0.3 Async Pre Release

[derflacco]

Hello there!
Before making a new release on Git, it might be better to post a pre-release here first.

Changelog:

• Fixed ASYNC + HDR crash

• Restored performance on Fast GL with HDR off (mostly useful on mtk, way less on Snapdragon)

• Dequeue us is now fixed instead of adaptive (you can still experiment through the app)

• CPU Boost is now fully configurable: you can choose which cluster to boost, and more

• Various FSR fixes

• The app now hints the Surface frame rate using the selected stream FPS, instead of the panel refresh rate (let me know how it works if you stream at es: 59.94hz)

• The ui needs to be refined a bit.

Let me know how it goes!

LINK

[Zayr64]

Afe these the new cpu boost settings?

Which settings would be 0.2 cpu boost on?

[derflacco]

Medium, all cores, periodic bursts disabled

[sve989]

Hello, any suggestions on how to stop microslutter on g99 cpu, the only fix I found is app cpu load using 3 buttons and loading 3 cores, but what will happen with the cpu if working like this every day 12+hours?

From your settings, I've tried Fsr, lowering the resolution to 720p, bitrate to 20, low latency, GPU raw, Immediate frame delivery, Sync options nothing help so far.

[derflacco]

Hello, I have a device with that chip. The best settings you can use are:

1440x900 resolution, 120fps, maximum 35 Mbps bitrate (25 is better), CPU boost set to medium, all cores, prefer lowest latency pacing. Unfortunately, that chip is limited.

[sve989]

Hello, I have a question about FSR. When I enable it with the option for upscaling: None, all stutters are gone, but the image is blurry. Is there any way to fix it or make it better? Sorry for my English.

[derflacco]

@sve989 0.3 or 0.2?

[sve989]

0.3, just test it on 0.2, same blurry image.

[VirtualGIT20]

Hi @derflacco !
Sorry, I've missed the notification of the new pre-release.
Will do some testing later after work!

[VirtualGIT20]

Hi! @derflacco

I was able to do some testing with the new build.

Regarding the fixes:

• ASYNC + HDR works fine on my side.

• The fixed dequeue timeout doesn’t seem to make any noticeable difference in streaming performance and smoothness compared to before, but I can confirm that works correctly.

*On a related note, I noticed that when resuming a stream using Smoothest, the decoding latency sometimes starts higher than usual and stays there for a short while before returning to normal. I’ve also observed the same behavior in vanilla Artemis. It’s not a major issue, just something worth mentioning.

• I still need to test more thoroughly, but from what I’ve seen so far, CPU Boost works very well, even on the Shield.

Using the preset from 0.2 (Medium, all cores), average delay at 4K60 went from ~3.5 ms down to ~1.5 ms.
I still need to test it properly on a less powerful device like the A52s, but the results so far are very impressive.

• I haven’t had the chance to properly test FSR yet.

Overall, although I still need to do more testing, this build feels quite solid to me. The only thing I would suggest is some minor UI refinement in the CPU Boost section.

On a side note, I’m currently trying to troubleshoot some encoding spikes on the server side (average ~6 ms, occasional spikes up to ~14 ms with small stutters even though everything seems under control host side).
Would it be possible, maybe in the future, to include encoding statistics in Light Mode as well? That would make that mode more complete, yet less invasive than standard one.

As always, thanks a lot for your work!

[derflacco]

Hello!

• The fixed dequeue timeout doesn’t seem to make any noticeable difference in streaming performance and smoothness compared to before, but I can confirm that works correctly.

Do you mean they’re working like the manually fixed dequeue? I made that fix/revert to address the creeping decoding latency that was reported in 0.2 when the dequeue wasn’t manually set.

Using the preset from 0.2 (Medium, all cores), average delay at 4K60 went from ~3.5 ms down to ~1.5 ms.
I still need to test it properly on a less powerful device like the A52s, but the results so far are very impressive.

Cool! It seems it’s working correctly for the shield now as well!

Would it be possible, maybe in the future, to include encoding statistics in Light Mode as well? That would make that mode more complete, yet less invasive than standard one.

I can try that it shouldn’t be difficult, since the hooks are already in place.

Overall, although I still need to do more testing, this build feels quite solid to me. The only thing I would suggest is some minor UI refinement in the CPU Boost section.

Yeah, I have to do that, but I wanted to know if the new options were useful.

Thank as always for the feedback!

[VirtualGIT20]

• The fixed dequeue timeout doesn’t seem to make any noticeable difference in streaming performance and smoothness compared to before, but I can confirm that works correctly.

Do you mean they’re working like the manually fixed dequeue? I made that fix/revert to address the creeping decoding latency that was reported in 0.2 when the dequeue wasn’t manually set.

Sorry, my wording wasn’t very clear. 😅

What I meant is that the fixed dequeue behavior now works correctly and I’m no longer seeing the creeping decoding latency issue. At the same time, I haven’t noticed any measurable loss in performance or smoothness compared to before.

[arushe0805]

Any ETA for a release for this? Overall I love this version Moonlight and the only bug I ran into was the ASYNC + HDR crash on .2 release. I originally didn't think the Snapdragon 8 Gen 2 could get a better decode time then I saw on the original fork but on yours I get around 1 to 2.5ms and looks the most smoothest over the other forks.

[derflacco]

Hello. As soon I will have more time spare, I think during the easter holidays :)

[realdavixx-tech]

Hello there!

Before making a new release on Git, it might be better to post a pre-release here first.

Changelog:

• Fixed ASYNC + HDR crash

• Restored performance on Fast GL with HDR off (mostly useful on mtk, way less on Snapdragon)

• Dequeue us is now fixed instead of adaptive (you can still experiment through the app)

• CPU Boost is now fully configurable: you can choose which cluster to boost, and more

• Various FSR fixes

• The app now hints the Surface frame rate using the selected stream FPS, instead of the panel refresh rate (let me know how it works if you stream at es: 59.94hz)

• The ui needs to be refined a bit.

Let me know how it goes!

LINK

Hey any chance you could update the link? It says it's not available for transfer. Or maybe someone else could upload it? That has it would appreciate! @derflacco @VirtualGIT20

[derflacco]

Hello there!

0.3 Async Pre Release 2

• merged new updates from moonlight common c
• switched to my common c fork, with some experimental changes
• attempt to fix the tonemap shift \ blur when enabling fsr \ direct present ( the blur should be gone, the tonemap shift it's still there, to be honest i like it). @sve989
• Server stats can be enabled on the lite overlay too @VirtualGIT20
• some minor Ui fixes
• hidden some non configured Cpu boost related stuff
• -O3 for modules, resource shrinking enabled for the perf builds

LINK

[VirtualGIT20]

Hello @derflacco and welcome back!

I'm testing the new build, I can confirm that new server stats works as intended.
I noticed that maybe there could be a repetition of the frame pacing option selected (LA after network parameters)

• attempt to fix the tonemap shift \ blur when enabling fsr \ direct present ( the blur should be gone, the tonemap shift it's still there, to be honest i like it).

Will try that on my A52s

Thanks again for your work!!

• merged new updates from moonlight common c
• switched to my common c fork, with some experimental changes

Edit:
Also, can I ask what changes should we expect with this fork?

[derflacco]

Hello!

I'm testing the new build, I can confirm that new server stats works as intended.

if you enable lite + you will also see min \ max latency

Also, can I ask what changes should we expect with this fork?

Nothing special honestly, maybe less input latency

I noticed that maybe there could be a repetition of the frame pacing option selected (LA after network parameters)

will fix that, thanks!