Skip to content

ci(audience): shrink xvfb screen to 160x120 for llvmpipe fill rate#761

Closed
ImmutableJeffrey wants to merge 1 commit intochore/sdk-318-linux-playmode-xvfbfrom
chore/audience-linux-perf-xvfb-screen
Closed

ci(audience): shrink xvfb screen to 160x120 for llvmpipe fill rate#761
ImmutableJeffrey wants to merge 1 commit intochore/sdk-318-linux-playmode-xvfbfrom
chore/audience-linux-perf-xvfb-screen

Conversation

@ImmutableJeffrey
Copy link
Copy Markdown
Collaborator

Summary

  • xvfb virtual screen drops from 320x240x24 (76 800 px) to 160x120x24 (19 200 px). 4x fewer pixels per frame.
  • Unity -screen-width and -screen-height track the new size.

Why

llvmpipe fragment-fill cost scales linearly with pixel count. The previous reduction (1280x720 to 320x240) cut fill 12x. Going to 160x120 cuts another 4x.

Risk

UI Toolkit may fail to lay out below some threshold. The SampleApp tests assert on the visual element tree, not on rendered pixels, so layout only needs to be valid; visible quality is irrelevant. If tests come back inconclusive on this PR's CI run, that means the floor is between 320x240 and 160x120 and we step back.

Experiment success criteria

  • Unity 6 Linux Mono cell wall time under 18 min (currently around 27 min).
  • All 4 Linux PR cells go green.

🤖 Generated with Claude Code

@ImmutableJeffrey
Copy link
Copy Markdown
Collaborator Author

Closing without merging. xvfb screen 160x120 did not reduce Unity 6 cell time (31 vs 27 min baseline). Profile data on PR #764 reveals xvfb-run --server-args was a no-op anyway (player log reports desktop 1280x1024 regardless of the screen flag). Removed in the new PR. Pixel-fill is also not the bottleneck on Unity 6 + llvmpipe; per-frame UI Toolkit cost is.

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

Labels

None yet

Development

Successfully merging this pull request may close these issues.

1 participant