v0.21.0

[LizardByte-bot]

Added

• (Input) Add support for automatically selecting the emulated controller type based on the physical controller connected to the client
• (Input/Windows) Add support for Applications (context menu) key
• (Input/Windows) Implement touchpad, motion sensors, battery state, and LED control for the emulated DualShock 4 controller
• (Input) Advertise support for new input features to clients
• (Linux/Debian) Added Debian Bookworm package
• (Prep-Commands) Expose connection environment variables
• (Input/Windows) Implement pen and touch support
• (Capture/Windows) Add standalone NVENC encoder
• (Capture) Implement AV1 encoding
• (Network) Implement IPv6 support
• (Capture/Windows) Add option to disable realtime hags
• (Graphics/NVIDIA) Add an option to decrease GPU scheduling priority to workaround HAGS video hang
• (Capture/Linux) Add FFmpeg powerpc64le architecture for self compiling Sunshine
• (Capture/Windows) Add support for capturing rotated displays
• (System Tray) Implement streaming event notifications
• (UI) Add port configuration table
• (Applications) Added option to automatically treat launcher type apps as detached commands
• (Input/Gamepad) Allow the Misc button to work as Guide on emulated Xbox 360 controllers

Changed

• (Input) Reduce latency by implementing input batching
• (Logging) Move input packet debug prints off the control stream thread
• (Input) Refactor gamepad emulation code to use DS4 extended report format
• (Graphics/NVIDIA) Modify and restore NVIDIA control panel settings before and after stream, respectively
• (Graphics/NVIDIA) New config page for NVENC
• (Graphics/Windows) Refactor DX shaders
• (Input/Windows) Use our own keycode mapping to avoid installing the US English keyboard layout

Fixed

• (UI) Fix update notifications
• (Dependencies/Linux) Replace libboost chrono and thread with standard chrono and thread
• (Input) Increase maximum gamepad limit to 16
• (Network) Allow use of multiple ENet channels
• (Network) Consider link-local addresses on LAN
• (Input) Fixed issue where button may sometimes stick on Windows
• (Input) Fix "ControllerNumber not allocated" warning when a gamepad is removed
• (Input) Fix handling of gamepad feedback with multiple clients connected
• (Input) Fix clamping mouse position to aspect ratio adjusted viewport
• (Graphics/AMD) Fix crash during startup on some older AMD GPUs
• (Logging) Fix crash when non-ASCII characters are logged
• (Prep-Commands) Fix resource exhaustion bug which could occur when many prep commands were used
• (Subprocesses) Fix race condition when inserting new processes
• (Logging) Log error if encoder doesn't produce IDR frame on demand
• (Audio) Improve audio capture logic and logging
• (Logging) Fix AMF logging to match configured log level
• (Logging) Log FFmpeg to log file instead of stdout
• (Capture) Reject codecs that are not supported by display device
• (Capture) Add fallbacks for unsupported codec settings
• (Capture) Avoid probing HEVC or AV1 codecs in some cases
• (Caputre) Remove DwmFlush()
• (Capture/Windows) Improvements to capture sleeps for better frame stability
• (Capture/Windows) Adjust capture rate to better match with display
• (Linux/ArchLinux) Fix package version in PKGBUILD and precompiled package
• (UI) Highlight fatal log messages in web ui
• (Commands) Allow stream if prep command fails
• (Capture/Linux) Fix KMS grab VRAM capture with libva 2.20
• (Capture/macOS) Fix video capture backend
• (Misc/Windows) Don't start the session monitor window when launched in command mode
• (Linux/AppImage) Use the linuxdeploy GTK plugin to correctly deploy GTK3 dependencies
• (Input/Windows) Fix reWASD not recognizing emulated DualShock 4 input

Dependencies

• Bump bootstrap from 5.2.3 to 5.3.2
• Bump third-party/moonlight-common-c from c9426a6 to 7a6d12f
• Bump gcc-10 in Ubuntu 20.04 docker image
• Bump furo from 2023.5.20 to 2023.9.10
• Bump sphinx from 7.0.1 to 7.2.6
• Bump cmake from 3.26 to 3.27 in Fedora docker images
• Move third-party/nv-codec-headers from sdk/11.1 branch to sdk/12.0 branch
• Automatic bump ffmpeg
• Bump actions/checkout from 3 to 4
• Bump boost from 1.80 to 1.81 in Macport manifest
• Bump @fortawesome/fontawesome-free from 6.4.0 to 6.4.2

Misc

• (Docs) Force badges to use svg
• (Docs) Add Linux SSH example
• (Docs) Add information about mesa for Linux
• (CI) Free additional space on Docker, Flatpak, and AppImage builds due to internal changes on GitHub runners
• (Docs/Logging/UI) Corrected various typos
• (Docs) Add blurb about Gamescope compatibility
• (Installer/Windows) Use system proxy to download ViGEmBus
• (CI) Ignore third-party directory for clang-format
• (Docs/Linux) Add Plasma-Compatible resolution example
• (Docs) Add Sunshine website available at https://app.lizardbyte.dev/Sunshine
• (Build/Windows) Fix audio code build with new MinGW headers
• (Build/Windows) Fix QoS code build with new MinGW headers
• (CI/Windows) Prevent winget action from creating an update when running in a fork
• (CI/Windows) Change winget job to ubuntu-latest runner
• (CI) Add CodeQL analysis
• (CI/Docker) Fix ArchLinux image caching issue
• (Windows) Manifest improvements
• (CI/macOS) Simplify macport build
• (Docs) Remove deprecated options from readthedocs config
• (CI/Docs) Lint rst files
• (Docs) Update localization information (after consolidating Crowdin projects)
• (Cmake) Split CMakelists into modules
• (Docs) Add Linux Headless/SSH Guide

---

This discussion was created from the release v0.21.0.

[grobuzga]

With this version most of the old NVENC settings are missing - how do I enable lossless quality with this version?

[ns6089]

how do I enable lossless quality with this version

You don't, we never properly supported it to begin with.
FEC coding used for transmitting video frames has hardcoded upper limit on video frame size,
and it can be easily exceeded if we don't limit bitrate to some sane level.

[grobuzga]

Guess I'll revert to 0.20 then. Can't stand color banding with lossy encoding, and couldn't find any other solution to eliminate or at least significantly reduce it.

[stickman89]

I can't even match the same streaming performance as Gamestream with Sunshine 0.21, no matter how I much I reduce the quality... It's just dropping frames a lot, despite no connection warning. I can actually sustain higher bitrates, whilst having a smoother experience with Gamestream, then I can at very low bitrate values on Sunshine. This is happening with or without default settings and with or without HAGs enabled. I never noticed this on prior versions, so I may also revert back. Also, as a separate issue; with vertical vsync set to Fast in NVIDIA control panel, it introduces an overlapping frames issue. In between frames you'll get the previous frame. Loading bars actually look like they stutter backwards, and it's quite noticable in gameplay client-side too. It's not happening on the host, and is erradicated when setting vsync to anything else. This happens both locally, on the same network and over WAN, despite 9ms latency. Gamestream operates fine with the 'fast' vsync setting.

Update: Reverted back to Sunshine 0.20 with the exact same higher bitrate moonlight client settings that worked fine with gamestream, but was stuttery over WAN with Sunshine 0.21 using default configuration... And to my surprise, streaming performance is great again. There are definitely some regressions, at least as a NVIDIA user, whilst on WAN with Sunshine 0.21 over 0.20. The quality even looks better whilst using default 0.20 settings. Happy again.

[stickman89]

@ns6089 - "we never properly supported it to begin with" ... Was something fundamentally wrong/broken with those settings then? Supported or not, I'm all for their return. They just worked as intended, and they offered us configurable options to suit our needs/requirements. And I don't understand what part of them created any demand for support. I bet if you created a poll, most of the power users would favor their return.

Now I'm curious, were there any code related hurdles preventing them from coming over to the current release? Or, just added because we could... Now, we don't want them there as we've switched up our path? Tempted to fork the code to re-implement them now for testing purposes.

[Docmine17]

Thanks!

[iMonZ]

Added

• (Input) Add support for automatically selecting the emulated controller type based on the physical controller connected to the client
• (Input/Windows) Add support for Applications (context menu) key
• (Input/Windows) Implement touchpad, motion sensors, battery state, and LED control for the emulated DualShock 4 controller
• (Input) Advertise support for new input features to clients
• (Linux/Debian) Added Debian Bookworm package
• (Prep-Commands) Expose connection environment variables
• (Input/Windows) Implement pen and touch support
• (Capture/Windows) Add standalone NVENC encoder
• (Capture) Implement AV1 encoding
• (Network) Implement IPv6 support
• (Capture/Windows) Add option to disable realtime hags
• (Graphics/NVIDIA) Add an option to decrease GPU scheduling priority to workaround HAGS video hang
• (Capture/Linux) Add FFmpeg powerpc64le architecture for self compiling Sunshine
• (Capture/Windows) Add support for capturing rotated displays
• (System Tray) Implement streaming event notifications
• (UI) Add port configuration table
• (Applications) Added option to automatically treat launcher type apps as detached commands
• (Input/Gamepad) Allow the Misc button to work as Guide on emulated Xbox 360 controllers

Changed

• (Input) Reduce latency by implementing input batching
• (Logging) Move input packet debug prints off the control stream thread
• (Input) Refactor gamepad emulation code to use DS4 extended report format
• (Graphics/NVIDIA) Modify and restore NVIDIA control panel settings before and after stream, respectively
• (Graphics/NVIDIA) New config page for NVENC
• (Graphics/Windows) Refactor DX shaders
• (Input/Windows) Use our own keycode mapping to avoid installing the US English keyboard layout

Fixed

• (UI) Fix update notifications
• (Dependencies/Linux) Replace libboost chrono and thread with standard chrono and thread
• (Input) Increase maximum gamepad limit to 16
• (Network) Allow use of multiple ENet channels
• (Network) Consider link-local addresses on LAN
• (Input) Fixed issue where button may sometimes stick on Windows
• (Input) Fix "ControllerNumber not allocated" warning when a gamepad is removed
• (Input) Fix handling of gamepad feedback with multiple clients connected
• (Input) Fix clamping mouse position to aspect ratio adjusted viewport
• (Graphics/AMD) Fix crash during startup on some older AMD GPUs
• (Logging) Fix crash when non-ASCII characters are logged
• (Prep-Commands) Fix resource exhaustion bug which could occur when many prep commands were used
• (Subprocesses) Fix race condition when inserting new processes
• (Logging) Log error if encoder doesn't produce IDR frame on demand
• (Audio) Improve audio capture logic and logging
• (Logging) Fix AMF logging to match configured log level
• (Logging) Log FFmpeg to log file instead of stdout
• (Capture) Reject codecs that are not supported by display device
• (Capture) Add fallbacks for unsupported codec settings
• (Capture) Avoid probing HEVC or AV1 codecs in some cases
• (Caputre) Remove DwmFlush()
• (Capture/Windows) Improvements to capture sleeps for better frame stability
• (Capture/Windows) Adjust capture rate to better match with display
• (Linux/ArchLinux) Fix package version in PKGBUILD and precompiled package
• (UI) Highlight fatal log messages in web ui
• (Commands) Allow stream if prep command fails
• (Capture/Linux) Fix KMS grab VRAM capture with libva 2.20
• (Capture/macOS) Fix video capture backend
• (Misc/Windows) Don't start the session monitor window when launched in command mode
• (Linux/AppImage) Use the linuxdeploy GTK plugin to correctly deploy GTK3 dependencies
• (Input/Windows) Fix reWASD not recognizing emulated DualShock 4 input

Dependencies

• Bump bootstrap from 5.2.3 to 5.3.2
• Bump third-party/moonlight-common-c from c9426a6 to 7a6d12f
• Bump gcc-10 in Ubuntu 20.04 docker image
• Bump furo from 2023.5.20 to 2023.9.10
• Bump sphinx from 7.0.1 to 7.2.6
• Bump cmake from 3.26 to 3.27 in Fedora docker images
• Move third-party/nv-codec-headers from sdk/11.1 branch to sdk/12.0 branch
• Automatic bump ffmpeg
• Bump actions/checkout from 3 to 4
• Bump boost from 1.80 to 1.81 in Macport manifest
• Bump @fortawesome/fontawesome-free from 6.4.0 to 6.4.2

Misc

• (Docs) Force badges to use svg
• (Docs) Add Linux SSH example
• (Docs) Add information about mesa for Linux
• (CI) Free additional space on Docker, Flatpak, and AppImage builds due to internal changes on GitHub runners
• (Docs/Logging/UI) Corrected various typos
• (Docs) Add blurb about Gamescope compatibility
• (Installer/Windows) Use system proxy to download ViGEmBus
• (CI) Ignore third-party directory for clang-format
• (Docs/Linux) Add Plasma-Compatible resolution example
• (Docs) Add Sunshine website available at https://app.lizardbyte.dev/Sunshine
• (Build/Windows) Fix audio code build with new MinGW headers
• (Build/Windows) Fix QoS code build with new MinGW headers
• (CI/Windows) Prevent winget action from creating an update when running in a fork
• (CI/Windows) Change winget job to ubuntu-latest runner
• (CI) Add CodeQL analysis
• (CI/Docker) Fix ArchLinux image caching issue
• (Windows) Manifest improvements
• (CI/macOS) Simplify macport build
• (Docs) Remove deprecated options from readthedocs config
• (CI/Docs) Lint rst files
• (Docs) Update localization information (after consolidating Crowdin projects)
• (Cmake) Split CMakelists into modules
• (Docs) Add Linux Headless/SSH Guide

This discussion was created from the release v0.21.0.

Well... thanks but the macOS Support is still completely broken :(
dyld[14335]: Library not loaded: /usr/local/opt/boost/lib/libboost_locale-mt.dylib
Referenced from: /usr/local/bin/sunshine-0.21.0
Reason: tried: '/usr/local/opt/boost/lib/libboost_locale-mt.dylib' (no such file), '/System/Volumes/Preboot/Cryptexes/OS/usr/local/opt/boost/lib/libboost_locale-mt.dylib' (no such file), '/usr/local/opt/boost/lib/libboost_locale-mt.dylib' (no such file), '/usr/local/lib/libboost_locale-mt.dylib' (no such file), '/usr/lib/libboost_locale-mt.dylib' (no such file, not in dyld cache)
fish: Job 1, 'sunshine' terminated by signal SIGABRT (Abort)

[iMonZ]

Right. I just installed the .dmg
Should I try again the macports version? A week ago I tried the latest dev. After 1 hour of compiling it was still broken.

[iMonZ]

I realize now that if Sunshine would switch from Macport to homebrew there would be no need for an Apple CI anymore and you could save money.

[ReenigneArcher]

How? It doesn't cost anything right now.

Also, we would still build the homebrew version to verify it builds.

[ReenigneArcher]

If you use dmg, you need to install all the dependencies yourself... that's why it's not supported right now, aka "experimental".

[iMonZ]

Gotcha!
Okay compiled all the deps.
Now I sadly stuck at this error:
https://hastebin.com/share/nodipurosa.rust

[FallingSnow]

How come constqp rate control was removed? It gave higher highs and lower lows when compared to CBR/VBR and looked great.

[FallingSnow]

For the time being I've reverted to using v0.20.

[stickman89]

Must agree. 0.21 is definitely slower imo. NVENC Tune and CBR/VBR options in 0.20 does indeed improve quality.

[utkvishwas]

Thank you so much for the pen support.
I was asking for it, and thank you so much for listening.

[Lynxen]

Is there any configuration needed to enable pen support? My host PC is a desktop without a touch / pen monitor, and I'm trying to connect with a Surface Pro as the client and use the pen on the host PC, but it's acting as if it's a mouse.

[Kenishi99]

Same issue here, Windows client (with a Wacom tablet / pen) connecting to a Windows host. It is not working and detected as a mouse / touchscreen input instead of a pen.

[dvd6]

Spen on my samsung tablet is working well, including pressure sensitivity (but not the button). The touchscreen is still working as a mouse (or touchpad).

[Sasoon]

Hey @dvd6 how'd you get pressure sensitivity working? I'm using a tab s8+ connecting to a windows machine, and pressure sensitivity flat out won't work in clip studio or blender. Been fiddling with settings but can't crack it :(

[dvd6]

I did nothing except updating the sunshine and moonlight.
My devices: tab s7+, windows11 pc
Software I tried: photoshop, onenote

[luan-digital]

@Sasoon I had to change a setting for it to work in Clip Studio Paint, but now it's working perfectly with my Tab S6 Lite and my Windows 11 PC. To get it to work, go into Preferences > Tablet > Using tablet service and select Tablet PC

[Wiggoggs]

Is there any plan to extend the motion controls support for PS4/PS5 controllers to linux hosts?

[besho1994]

hi. does anybody know how can i stream on iphone 14 pro in full screen? i get black vertical bars i tried to change resolution to match iphone but i doesnt work. any ideas?

[dkr88]

For me, 0.21 has introduced micro-stuttering or some kind of frame pacing problems. I've been trying nightlies for the last month or so, and noticed the same problem in them that I'm seen in this release. Going back to 0.20 and it's perfectly smooth at 60fps. All default Sunshine settings, 3080 + 5600X + Windows 10 w/HAGS. Problem is apparent on Android and Windows clients, and I noticed it before and after the Moonlight updates.

[ReenigneArcher]

Probably would have been better to report it when you noticed it months ago.

[dkr88]

Probably, yeah... but it's so hard to quantify issues like this. I assumed if it was a wider issue it would have already been reported/being worked on. I really have no idea how to go about troubleshooting it - just wanted to see if anyone had a similar experience.

[nicanecreed]

I am getting the same issue on 0.21. It seems to be between 20-30 fps with the stuttering. Back to 0.20 and works fine with solid 60 fps. It's an older i7-4960X + GTX 1060 + Windows 11.

[stickman89]

Confirmed the same thing. Vsync set to 'Fast' in NVIDIA Control Panel introduces a strange visual oddity where the previous frame is shown in between frames, worsening the stutter - frames peddle backwards by one. Setting Vsync to anything else does get rid of that odd behavior, but it doesn't fix the stuttery performance in general with 0.21, and what feels like frame drop. No issues on 0.20.

[jamesdavis36]

Thanks for the work on this update, I've noticed particularly an increase in usability in areas where I was getting just not quite enough bandwidth from available wifi prior.

[Senorsen]

For me, 0.21 has introduced micro-stuttering or some kind of frame pacing problems. I've been trying nightlies for the last month or so, and noticed the same problem in them that I'm seen in this release. Going back to 0.20 and it's perfectly smooth at 60fps. All default Sunshine settings, 3080 + 5600X + Windows 10 w/HAGS. Problem is apparent on Android and Windows clients, and I noticed it before and after the Moonlight updates.

Me too, upgrading to 0.21, there is some stuttering and flickering (streaming in 2.5Gbps LAN, sunshine host is using 4070ti, client is the latest moonlight 5.0.1 on a M1 Macbook Air).
After downgrading to 0.20, the stuttering and flickering is gone.
Really want to use the 0.21 because the new notification feature...
Maybe I'll try to export some logs and report through the github issues.

[dkr88]

Based on some other posts, I was able to partially narrow the problem down to v-sync weirdness. In apps where I was experiencing the problem, I forced v-sync on in NVCP (despite it already being turned on in the app) and it seems to have pretty much fixed my stuttering. Huge ymmv obviously, but it's worth a shot.

EDIT: Making the assumption that your v-sync refresh rate is the same as your streaming FPS.

[Mothamoz]

It's probably because the NVENC preset got changed from P4 (default 0.20) to P1 (default 0.21).

Change it back to P4 and you'll hopefully be good to go, the only difference 1-2ms extra encoding latency, unnoticeable.

[Senorsen]

@Mothamoz Wow, good to know, I'll give it a try. But why it happens on P1 (fastest)? Because it should be "faster" than P4 (medium)...

[Mothamoz]

@Mothamoz Wow, good to know, I'll give it a try. But why it happens on P1 (fastest)? Because it should be "faster" than P4 (medium)...

Encoding time is 2 ms lower on my end on P1, not noticeable. But its heavy for alot of decoders combined with high bitrate

[Senorsen]

Hi @Mothamoz, I’ve tried your solution a few times and noticed that the issue is partially gone, but there are still a very few stutterings sometimes.

And I found a very interesting thing - when I move the mouse very “fast” - keep shaking it, both the screen and the audio gets jitters. (find this just after i put the mouse on my “furry” sofa, and the pointer keeps shaking by itself, very quickly.) i’m not very sure about this but you can give a try too..

ps. My setup:
Sunshine: 0.21, Windows 11 23H1, RTX 4070Ti, i5-8500 (quite an old CPU, but the single core usage is not high, i assume it is not the bottleneck. also it works perfectly in the old version 0.20/0.19); Realtek 8125BG 2.5Gbps wired connection
Moonlight: 5.0.1, Macbook Air M1, Wifi 6 connected to an AX86U Pro router within 3 meters, both in the living room. Local speedtest (librespeed) gives ~800mbps 9ms result.
Stream settings: 4k 60fps, HEVC/h265, 100-120Mbps (seems no difference), full screen

[sergeymetallic]

I have windows PC and Nvidia Shield 2017 as a client with latest moonlight. After the upgrade to 0.21 none of my gamepads(sony dualsense and sony dualshock) are recognized. Rolling back to 0.20 solved the issue.

[hifihuang]

Huang@iMac ports % sudo port install sunshine
Portfile changed since last build; discarding previous state.
--->  Computing dependencies for Sunshine
--->  Fetching distfiles for Sunshine
--->  Verifying checksums for Sunshine
--->  Extracting Sunshine
--->  Configuring Sunshine
--->  Building Sunshine                                  
Error: Failed to build Sunshine: command execution failed
Error: See /opt/local/var/macports/logs/_Users_Huang_ports_multimedia_sunshine/Sunshine/main.log for details.
Error: Follow https://guide.macports.org/#project.tickets if you believe there
is a bug.
Error: Processing of port sunshine failed

[ReenigneArcher]

Nothing in here to help diagnose.

[Sasoon]

Anyone having issues with pressure sensitivity on android? Trying to use s-pen on a tab s8+ to remote into Win11 pc.. no pressure sensitivty in Clip Studio or Blender :(