Skip to content

Affinitize benchmark process to performance cores by default #3266

Description

@MichalPetryka

Some modern CPUs have non heterogenous core configurations, meaning that some cores are faster than others.

On x86 Intel is currently doing it, AMD does it in limited capacity on laptops while on Arm64 most CPUs have it AFAIR.

This can result in wildly different timings between benchmark runs or even timing changes in the middle due to the OS moving the code to a different core.

BDN should probably default to affinitizing the benchmark process to only "performance" cores, witha way to somehow opt out/force "efficiency" ones.

On X86 the core types can be discovered with cpuid and information from CPU vendor manuals, no idea how to do it on arm64.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions