Explore checkpoint and restore for GPU functions
Open
@balajinvda is already working on this.
Since Jul 15, 2026.
roadmap
- Dominant language
- Go
- Stars
- 218
- Forks
- 72
- Avg merge
- 1d 12h
- Merged PRs (30d)
- 427
Description
Description
Explore checkpoint and restore support for NVCF GPU functions as a way to reduce cold-start latency and improve recovery. The proof of concept should determine whether an initialized function can be checkpointed after expensive startup work and restored on a compatible GPU node to serve traffic faster than a normal cold start.
Definition of Done
- A representative GPU function is checkpointed after model or runtime initialization.
- A restored function instance serves at least one validated request successfully.
- Cold-start and restore latency are measured across multiple runs.
- Constraints involving CUDA, drivers, GPU type, container runtime, process model, storage, security, scheduling, and observability are documented.
- The proof of concept ends with a recommendation to proceed, pause, or reject a product implementation.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.