Analyzing soft-error vulnerability on GPGPU microarchitecture

Authors:
Jingweijia Tan;Nilanjan Goswami;Tao Li;Xin Fu
Affiliations:
EECS Department, University of Kansas, USA;ECE Department, University of Florida, USA;ECE Department, University of Florida, USA;EECS Department, University of Kansas, USA
Venue:
IISWC '11 Proceedings of the 2011 IEEE International Symposium on Workload Characterization
Year:
2011

Citing 0
Cited 3

RISE: improving the streaming processors reliability against soft errors in gpgpus

Proceedings of the 21st international conference on Parallel architectures and compilation techniques
Warped-DMR: Light-weight Error Detection for GPGPU

MICRO-45 Proceedings of the 2012 45th Annual IEEE/ACM International Symposium on Microarchitecture
CPU-GPU hybrid bidiagonal reduction with soft error resilience

ScalA '13 Proceedings of the Workshop on Latest Advances in Scalable Algorithms for Large-Scale Systems

Quantified Score

Hi-index	0.00

Visualization

Abstract

The general-purpose computation on graphic processing units (GPGPU) becomes increasingly popular due to their high computational throughput for data parallel applications. Modern GPU architectures have limited capability for error detection and tolerance since they are originally designed for graphics processing. However, the rigorous execution correctness is required for general-purpose applications. This makes reliability a growing concern in GPGPU architecture design. With CMOS processing technologies continuously scaling down to the nano-scale, on-chip soft error rate (SER) has been predicted to increase exponentially. GPGPUs with hundreds of cores integrated in a single chip are prone to manifest high SER. This paper explores a first step to characterize GPGPU reliability in light of soft errors. We develop GPGPU-SODA (GPGPU Software Dependability Analysis), a framework to estimate the soft-error vulnerability of GPGPU microarchitecture. By using GPGPU-SODA, we observe that several microarchitecture structures in GPGPUs exhibit high soft-error susceptibility, and the structure vulnerability is sensitive to workload characteristics (e.g. branch divergences, memory coalescing). We further investigate several architectural optimizations. We find that both dynamic warp formation and increasing the number of threads supported by GPU largely affect the GPGPU soft-error robustness. However, changing the warp scheduling policy has minor impact on the structure vulnerability. The observations made in this study provide designers the useful guidance to build resilient GPGPUs: a comprehensive resiliency solution for GPGPUs should consider the entire GPGPU design instead of just focusing on a particular structure.