Argus-G: A Low-Cost Error Detection Scheme for GPGPUs
Ralph Nathan, Daniel J. Sorin · 2010
have developed Argus-G, a low-cost error detection mechanism for the SIMT cores found in GPGPUs. As GPUs make the transition into general purpose computing, detecting errors and dealing with them will become a more pressing issue. General purpose graphics processing units are increasingly used for scientific computing, where errors, if not detected, can significantly distort the results of these scientific simulations. Argus-G is an adaptation of the Argus error detection scheme for general purpose cores that has been tailored for GPUs. Our experiments show that, on average, our implementation incurs a 4% overhead in runtime and a 10% increase in the number of instructions executed.