IDLE POLL
— ms
Firefox 155.0.1 host completion, idle
GPU complete
— ms
host notified
— ms
GPU work finishes; the completion report does not. Consistent with Firefox Bug 1870699 (WebGPU polled on a timer).
COMPLETION RECEIPT
GPU finished. Firefox waited.
WEBGPU / HOST OBSERVATION
MANUAL · NOTHING RUNS UNTIL YOU PRESS RUN CAPTURE
Same WebGPU workload, local only. Nothing is measured until you press Run capture inside the panel; nothing is uploaded.
NO TELEMETRY
NO AUTO-RUN
CLOSE PANEL TO STOP
COMPLETION TIMELINE
Shared 0–100 ms scale · submit → GPU complete → host notified
020406080100 ms
-
Chrome 152
-
Firefox 155 idle
-
Firefox + fast kicks
-
Firefox + 20 ms kick
Solid bar = elapsed host-completion wait. Marker = when the GPU itself actually finished. The gap between them is the reported joint, not GPU slowness.
PREDICTION RECEIPT
Diagnostic probe: does forcing observation cadence move the reported floor?
predicted
20 ms kick cadence → ~20 ms completion
observed
— ms
kicks / run
The kick is a diagnostic probe of the polling mechanism only — not a fix to run in a real app. It does not make the GPU workload faster.
DISTRIBUTION · RUNS EACH
| run | gpu median | host min | host median | host max |
|---|
GPU column shows median only. Kick-mode captures include empty diagnostic submissions in their raw min/mean — those are not workload timings and are not displayed here.
LIMITS / PROVENANCE
- One Linux host, Intel Iris Xe integrated GPU. No cross-platform or cross-vendor claim.
- Firefox 155.0.1 WebGPU was manually enabled via flag; not default release support.
- Canonical evidence is headless, on the real Iris Xe GPU. Ordinary headed-Firefox parity on that real GPU remains untested (see PROVENANCE.md for a separate, non-canonical software-rendered headed smoke that is not a substitute).
- Consistent with Mozilla Bug 1870699 (polling on a timer). No source-level confirmation was performed — this is black-box timing evidence offered to an already-open, already-triaged issue.
- Not claimed as a first-discovery finding, a product, a performance benchmark, or evidence of demand. No cross-browser prevalence claim.