ARGO: Aging-aware GPGPU register file allocation

ARGO: Aging-aware GPGPU register file allocation
复制标题

DOI:
10.1109/codes-isss.2013.6659017
复制
发表时间:
2013-09
期刊:
2013 International Conference on Hardware/Software Codesign and System Synthesis (CODES+ISSS)
影响因子:
--
通讯作者:
Majid Namaki-Shoushtari;Abbas Rahimi;N. Dutt;Puneet Gupta;Rajesh K. Gupta
Majid Namaki-Shoushtari;Abbas Rahimi;N. Dutt;Puneet Gupta;Rajesh K. Gupta
中科院分区:
其他
文献类型:
--
作者:
Majid Namaki-Shoushtari;Abbas Rahimi;N. Dutt;Puneet Gupta;Rajesh K. Gupta

文献摘要

被引文献

相似文献

采用纳米级 CMOS 技术实现的最先进的通用图形处理单元 (GPGPU) 使用数百个集成片上资源为高度并行应用提供非常高的计算吞吐量。这些资源在应用程序执行过程中会受到压力,使其受到诸如负偏置温度不稳定性 (NBTI) 等退化机制的影响,从而对其可靠性产生不利影响。为了支持高度并行执行,GPGPU 包含大型寄存器文件 (RF),这是压力最大的 GPGPU 组件之一;然而,我们观察到典型通用内核的 RF 利用率严重不足(平均只有 46%)。我们推出了 ARGO,这是一种老化感知 GPGPU RF allOcator,它通过在整个 RF 中分配压力来机会性地利用 RF 未充分利用的情况。 ARGO 通过对有压力的银行进行故意的电源门控来实现射频银行的适当平衡。我们在 AMD Evergreen GPGPU 架构上展示了我们的技术,并表明 ARGO 将 NBTI 引起的阈值电压降级改善了高达 43%(平均 27%),从而将 RF 静态噪声容限改善了高达 46%(平均 30%)。此外,我们估计通过为未使用的组提供睡眠状态,泄漏功率可同时减少 54%。
State-of-the-art general-purpose graphic processing units (GPGPUs) implemented in nanoscale CMOS technologies offer very high computational throughput for highly-parallel applications using hundreds of integrated on-chip resources. These resources are stressed during application execution, subjecting them to degradation mechanisms such as negative bias temperature instability (NBTI) that adversely affect their reliability. To support highly parallel execution, GPGPUs contain large register files (RFs) that are among the most highly stressed GPGPU components; however we observe heavy underutilization of RFs (on average only 46%) for typical general-purpose kernels. We present ARGO, an Aging-awaRe GPGPU RF allOcator that opportunistically exploits this RF underutilization by distributing the stress throughout RF. ARGO achieves proper leveling of RF banks through deliberated power-gating of stressful banks. We demonstrate our technique on the AMD Evergreen GPGPU architecture and show that ARGO improves the NBTI-induced threshold voltage degradation by up to 43% (on average 27%), that yields improving RFs static noise margin up to 46% (on average 30%). Furthermore, we estimate a simultaneous reduction in leakage power of 54% by providing sleep states for unused banks.