Kernel-Level Tolerance

Abstract

Negative bias temperature instability (NBTI) adversely affects the reliability of a processor by introducing new delay-induced faults. However, the effect of these delay variations is not uniformly spread across functional units and instructions: some are affected more (hence less reliable) than others. For massive number of kernels executing on functional units in GPUs, we propose a preventive method to ensure the absence of NBTI-induced timing errors during GPU lifetime . This chapter presents an NBTI-aware compiler-directed very long instruction word (VLIW) assignment scheme that uniformly distributes the stress of instructions with the aim of minimizing aging of GP-GPU architecture without any performance penalty. The proposed solution is an entirely software technique based on static workload characterization and online execution with NBTI monitoring that equalizes the expected lifetime of each processing element by regenerating aging-aware healthy kernels that respond to the specific health state of GP-GPU. We demonstrate our approach on AMD Evergreen architecture where iso-throughput executions of the healthy kernels reduce NBTI-induced voltage threshold shift up to 49% (11%) compared to naive kernel executions, with (without) architectural support for power-gating. The kernel adaption flow takes average of 13 ms on a typical host machine thus making it suitable for practical implementation.