1

Examining Failures and Repairs on Supercomputers with Multi-GPU Compute Nodes