Why avisynth bottlenecks ffmpeg nvenc-oding ?

[peter71]

Hi,

Transcoding of a H.264 file with my RTX2060 goes with a blazingly fast ~400 fps, when decoding is hw. accelerated too.

The same thing when served with AviSynth FFmpegSource2 (AviSynth+ 3.4 (r2923, 3.4, x86_64)) falls down to ~130fps.

Now I understand, that frameserving+decoding is a CPU intense job, and my CPU is kind of a relic (Xeon E5450), but Avsmeter says:

FPS (min | max | average): 121.2 | 251.4 | 196.3
Process memory usage (max): 47 MiB
Thread count: 9
CPU usage (average): 87.9%

This significant difference between 130 and 196 is the problem, I am trying to track down. Any idea please?

[Atak_Snajpera]

Ffmpegsource uses CPU while decoding hence slower processing. L-smash on other hand supports hw decoding.

[huhn]

the up and download of the video from the GPU Vram cost performance.

[peter71]

Ffmpegsource uses CPU while decoding hence slower processing. L-smash on other hand supports hw decoding.

Thanks for the hint. Ffmpeg nvenc encoding reaches 111 fps with LSMASHVideoSource decoding on CPU, and 96 fps with GPU (decoder = "h264_cuvid") decoding. My motherboard's PCIE x16 slot is obviously too old for this high amount of transferred data.

[poisondeathray]

Thanks for the hint. Ffmpeg nvenc encoding reaches 111 fps with LSMASHVideoSource decoding on CPU, and 96 fps with GPU (decoder = "h264_cuvid") decoding. My motherboard's PCIE x16 slot is obviously too old for this high amount of transferred data.

That might be some other problem , maybe incompatible video (some video profiles not decoded or supported by cuvid), or wrong lsmash version.
Try these
https://github.com/HolyWu/L-SMASH-Works/releases

Are you sure GPU is working for decoding? Look at GPU-z with the avs h264_cuvid script (or start up GPU-z, then use avsmeter script.avs -gpu)

[peter71]

That might be some other problem , maybe incompatible video (some video profiles not decoded or supported by cuvid), or wrong lsmash version.
Try these
https://github.com/HolyWu/L-SMASH-Works/releases

Are you sure GPU is working for decoding? Look at GPU-z with the avs h264_cuvid script (or start up GPU-z, then use avsmeter script.avs -gpu)

I guess the GPU must be decoding, because GPU usage is sustained at 20% during the process. If lsmash is called with sw. decoding, it is only about 10%. I tried out the 20191127 build from your link, it performed at 99 fps.

The command and output is:

$ /cygdrive/C/util/dv/ffmpeg-20200101-7b58702-win64-static/bin/ffmpeg.exe -vsync 0 -hwaccel cuvid -i lsmash.avs -b:v 5M -c:v h264_nvenc -c:a libmp3lame -b:a 128k gpu_5M_avisynth_lsmash_cuvid.mp4
ffmpeg version git-2020-01-01-7b58702 Copyright (c) 2000-2020 the FFmpeg developers
built with gcc 9.2.1 (GCC) 20191125
configuration: --enable-gpl --enable-version3 --enable-sdl2 --enable-fontconfig --enable-gnutls --enable-iconv --enable-libass --enable-libdav1d --enable-libbluray --enable-libfreetype --enable-libmp3lame --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenjpeg --enable-libopus --enable-libshine --enable-libsnappy --enable-libsoxr --enable-libtheora --enable-libtwolame --enable-libvpx --enable-libwavpack --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxml2 --enable-libzimg --enable-lzma --enable-zlib --enable-gmp --enable-libvidstab --enable-libvorbis --enable-libvo-amrwbenc --enable-libmysofa --enable-libspeex --enable-libxvid --enable-libaom --enable-libmfx --enable-ffnvcodec --enable-cuvid --enable-d3d11va --enable-nvenc --enable-nvdec --enable-dxva2 --enable-avisynth --enable-libopenmpt --enable-amf
libavutil 56. 38.100 / 56. 38.100
libavcodec 58. 65.100 / 58. 65.100
libavformat 58. 35.101 / 58. 35.101
libavdevice 58. 9.102 / 58. 9.102
libavfilter 7. 70.101 / 7. 70.101
libswscale 5. 6.100 / 5. 6.100
libswresample 3. 6.100 / 3. 6.100
libpostproc 55. 6.100 / 55. 6.100
Guessed Channel Layout for Input Stream #0.1 : stereo
Input #0, avisynth, from 'lsmash.avs':
Duration: 00:00:49.32, start: 0.000000, bitrate: 0 kb/s
Stream #0:0: Video: rawvideo (I420 / 0x30323449), yuv420p, 1920x1080, 30.03 fps, 30.03 tbr, 30.03 tbn, 30.03 tbc
Stream #0:1: Audio: pcm_f32le, 48000 Hz, stereo, flt, 3072 kb/s
File 'gpu_5M_avisynth_lsmash_cuvid.mp4' already exists. Overwrite? [y/N] y
Stream mapping:
Stream #0:0 -> #0:0 (rawvideo (native) -> h264 (h264_nvenc))
Stream #0:1 -> #0:1 (pcm_f32le (native) -> mp3 (libmp3lame))
Press [q] to stop, [?] for help
Output #0, mp4, to 'gpu_5M_avisynth_lsmash_cuvid.mp4':
Metadata:
encoder : Lavf58.35.101
Stream #0:0: Video: h264 (h264_nvenc) (Main) (avc1 / 0x31637661), yuv420p(progressive), 1920x1080, q=-1--1, 5000 kb/s, 30.03 fps, 10930 tbn, 30.03 tbc
Metadata:
encoder : Lavc58.65.100 h264_nvenc
Side data:
cpb: bitrate max/min/avg: 0/0/5000000 buffer size: 10000000 vbv_delay: N/A
Stream #0:1: Audio: mp3 (libmp3lame) (mp4a / 0x6134706D), 48000 Hz, stereo, fltp, 128 kb/s
Metadata:
encoder : Lavc58.65.100 libmp3lame
frame= 1481 fps= 99 q=29.0 Lsize= 31618kB time=00:00:49.28 bitrate=5255.1kbits/s speed= 3.3x
video:30816kB audio:770kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.101611%


The avs script is:

LoadPlugin("c:\util\dv\LSMASH_20191127\LSMASHSource.dll")

v = LSMASHVideoSource("VID_20190119_145158.mp4", decoder = "h264_cuvid" )
a = LSMASHAudioSource("VID_20190119_145158.mp4")

audiodub(a,v)