AMD EPYC™ 7002 Series – A Processor with Improved Soft Error Resilience

Computing and networking equipment power most of the applications we use daily. Therefore, dependability of these systems is of utmost importance. The impact of the neutron induced soft errors on the AMD EPYC™ 7002 Series processor is examined in this paper. The architecture of the processor and fault/error handling mechanisms are briefly presented. The experimental setup, used at Los Alamos Neutron Science Center (LANSCE) for soft error characterization, is described and the test results are analyzed. Two workloads are employed in the accelerated measurements, the high performance Linpack (HPL) benchmark and the STREAM benchmark. The neutron impact is quantified by the mean-time to failure (MTTF). The AMD EPYC 7002 MTTF is also compared against the MTTF of the previous generation of processors, AMD EPYC™ 7001, and confidence intervals are derived. The OS error logs are collected throughout the experiment. The measurement results are discussed.