cuda.core.WorkqueueResourceOptions

class cuda.core.WorkqueueResourceOptions(

sharing_scope: str | None = None,

concurrency_limit: int | None = None,

)

Customizable WorkqueueResource.configure options.

sharing_scope

Workqueue sharing scope. Accepted values: "device_ctx" or "green_ctx_balanced". (Default to None)

Type:

str, optional

concurrency_limit

Expected maximum number of concurrent stream-ordered workloads. Must be >= 1 when set. The effective driver-side cap is CUDA_DEVICE_MAX_CONNECTIONS (typically [1, 32]); configurations may exceed this cap, but the driver will not guarantee that work submission remains non-overlapping. (Default to None)

Type:

int, optional