GPU Snapshot

Summary

GPU Snapshot provides architectural support for GPU resilience, enabling efficient checkpoint offloading in GPU-dense systems where node-level reliability is increasingly important.

This work was published at the International Conference on Supercomputing (ICS) in 2019 and is closely related to my study of resilience trends in GPU-dense systems (Best Paper at SELSE 2019, invited to DSN 2019).

Links

• Legacy project page: kyushick.github.io/project/gpusnap/
• Publication: GPU Snapshot: Checkpoint Offloading for GPU-Dense Systems, ICS 2019