Skip to content

GitLab

  • Menu
Projects Groups Snippets
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
    • Contribute to GitLab
  • Sign in / Register
  • A amd
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
  • Issues 1,283
    • Issues 1,283
    • List
    • Boards
    • Service Desk
    • Milestones
  • Merge requests 0
    • Merge requests 0
  • CI/CD
    • CI/CD
    • Pipelines
    • Jobs
    • Schedules
  • Deployments
    • Deployments
    • Environments
    • Releases
  • Monitor
    • Monitor
    • Incidents
  • Packages & Registries
    • Packages & Registries
    • Container Registry
  • Analytics
    • Analytics
    • Value stream
    • CI/CD
    • Repository
  • Wiki
    • Wiki
  • Snippets
    • Snippets
  • Activity
  • Graph
  • Create a new issue
  • Jobs
  • Commits
  • Issue Boards
Collapse sidebar
  • drm
  • amd
  • Issues
  • #1480

Closed
Open
Created Feb 08, 2021 by Alexander Monakov@amonakov

Renoir: wrong GPU timestamps

I am using Vulkan timestamp queries, and unfortunately I see a discrepancy about a factor of two: with a long-running Vulkan command buffer (from 10 ms up to 400ms), GPU timestamping reports 2x time compared to CPU monotonic clock.

Vulkan timestamps are in abstract units that the application needs to scale by timestampPeriod value. On Renoir, I see timestampPeriod 20.8333ns, which corresponds to a 48MHz clock, but a value closer to 10ns, corresponding to a 96MHz or 100MHz clock, would give more realistic results.

From inspecting Mesa, libdrm, and Linux kernel, I think it boils down to the reference clock rate returned by drivers/gpu/drm/amd/amdgpu/soc15.c:soc15_get_xclk. I see it was already patched once for a previous APU generation.

Assignee
Assign to
Time tracking