You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Two related cleanups around the helper's GPU-sharing model.
Drop the GpuLimitPercent setting (5..75, default 25). The knob name read
as "max % of GPU the helper uses," but the implementation actually
controlled when to yield based on overall GPU-busy %. Everyone who saw
"25" reasonably worried the helper would constantly pin a quarter of
their GPU, which it never did. The dedicated encoder block (NVENC,
AMD VCN) runs in parallel with the game's render pipeline, so a low
yield threshold cost helper throughput for no measurable game-FPS
benefit. Helper now uses a hardcoded 95% back-off threshold internally
and exposes no equivalent knob.
Refuse Intel QSV and Windows MediaFoundation at encoder selection.
Both run on integrated graphics on the vast majority of installs,
where the encoder shares silicon with the render pipeline and
accepting a lease costs real frames in-headset. Helper opts out of
co-watching entirely when only integrated encoders are available;
the rest of WKVRCProxy (relay, watchdog, mesh client) continues to
run normally. A helper_encoder_refused log line lists the rejected
encoders so operators see why their iGPU machine isn't sharing.
Wire field gpu_limit_percent retained on HelperStatusFrame for
backward compat with older servers that still log it; always emitted
as 0 from this build (no user override).
0 commit comments