The design chapter on demos, section “The graphics card”, before and after the correction. The code change of the pull request is unchanged.
Before
The graphics card
On a laptop with two graphics chips the recording browser paints on the stronger one (on a Windows laptop with an NVIDIA card it would otherwise take the integrated Intel chip): the screencast hands over more frames while the machine is busy (28–44 instead of 22–27 a second). Encoding stays with libx264 on the processor: NVIDIA's encoder was no faster, since reading the JPEG frames takes most of the time, and at the same quality its files were twice the size (2026-10-28).
Decisions
(nothing about the graphics chip or the encoder)
After
The graphics card
The recording browser asks for the stronger graphics chip. On a Windows or macOS laptop with two (a Windows laptop with an NVIDIA card, say, where it would otherwise take the integrated Intel chip) it paints on that one while it records; Linux ignores the request, and a machine with one graphics chip records as before.
Decisions
- The recording browser asks for the stronger graphics chip (2026-10-28): on a busy Windows laptop with an NVIDIA card, the screencast of a test page handed over 28–44 frames a second instead of 22–27 on the Intel chip. The cost: the stronger chip draws power while a demo records.
- Encoding stays with libx264 on the processor (2026-10-28). Rejected: NVIDIA's encoder, no faster (reading the JPEG frames takes most of the time) and, at the same quality, files twice the size.