FFmpeg¶

About¶

FFmpeg is a complete, cross-platform solution to record, convert and stream audio and video. The version installed on Discoverer HPC is one compiled by John Van Sicle.

Versions avaiable¶

Supported versions¶

To check which FFmpeg versions are currently supported on Discoverer, execute on the login node:

module avail ffmpeg

User-supported installations¶

Important

Users are welcome to bring and install FFmpeg within their scratch folders, but those installations will not be supported by the Discoverer HPC team.

Running FFmpeg as Slurm job¶

Warning

Do not run FFmpeg on the login node. Always use Slurm batch script to invoke FFmpeg from within a job submitted to the queue.

Shown below is a Slurm batch script that handles an AVI video container and transforms the content into MPEG-4 format, after getting rid of the packet B-frames (some old video captures installed in experimental aparature produce such AVI video containers):

#!/bin/bash
#
#SBATCH --partition=cn         # Partition
#SBATCH --job-name=ffmpeg      # Job Name
#SBATCH --time=01:00:00        # WallTime

#SBATCH --nodes           1    # Use 1
#SBATCH --ntasks-per-node 1    # Use 1
#SBATCH --cpus-per-task   12   # Do not set here more than 12 threads

#SBATCH -o slurm.%j.out        # STDOUT
#SBATCH -e slurm.%j.err        # STDERR

module purge
module load ffmpeg/latest

cd $SLURM_SUBMIT_DIR

# Fixing the problem with the wasteful store of packet B-frames in AVI encoders:
#
ffmpeg -i source.avi -codec copy -bsf:v mpeg4_unpack_bframes source.fixed.avi

# Perform a two-pass encoding:
#
ffmpeg -y -i source.fixed.avi -c:v libx264 -b:v 1M -pass 1 -threads ${SLURM_CPUS_PER_TASK} -an -f mp4 /dev/null
ffmpeg -y -i source.fixed.avi -c:v libx264 -b:v 1M -pass 2 -threads ${SLURM_CPUS_PER_TASK} -c:a aac source.fixed.mp4

Specify the parmeters and resources required for successfully running and completing the job:

the Slurm partition of compute nodes, based on your project resource reservation (--partition)

the job name, under which the job will be seen in the queue (--job-name)

the wall time for running the job (--time)

the requested numbef of threads (--cpus-per-task)

Note

FFmpeg has its limit when using threading. So try to vary --cpus-per-task until obtain some better productivity.

Save the complete Slurm job description as a file, for example /discofs/$USER/ffmpeg/run_ffmpeg.batch, and submit it to the queue:

cd /discofs/$USER/ffmpeg
sbatch run_ffmpeg.batch

Upon successful submission, the standard output will be directed into the file /discofs/$USER/ffmpeg/slurm.%j.out (where %j stands for the Slurm job ID.

Getting help¶

See Getting help