Cuda Driver API checkpoint limitations

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi! I noticed that recently the cuda driver API exported checkpoint functionality (CUDA Driver API :: CUDA Toolkit Documentation) which seems very similar to the cuda checkpoint utility (GitHub - NVIDIA/cuda-checkpoint: CUDA checkpoint and restore utility).

Does the driver API have the same limitations as cuda-checkpoint (like not supporting UVM)?

I can’t find anything in the API docs about this but I’d imagine that the cuda-checkpoint utility is calling into these driver functions under the hood so they should have the same limitations.

Thanks!