Experimental Findings on the Sources of Detected Unrecoverable Errors in GPUs

TL;DR

This paper investigates the sources of Detected Unrecoverable Errors in GPUs exposed to neutron beams, identifying key causes and evaluating the impact of ECC on error reduction.

Contribution

It provides empirical data on error sources in GPUs under neutron exposure and quantifies the effectiveness of ECC in reducing specific error types.

Findings

ECC reduces DUEs caused by Illegal Address access by up to 92% in Kepler GPUs.

ECC reduces DUEs caused by Illegal Address access by up to 98% in Volta GPUs.

Illegal memory accesses and interface errors are significant sources of DUEs.

Abstract

We investigate the sources of Detected Unrecoverable Errors (DUEs) in GPUs exposed to neutron beams. Illegal memory accesses and interface errors are among the more likely sources of DUEs. ECC increases the launch failure events. Our test procedure has shown that ECC can reduce the DUEs caused by Illegal Address access up to 92% for Kepler and 98% for Volta.