Skip to content

Optimal number of workers for parallel computations should be lower than cpu count #161

Description

@dbrakenhoff

Perhaps not very surprising, but the setting the number of workers to mp.cpu_count() when parallel=True does not yield optimal performance. From testing on different users computers, results were a little hard to interpret, for some it worked fine with parallel=True, for others the computation slowed down compared to the single core computation. Overall though, I think it would better to default to mp.cpu_count() // 2 or maybe even a bit less as the number of workers, which should deliver a speedup for most people.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions