Module gpu_sharing 

GPU sharing support via NVIDIA MPS (Multi-Process Service) and time-slicing.

MPS allows multiple containers to share a single GPU with hardware-level isolation of compute resources. Time-slicing provides simpler round-robin GPU sharing without concurrent kernel execution.

Structs

MpsDaemonManager

Manages NVIDIA MPS daemon instances, one per physical GPU.

Enums

MpsError

Errors from MPS daemon operations