QSVEnc Documentation

QSVEnc is a tool designed to experiment with the performance (quality and speed) of Intel hardware encoders (QuickSync Video - QSV). It is provided in two main forms:

1. QSVEncC.exe: A standalone command-line version for direct use.
2. QSVEnc.auo: An output plugin for AviUtl that uses oneVPL for encoding.

Developers and users can choose between the CLI for automated or scripted workflows and the AviUtl plugin for integration into a GUI-based video editing environment.

QSVEnc is designed to investigate the performance and image quality of Intel's QuickSyncVideo (QSV) hardware encoder. It is distributed in two forms:

1. QSVEncC.exe: A standalone command-line version used for transcoding tasks.
2. QSVEnc.auo: An output plugin specifically for the AviUtl video editing software.

Both versions share core encoding features such as support for H.264/AVC, H.265/HEVC, MPEG2, VP9, and AV1, along with various encode modes (CQP, CBR, VBR, etc.) and VPP (Video Pre-Processing) filters.

QSVEnc is licensed under the MIT License. It utilizes several third-party libraries, including:

• oneVPL
• ffmpeg
• libplacebo
• libhdr10plus
• libdovi
• libass
• tinyxml2
• ttmath
• clRNG
• dtl
• Caption2Ass

For specific license details regarding these dependencies, please refer to their respective header files or the license.txt file in the repository.

The standalone command-line version QSVEncC includes additional features:

• Hardware Decoding Support: Supports MPEG2, H.264/AVC, HEVC, VP8, VP9, and AV1.
• Flexible Pipeline: Allows free combination of QSV-based decoding, VPP, and encoding.
• Passthrough: Supports outputting without encoding.
• Format Support: Supports various formats including avi(vfw), avs, vpy, y4m, and raw.
• Audio Processing: Uses libavcodec/libavformat for audio processing.
• Muxing: Supports muxing while outputting using libavcodec/libavformat.

Starting from version 8.00, you can use VPP (Video Processing Pipeline) filters even when using avcodec encoders (e.g., av_libsvtav1, av_libvvenc, av_libvpx-vp9) by specifying them with the -c flag. You can pass encoder-specific parameters using the --avcodec-prms option.

Note: This feature is available when using the -c av_xxx syntax.

-c av_libsvtav1 --avcodec-prms "preset=6,crf=30,svtav1-params=enable-variance-boost=1:variance-boost-strength=2"

The AviUtl output plugin QSVEnc.auo provides:

• Audio encoding
• Muxing functionality for audio and chapters
• Automatic field shift support

In Windows environments with multiple QSV-capable GPUs, QSVEncC performs automatic selection based on the following logic:

1. Capability Check: It verifies if a GPU supports the requested Codec, Profile, and Level. It also checks for 10bit depth encoding and HW decoding if specified in the options.
2. Preference Logic: If multiple GPUs meet the requirements, the software prefers the GPU with the lowest Video Engine (VE) utilization and lowest GPU core utilization.

Note on Timing: Utilization values are checked during the initialization phase. If multiple tasks are started simultaneously, they may all target the same GPU before the utilization values are updated, potentially leading to uneven distribution.

Manual Selection: You can manually specify a GPU using the --device option.

The command-line version QSVEncC includes additional capabilities beyond the core encoding engine:

• Hardware Decoding: Supports QSV(hw) decoding for MPEG2, H.264/AVC, HEVC, VP8, VP9, and AV1.
• Input Formats: Supports various formats including .avs, .vpy, .y4m, and raw data.
• Muxing/Demuxing: Uses libavformat for demuxing and muxing.
• Decoding: Uses libavcodec for decoding.

To install QSVEncC on Windows 10, follow these steps:

1. Install Intel Graphics driver: Ensure your Intel graphics drivers are up to date.
2. Download Windows binary: Download the appropriate binary from the QSVEnc releases page.
   • Use QSVEncC_x.xx_Win32.7z for 32-bit systems.
   • Use QSVEncC_x.xx_x64.7z for 64-bit systems.

QSVEncC can be run directly from the extracted directory without a formal installation process.

Building VapourSynth allows QSVEncC to load .vpy scripts. This is an optional step.

1. Install Build Tools

sudo apt install python3-pip autoconf automake libtool meson

2. Install zimg

git clone https://github.com/sekrit-twc/zimg.git --recursive
cd zimg
./autogen.sh
./configure
make && sudo make install
cd ..

3. Install Cython

sudo pip3 install Cython

4. Build VapourSynth

git clone https://github.com/vapoursynth/vapoursynth.git
cd vapoursynth
./autogen.sh
./configure
make && sudo make install

# Link vapoursynth for automatic loading (replace python3.x with your version)
sudo ln -s /usr/local/lib/python3.x/site-packages/vapoursynth.so /usr/lib/python3.x/lib-dynload/vapoursynth.so
sudo ldconfig

5. Verify Installation

vspipe --version