Moonlight M2 Mac Mini High Decoding Latency Tied to Mouse Movement

[Pointyguide]

Moonlight streaming on a M2 Mac mini as well as a M1 MacBook Pro. The decoding latency is between 2-3ms but if I move the mouse around very fast the decoding latency drops to 1.2ms. The latency is tied to how fast the mouse is moved around the screen. Also on windows moonlight decoding latency is in the sub 1 ms range and feels very responsive but on M1 and M2 the decoding latency of 2-3ms is very high and does not feel responsive at all. Is there any way to bring the decoding latency down to Windows Levels?

Moonlight Settings:
1440P 144HZ

Host PC:
7800x3d
RTX 3060 Ti

[Pointyguide]

After further investigation. I have noticed that without high mouse movement Moonlight runs mainly on the E cores. If moving the mouse wildly the load shifts to P cores and Decoding latency goes down. There must be a way to force moonlight to always request execution on the P cores?

[cgutman]

I doubt the drop from 2-3ms to 1ms is actually responsible for what you're feeling. 1ms of latency difference is not something that's going to be perceptible. I think you're actually experiencing latency caused by the AVSampleBufferDisplayLayer renderer.

Please try this nightly build which includes the new Metal-based renderer: https://ci.appveyor.com/project/cgutman/moonlight-qt/builds/49795697/job/q2f3adjbrm2uhjap/artifacts

[Pointyguide]

I think it might feel a little better but I do not see any major changes, the decoding latency is still tied to mouse movements. Faster I move the mouse around the more it drops close to 1 ms. lowest I see is 1.31 ms. for some reason Macs just don't feel as native as a windows client I am testing with a 13700 and a RTX 4060. My windows client have sub millisecond decoding latencies and mouse feels more fluid and not as heavy.

Also 4k streaming the decoding latency hovers around 2-3ms as well. I believe there is a small improvement with this build but not immensely huge like I expected.

Also moonlight primarily runs on the Efficiency cores instead of Performance cores, unless there is high mouse movements.