Go to file
2022-12-20 15:32:42 +01:00
external Remove cuda.h 2022-09-29 20:07:54 +02:00
include follow focused 2022-12-20 15:32:42 +01:00
scripts Move scripts to script dir, add youtube-hls-stream script, fix twitch stream script 2022-09-19 16:19:41 +02:00
src follow focused 2022-12-20 15:32:42 +01:00
.gitignore wip intel 2022-11-30 19:36:33 +01:00
build.sh wip intel 2022-11-30 19:36:33 +01:00
install_ubuntu.sh Refactor nvfbc into abstract capture api 2022-10-14 01:22:43 +02:00
install.sh Stupid mistake. Old /usr/local/bin/gpu-screen-recorder overwrites /usr/bin/gpu-screen-recorder path. Make installer install to both paths to fix past mistakes 2022-10-04 01:37:08 +02:00
LICENSE Update LICENSE 2020-03-30 21:40:41 +02:00
project.conf follow focused 2022-12-20 15:32:42 +01:00
README.md follow focused 2022-12-20 15:32:42 +01:00
TODO follow focused 2022-12-20 15:32:42 +01:00

GPU Screen Recorder

This is a screen recorder that has minimal impact on system performance by recording a window using the GPU only, similar to shadowplay on windows. This is the fastest screen recording tool for Linux.

This screen recorder can be used for recording your desktop offline, for live streaming and for nvidia shadowplay-like instant replay, where only the last few seconds are saved.

Note

This software works only on x11.
If you are using a variable refresh rate monitor, then choose to record "screen-direct". This will allow variable refresh rate to work when recording fullscreen applications. Note that some applications such as mpv will not work in fullscreen mode. A fix is being developed for this.
For screen capture to work with PRIME (laptops with a nvidia gpu), you must set the primary GPU to use your dedicated nvidia graphics card. You can do this by selecting "NVIDIA (Performance Mode) in nvidia settings:

and then rebooting your laptop.

TEMPORARY ISSUE

screen-direct capture has been temporary disabled as it causes issues with stuttering. This might be a nvfbc bug.

Performance

When recording Legend of Zelda Breath of the Wild at 4k, fps drops from 30 to 7 when using OBS Studio + nvenc, however when using this screen recorder the fps remains at 30.
When recording GTA V at 4k on highest settings, fps drops from 60 to 23 when using obs-nvfbc + nvenc, however when using this screen recorder the fps only drops to 55. The quality is also much better when using gpu-screen-recorder.
It is recommended to save the video to a SSD because of the large file size, which a slow HDD might not be fast enough to handle.
Using NvFBC (recording the monitor/screen) is not faster than not using NvFBC (recording a single window) with gpu screen recorder, in fact it might be a tiny bit slower.

Installation

If you are running an Arch Linux based distro, then you can find gpu screen recorder on aur under the name gpu-screen-recorder-git (yay -S gpu-screen-recorder-git).
If you are running an Ubuntu based distro then run install_ubuntu.sh as root: sudo ./install_ubuntu.sh. You also need to install the libnvidia-compute version that fits your nvidia driver to install libcuda.so to run gpu-screen-recorder and libnvidia-fbc.so.1 when using nvfbc. But it's recommended that you use the flatpak version of gpu-screen-recorder if you use an older version of ubuntu as the ffmpeg version will be old and wont support the best quality options.
If you are running another distro then you can run install.sh as root: sudo ./install.sh, but you need to manually install the dependencies, as described below.
You can also install gpu screen recorder (the gtk gui version) from flathub.

Dependencies

libglvnd (which provides libgl and libegl), (mesa if you are using an amd or intel gpu), ffmpeg, libx11, libxcomposite, libpulse. You need to additionally have libcuda.so installed when you run gpu-screen-recorder and libnvidia-fbc.so.1 when using nvfbc.
Recording monitors requires a gpu with NvFBC support (note: this is not required when recording a single window!). Normally only tesla and quadro gpus support this, but by using nvidia-patch or nvlax you can do this on all gpus that support nvenc as well (gpus as old as the nvidia 600 series), provided you are not using outdated gpu drivers.

How to use

Run scripts/interactive.sh or run gpu-screen-recorder directly, for example: gpu-screen-recorder -w $(xdotool selectwindow) -c mp4 -f 60 -a "$(pactl get-default-sink).monitor" -o test_video.mp4
Then stop the screen recorder with Ctrl+C, which will also save the recording.
Send signal SIGUSR1 (killall -SIGUSR1 gpu-screen-recorder) to gpu-screen-recorder when in replay mode to save the replay. The paths to the saved files is output to stdout after the recording is saved.
You can find the default output audio device (headset, speakers (in other words, desktop audio)) with the command pactl get-default-sink. Add monitor to the end of that to use that as an audio input in gpu-screen-recorder.
You can find the default input audio device (microphone) with the command pactl get-default-source. This input should not have monitor added to the end when used in gpu-screen-recorder.
Example of recording both desktop audio and microphone: gpu-screen-recorder -w $(xdotool selectwindow) -c mp4 -f 60 -a "$(pactl get-default-sink).monitor" -a "$(pactl get-default-source)" -o test_video.mp4.
Note that if you use multiple audio inputs then they are each recorded into separate audio tracks in the video file. There is currently no option to merge audio tracks, but it's a planned feature. For now I recommend using gpwgraph if you are using pipewire. Gpwgraph allows you to merge multiple audio inputs into one with a simple gui. If you use pulseaudio then you need to create a virtual sink, which is a bit more complex.

There is also a gui for the gpu-screen-recorder called gpu-screen-recorder-gtk.

Demo

Click here to watch a demo video on youtube

FAQ

How is this different from using OBS with nvenc?

OBS only uses the gpu for video encoding, but the window image that is encoded is copied from the GPU to the CPU and then back to the GPU (video encoding unit). These operations are very slow and causes all of the fps drops when using OBS. OBS only uses the GPU efficiently on Windows 10 and Nvidia.
This gpu-screen-recorder keeps the window image on the GPU and sends it directly to the video encoding unit on the GPU by using CUDA. This means that CPU usage remains at around 0% when using this screen recorder.

How is this different from using OBS NvFBC plugin?

The plugin does everything on the GPU and gives the texture to OBS, but OBS does not know how to use the texture directly on the GPU so it copies the texture to the CPU and then back to the GPU (video encoding unit). These operations are very slow and causes a lot of fps drops unless you have a fast CPU. This is especially noticable when recording at higher resolutions than 1080p.

How is this different from using FFMPEG with x11grab and nvenc?

FFMPEG only uses the GPU with CUDA when doing transcoding from an input video to an output video, and not when recording the screen when using x11grab. So FFMPEG has the same fps drop issues that OBS has.

TODO

  • Dynamically change bitrate/resolution to match desired fps. This would be helpful when streaming for example, where the encode output speed also depends on upload speed to the streaming service.
  • Show cursor when recording. Currently the cursor is not visible when recording a window.
  • Implement opengl injection to capture texture. This fixes VRR without having to use NvFBC direct capture.
  • Always use direct capture with NvFBC once the capture issue in mpv fullscreen has been resolved (maybe detect if direct capture fails in nvfbc and switch to non-direct recording. NvFBC says if direct capture fails).