Microarchitecture level reliability comparison of modern GPU designs: First findings

Abstract

State-of-the-art GPU chips are designed to deliver extreme throughput for graphics as well as for data-parallel general purpose computing workloads (GPGPU computing). Unlike graphics computing, GPGPU computing requires highly reliable operation. The performance-oriented design of GPUs requires to jointly evaluate the vulnerability of GPU workloads to soft-errors with the performance of GPU chips. We briefly present a summary of the findings of an extensive study aiming at the evaluation of the reliability of four GPU architectures and corresponding chips, orrelating them with the performance of the workloads.