Skip to content

Load balance when one GPU gets close to memory overflow #714

Open
@MaxThevenet

Description

@MaxThevenet

So far, using GPUs efficiently means filling the GPU memory as much as possible. We could have the option to load balance when the free memory on a GPU gets dangerously small (e.g., 10% of available memory).

Metadata

Metadata

Assignees

No one assigned

    Labels

    backend: cudaSpecific to CUDA execution (GPUs)component: load balancingLoad balancing strategies, optimization etc.

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions