而通过减少存储器访问时间来提高整体性能。
更新日期:2021-05-25
down
wechat
bug