HybridMemoryModeOptions#

class nvmath.sparse.advanced.HybridMemoryModeOptions( hybrid_memory_mode: bool = False, hybrid_device_memory_limit: int | str | None = None, register_cuda_memory: bool = True, )[source]#

A data class for providing options related to the use of hybrid (CPU-GPU) memory to those execution spaces that support it.

hybrid_memory_mode#

If set to True, use CPU memory to store factors (default: False). See nvmath.bindings.cudss.ConfigParam.HYBRID_MODE.

Type:: bool

device_memory_limit#: The maximum device memory available for execution. It can be specified as a value (with optional suffix like K[iB], M[iB], G[iB]) or as a percentage. The default is based on internal heuristics. See nvmath.bindings.cudss.ConfigParam.HYBRID_DEVICE_MEMORY_LIMIT.

register_cuda_memory#

Specify whether to register memory using cudaHostRegister() if hybrid memory mode is used. The default is True. See nvmath.bindings.cudss.ConfigParam.USE_CUDA_REGISTER_MEMORY.

Type:: bool

See also

ExecutionHybrid, DirectSolver, direct_solver().