FFmpeg 7 x86_64 Transcoding Performance Test Report

1. FFmpeg Information

# ffmpeg -version
ffmpeg version 7.0.1 Copyright (c) 2000-2024 the FFmpeg developers
built with gcc 13 (Ubuntu 13.2.0-23ubuntu4)
configuration: --disable-debug --disable-doc --disable-ffplay --enable-alsa --enable-cuvid --enable-ffprobe --enable-gpl --enable-libaom --enable-libass --enable-libfdk_aac --enable-libfontconfig --enable-libfreetype --enable-libfribidi --enable-libharfbuzz --enable-libkvazaar --enable-libmp3lame --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenjpeg --enable-libopus --enable-libplacebo --enable-librav1e --enable-libshaderc --enable-libsvtav1 --enable-libtheora --enable-libv4l2 --enable-libvidstab --enable-libvmaf --enable-libvorbis --enable-libvpl --enable-libvpx --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxml2 --enable-libxvid --enable-libzimg --enable-nonfree --enable-nvdec --enable-nvenc --enable-cuda-llvm --enable-opencl --enable-openssl --enable-stripping --enable-vaapi --enable-vdpau --enable-version3 --enable-vulkan
libavutil      59.  8.100 / 59.  8.100
libavcodec     61.  3.100 / 61.  3.100
libavformat    61.  1.100 / 61.  1.100
libavdevice    61.  1.100 / 61.  1.100
libavfilter    10.  1.100 / 10.  1.100
libswscale      8.  1.100 /  8.  1.100
libswresample   5.  1.100 /  5.  1.100
libpostproc    58.  1.100 / 58.  1.100

2. CPU Information

CPU Brand	CPU Model	CPU Frequency	CPU Cores	Threads per CPU	Total Threads
Intel	Xeon Gold 5218R	2.1GHz	2	40	80

3. Input Video Information

Resolution	Frame Rate	Bitrate	Codec
1080P	25	4M	H.264
1080P	25	3M	H.265

4. tmpfs 4KiB Read/Write Performance

To avoid disk IO or network IO bottlenecks, we used a RAM disk for transcoding input and output. The tmpfs 4K read/write performance was 3.GB/s for read and 1.9GB/s for write, meeting our maximum requirement of 400MB/s.

# df -h | grep "/tmp"
tmpfs                    6.0G  3.9G  2.2G   64% /tmp

dd if=/tmp/testfile of=/dev/null bs=4K count=10000000
1000000+0 records in
1000000+0 records out
4096000000 bytes (4.1 GB) copied, 1.15974 seconds, 3.5 GB/s

dd if=/dev/zero of=/tmp/testfile bs=4K count=10000000
dd: error writing '/tmp/testfile': No space left on device
1572848+0 records in
1572847+0 records out
6442381312 bytes (6.4 GB) copied, 3.32676 seconds, 1.9 GB/s

5. FFmpeg CLI Test Results

x10 represents 10 processes, x20 represents 20 processes, and -re x50 represents a fixed decoding frame rate with 50 processes.

# Transcoding Command Example
docker run -it -v /tmp:/media linuxserver/ffmpeg -nostdin -re -threads 1 -i /media/1080p.25fps.3M.265 -vf scale=720:480 -filter\_threads 1 -threads 1 -c:v libx264 -b:v 1M -bf 0 -g 50 -preset ultrafast -tune zerolatency -f h264 /media/outputs/d1\_1.25fps.1M.264

Input Video Codec	Output Video Resolution	Output Video Frame Rate	Output Video Bitrate	Output Video Codec	Transcoding Paths	CPU Usage	Memory Usage
H.264	D1	25	1M	H.264	x10: 38 x20: 35 -re x50: 37	x10: 80% x20: 99% -re x50: 100%	x10: 3.2G x20: 5.5G -re x50: 7.7G
H.264	CIF	25	128K	H.264
H.265	D1	25	1M	H.265	x10: 18 x20: 17 -re x50: 19	x10: 92% x20: 99% -re x50: 100%	x10: 3.9G x20: 8.2G -re x50: 16.4G
H.265	CIF	25	128K	H.265
H.265	D1	25	1M	H.264	x10: 30 x20: 28 -re x50: 34	x10: 84% x20: 100% -re x50: 100%	x10: 3.5G x20: 8.8G -re x50: 9.3G
H.265	CIF	25	128K	H.264
H.264	D1	25	1M	H.264	x10: 51 x20: 51 -re x50: 50	x10: 75% x20: 100% -re x50: 50%	x10: 2.4G x20: 7.4G -re x50: 7.6G
H.264	CIF	25	128K	H.264	x10: 65 x20: 65 -re x50: 50	x10: 67% x20: 100% -re x50: 40%	x10: 2.3G x20: 7.1G -re x50: 6.5G
H.265	D1	25	1M	H.265	x10: 27 x20: 24 -re x50: 24	x10: 83% x20: 95% -re x50: 100%	X10: 3.2G x20: 7.8G -re x50: 11.2G
H.265	CIF	25	128K	H.265	x10: 40 x20: 36 -re x50: 45	x10: 79% x20: 98% -re x50: 100%	x10: 2.8G x20: 6.9G -re x50: 9.8G
H.265	D1	25	1M	H.264	x10: 43 x20: 39 -re x50: 47	x10: 81% x20: 99% -re x50: 100%	x10: 3.2G x20: 7.7G -re x50: 8.1G
H.265	CIF	25	128K	H.264	x10: 52 x20: 47 -re x50: 50	x10: 73% x20: 96% -re x50: 49%	x10: 3.1G x20: 6.4G -re x50: 5.2G
H.264	FULL HD	25	74M	I420	x10: 91 x20: 89 -re x50: 50	x10: 98% x20: 100% -re x50: 32%	x10: 2.2G x20: 4.9G -re x50: 5.1G
H.265	FULL HD	25	74M	I420	x10: 61 x20: 62 -re x50:50	x10: 65% x20: 90% -re x50: 43%	x10: 2.9G x20: 5.1G -re x50: 5.6G

6. libavcodec Transcoding Load Performance

The following measurements show the 90th percentile under a load of 40 threads.

Load	Codec	Input Resolution	Output Format	Output Resolution	Time
Decoding	H.264	1080P	YUV420P	1080P	23 ms/frame
Decoding	H.265	1080P	YUV420P	1080P	27 ms/frame
Scaling	YUV420P	1080P	YUV420P	D1	6 ms/frame
Scaling	YUV420P	1080P	YUV420P	CIF	4 ms/frame
Encoding	YUV420P	D1	H.264	D1	2.39 ms/frame
Encoding	YUV420P	CIF	H.264	CIF	1.02 ms/frame
Encoding	YUV420P	D1	H.265	D1	30.86 ms/frame
Encoding	YUV420P	CIF	H.265	CIF	14.67 ms/frame

7. libavcodec Test Results

The following measurements show the 90th percentile under a load of 40 threads.

The results are slightly better than the FFmpeg CLI tests, likely because the video frames are pre-read into std::vector, and the encoding results are discarded directly.

CPU and memory usage are consistent with the FFmpeg CLI, so they will not be elaborated here.

Input Video Codec	Output Video Resolution	Output Video Frame Rate	Output Video Bitrate	Output Video Codec	Transcoding Paths
H.264	FULL HD	25	74M	YUV420P	152
H.265	FULL HD	25	74M	YUV420P	86
H.264	D1	25	1M	H.264	65
H.264	CIF	25	128K	H.264	93
H.265	D1	25	1M	H.265	19
H.265	CIF	25	128K	H.265	40
H.265	D1	25	1M	H.264	53
H.265	CIF	25	128K	H.264	68
H.264	D1+CIF	25	1M+128K	H.264	49
H.265	D1+CIF	25	1M+128K	H.265	15
H.265	D1+CIF	25	1M+128K	H.264	41