Package Details: sunshine 2026.516.143833-1

Git Clone URL: https://aur.archlinux.org/sunshine.git (read-only, click to copy)
Package Base: sunshine
Description: Self-hosted game stream host for Moonlight
Upstream URL: https://app.lizardbyte.dev/Sunshine
Licenses: GPL-3.0-only
Submitter: hadogenes
Maintainer: LizardByte
Last Packager: LizardByte
Votes: 81
Popularity: 3.58
First Submitted: 2021-04-26 11:33 (UTC)
Last Updated: 2026-05-16 23:04 (UTC)

Pinned Comments

LizardByte commented on 2025-09-09 01:59 (UTC)

This package is automatically synced from https://github.com/LizardByte/pacman-repo

GitHub provides the ability for anyone to contribute fixes/improvements instead of only one maintainer. Please submit fixes, issues, etc. on GitHub.

Temporary build fixes for stable version should be submitted to https://github.com/LizardByte/pacman-repo

Permanent fixes that should persist with future versions should also be submitted to https://github.com/LizardByte/Sunshine

Latest Comments

1 2 3 4 5 6 .. 22 Next › Last »

torquetom commented on 2026-06-03 17:59 (UTC)

@Neurofibromin When updating/changing CUDA, you need to reboot to reload env vars, otherwise NVCC_CCBIN and CUDA_PATH isn't set properly and that causes it to use the wrong gcc version. It should use gcc15 and only gcc15.

torquetom commented on 2026-06-03 17:27 (UTC) (edited on 2026-06-03 18:00 (UTC) by torquetom)

This is incompatible with anything older than nvidia-580xx. Lower driver versions use CUDA 12.9. CUDA 12.9 is incompatible with gcc15. gcc15 is hardcoded.

The only way this can compile is against CUDA 13+. With the mismatching CUDA on the driver (12.x) and at compile-time (13+), it crashes at trying to create encoder, something about a "no kernel image is available for execution on the device".

A workaround without editing PKGBUILD is simply deleting the CUDA package and building without it. It works fine for me. AFAIK (plus I asked) it's only used for NvFBC, which doesn't even work on anything lower than 585.

Neurofibromin commented on 2026-05-17 13:57 (UTC)

fails to build after update to gcc 16

urbenlegend commented on 2025-09-21 06:48 (UTC) (edited on 2025-09-21 06:48 (UTC) by urbenlegend)

@Xarth

It does seem like this package puts udev rules already in /usr/lib/udev/rules.d/60-sunshine.rules, so you most likely don't have to add your own rules in /etc/udev/rules.d to get input working again.

I also ran into permissions issues today and all I had to do was add myself to the input group to get things working again.

Nocifer commented on 2025-09-20 19:21 (UTC) (edited on 2025-09-20 19:23 (UTC) by Nocifer)

So let me see if I get this correctly:

According to the previous logic, if the if [[ "${_use_cuda::1}" == "t" ]] block read as true (which would mean either that CUDA has been detected as installed on the system or that the user has manually set _use_cuda to true) then CUDA would be added both to the makedepends array and to the optdepends array, so the build would pull it in as a make dependency. This would work perfectly fine even on an AMD system, even if CUDA would obviously be a no go during runtime.

According to the new logic in the latest PKGBUILD, if the if [[ "${_use_cuda::1}" == "t" ]] block reads as true then CUDA is now only added to the optdepends array but not to the makedepends array. This means that if I set _use_cuda to true, the build will fail because CUDA is no longer getting pulled as a make dependency but SUNSHINE_ENABLE_CUDA and CUDA_FAIL_ON_MISSING are still set to ON - in other words, building with CUDA will only work if and only if I have CUDA preinstalled on my system.

Why was this change implemented in such a way that it breaks the previously working procedure?

I'm asking out of curiosity; I'm on an AMD system so this doesn't affect me in the slightest.

mdhen commented on 2025-09-19 21:12 (UTC) (edited on 2025-09-20 00:35 (UTC) by mdhen)

CUDA is now set as an optional dependency - issue on AMD systems fixed.

HatsuneCthulhu commented on 2025-09-19 02:58 (UTC) (edited on 2025-09-19 03:10 (UTC) by HatsuneCthulhu)

This is now broken on AMD systems since LizardByte took over as maintainer (as in rev 4 onwards), it's requiring cuda as a build dependency, which obviously AMD systems will not have.

1 2 3 4 5 6 .. 22 Next › Last »