This is a Testing Release
If you want a stabler version use this:
https://github.com/derflacco/moonlight-android/releases/tag/Experimental_6.2
Based on Exp_6.2.
Changelog
Highlights
-
New jitter smoothing pipeline (EWMA → IJH → Quantile Hybrid) plus multiple timing/stability fixes.
-
Added AdaptX, a fully auto-adaptive pacing profile with deadline-based early drops and dedicated LFR rules.
-
GpuRaw now uses Warp Factor (Test)
-
Integrated thermal-aware CpuWarmUp to keep DVFS responsive without heavy drain or runaway throttling.
-
Critical FSR/EGL fixes and improved HDR + perf overlay coexistence.
-
Mouse emulation fixed and made smoother, including when Snappy Input is enabled.
Added
AdaptX frame pacing profile
-
Vsync-aligned pacing with dynamic guard and fully adaptive behavior for host/display Hz mismatches.
-
Early deadline drop to avoid queue build-up when the next vsync window is already missed.
-
lateStreak tracking to react better to persistent lateness.
CpuWarmUp helper
-
Introduced cleanly and wired into the renderer.
-
Thermal-aware mode with self-throttling when the device is hot.
-
Cluster spreading to distribute load across big/little cores.
Changed
Renderer / Pacing
-
Render loop fixes and present stabilization.
-
SPSC ring for output indices to reduce locks/overhead in the hot path.
-
Sanity checks for timebase and frame deltas to prevent drift.
-
Recovery (drop/catch-up) enabled for non-Balanced profiles too: GPU_RAW / Max Smoothness / Cap FPS.
-
Improved refresh-rate management and pacing refinements.
Jitter smoothing
-
Switched to IJH for more responsive jitter tracking.
-
Upgraded to Quantile Hybrid to reduce spikes and variability.
-
Fixed PLL phase + render-time feedback to improve vsync alignment.
LFR
-
Optimized to reduce CPU burn without hurting latency.
-
Targeted bypass/skip logic for low-latency scenarios and for AdaptX.
Fixed
FSR
- Fixed EGL_BAD_ACCESS in the FSR render path.
HDR / Overlay
- HDR window mode now updates before the perf overlay to avoid conflicts.
CpuWarmUp / ADPF
-
Added ProGuard keeps for CpuWarmUp.
-
Fixed interaction with ADPF/PerfHint.
-
Corrected behavior on app-switch.
Render safety
-
Clamp negative frame-age to 0 to avoid wrong gating decisions.
-
Pre-allocated BufferInfo to eliminate frequent hot-path allocations.
Input
-
Fixed mouse emulation breaking with Snappy Input ON.
-
Smoother mouse movement (less “low-polling-rate” feel).
-
More consistent, softer scrolling.
-
LT = fast-scroll while in mouse emulation mode.
A quick note about CpuWarmUp:
The basic idea is to place a small load on the CPU to keep its clock speeds higher, which can potentially reduce decoding latency.
It's essentially an hack, a quick and crude way to achieve a better stream on hardware that isn't low-latency capable (like my G99 tablet). Since rooting is less common these days, this can be a helpful workaround.
However, if you have root access, do not use this. A much more effective method is to set your CPU governor to "Performance" and configure the TopApp/Foreground cpuset to use only the big cores.
I can't stress this enough: That is the most effective way to drastically reduce decoding latency on hardware without native low-latency decoding support.
For newer hardware, like the S20 FE or Poco X6 Pro, you should just skip this feature. It won't provide any benefit and will only result in higher battery drain and increased thermal throttling.
A final warning: Enable this at your own risk (default disabled). It will increase power consumption and generate more heat.
As always,
⚠️ Heads up: this is experimental. I’ve only tested on my hardware, so things may break on yours. Use at your own risk.
Found a bug? Tell me and I’ll try to fix it.
Enjoy!