Install FFmpeg on Nvidia CUDA Container

This documentation describes to install FFmpeg on nvidia/cuda container to use the Nvidia GPU to accelerate encoding.

If you want to know how to install FFmpeg with NVIDIA GPU on Linux, go to see that.

FFmpeg can support hardware-based decoding and encoding for Nvidia GPU card. With the help of Nvidia GPU, h264_nvenc can lead encoding speed with 5x faster than libx264 in GTX1080 card.

Let's see how to install everything one by one on the Nvidia CUDA Docker container nvidia/cuda:12.2.0-devel-ubuntu20.04, in which CUDA toolkit and GPU driver are already included.

note

It must use nvidia/cuda:xxx-devel-xxx image to build FFmpeg, because the dev image contain all the necessary libraries.

Prerequisites

Make sure Nvidia GPU Driver is installed in your host machine! As it will be mounted into the container.

Use ldconfig to check if the required Nvidia GPU driver libraries are available inside the container. Such as,

ldconfig -p | grep libcuda

note

When running in the nvidia/cuda Docker container, what Nvidia libraries(from the host machine) should be mounted inside the container are specified by the NVIDIA_DRIVER_CAPABILITIES env variable, see driver-capabilities. Here for FFmpeg to employ GPU, it should be included at least as NVIDIA_DRIVER_CAPABILITIES=video,utility.

Step by Step

docker run --rm --runtime=nvidia \
    -e NVIDIA_VISIBLE_DEVICES=all \
    -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
    nvidia/cuda nvidia-smi

docker run --rm --runtime=nvidia \
    -e NVIDIA_VISIBLE_DEVICES=all \
    -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
    nvidia/cuda bash

Complete Dockerfile

The source code is available at Dockerfile

nvidia-cuda-ffmpeg/Dockerfile
# pull official base image, NOTE to use `devel`
FROM nvidia/cuda:12.2.0-devel-ubuntu20.04

# Set environment variables to prevent interactive prompts during installation
ENV DEBIAN_FRONTEND=noninteractive

# Install dependencies
RUN apt-get update && \
    apt-get install -y wget git make build-essential pkg-config yasm cmake libtool libc6 libc6-dev unzip libnuma1 libnuma-dev && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

# Install FFmpeg 6.1
RUN mkdir -p /opt

RUN cd /opt/ && wget https://github.com/FFmpeg/nv-codec-headers/releases/download/n12.1.14.0/nv-codec-headers-12.1.14.0.tar.gz -O nv-codec-headers-12.1.14.0.tar.gz && tar -xf nv-codec-headers-12.1.14.0.tar.gz
RUN cd /opt/nv-codec-headers-12.1.14.0 && \
    make install PREFIX=/usr

RUN cd /opt && wget https://ffmpeg.org/releases/ffmpeg-6.1.tar.xz -O ffmpeg-6.1.tar.xz && tar -xf ffmpeg-6.1.tar.xz
RUN cd /opt/ffmpeg-6.1 && \
    ./configure --enable-cuda --enable-cuvid --enable-nvdec --enable-nvenc --enable-nonfree --enable-libnpp --extra-cflags=-I/usr/local/cuda/include --extra-ldflags=-L/usr/local/cuda/lib64 --enable-libfreetype --enable-libharfbuzz --enable-libfontconfig --enable-libfribidi --enable-filter=drawtext && \
    make -j 8 && \
    make install PREFIX=/usr

CMD ["ffmpeg", "-version"]

Known issues

Nvidia Docker encoding stops after long running time with such error message: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected.

[Issue]: NVidia Docker transcoding randomly stops working after 5 minutes to 4 hours later. · Issue #9287 · jellyfin/jellyfin · GitHub

Possible solution:

Edit /etc/defautls/grub,

GRUB_CMDLINE_LINUX_DEFAULT="quiet splash systemd.unified_cgroup_hierarchy=0"

Then run update-grub and reboot.

Install FFmpeg on Nvidia CUDA Container

Prerequisites​

Step by Step​

Complete Dockerfile​

Known issues​

Prerequisites

Step by Step

Complete Dockerfile

Known issues