**This is an old revision of the document!**
Encoding-Quality-Testing
Created with the vmaf-docker repo.
Testing VAAPI vs. software encoding for live streaming
The graphs currently contain no confidence intervals (values may differ, but the difference may not be relevant)
The codec settings are the ones currently in use by the C3VOC for live streaming.
All the reference content is 1080p25.
Codec Settings
- vaapi: nothing but keyint and bitrates set
-vaapi_device /dev/dri/renderD128
-hwaccel vaapi -hwaccel_output_format vaapi
-i $ref
-vf 'format=nv12|vaapi,hwupload'
-c:v {codec}_vaapi
-keyint_min:v 75 -g:v 75
-b:v {rate}k -maxrate:v {rate}k -bufsize {rate}k
- libvpx-vp9 (current C3VOC transcoding settings)
-i $ref
-c:v libvpx-vp9
-deadline:v realtime -cpu-used:v 8
-threads:v 8
-frame-parallel:v 1 -tile-columns:v 2
-keyint_min:v 75 -g:v 75
-crf:v 23
-b:v {rate}k -maxrate:v {rate}k -bufsize {rate}k
- x264 (current C3VOC encoder-stream settings)
-i $ref
-c:v libx264 -preset:v veryfast
-profile:v main -flags +cgop
-threads:v 0 -g:v 75
-crf:v 21
-maxrate:v {rate}k -bufsize {rate}k
The codecs are tested with bitrates from 1Mbit/s to 5.2Mbit/s. Currently the 1080p25 streams are coded with 2.8Mbit/s maxrate.
Versions
- CPU: Intel(R) Core(TM) i7-8665U CPU @ 1.90GHz
- ffmpeg-4.1.4
- libvmaf-1.3.15
- libvpx-1.7.0 (buster)
- libx264-2:0.155.2917+git0a84d98-2 (buster)
Reference
- Ref1 - Noisy MPEG2 36C3
- Ref2 - MPEG2 36C3
- Red Kayak - Xiph (converted to 8Bit 25FPS)
- Rush Hour - Xiph
Results
| Reference | Intel 8th Gen (i7-8665U) | AMD Navi (RX 5700 XT) + AMD Ryzen 5 2600 navi_scores.json |
|---|---|---|
| Ref1 | bitrates_ref1.pdf speed_ref1.pdf * no improvement beyond 2Mbit/s * VAAPI HEVC & h.264 with best quality * libvpx-VP9 better and faster than x264 * VAAPI HEVC not faster than software encoding | navi_bitrates_ref1.nut.pdf navi_speed_ref1.nut.pdf |
| Ref2 | bitrates_ref2.pdf speed_ref2.pdf * no improvement beyond 1.5Mbit/s * similar Mean for all codecs * better Minimum scores for libvpx/x264 * high speeds for libvpx/x264 | navi_bitrates_ref2.nut.pdf navi_speed_ref2.nut.pdf |
| Red Kayak | bitrates_red_kayak.pdf speed_red_kayak.pdf * very good quality never reached * VAAPI VP9 has rate control problems at low bitrates (resulting bitrate large but quality low) * VAAPI HEVC > VAAPI h.264 > libvpx-VP9 at high bitrates * libvpx-VP9 best at low bitrates * libvpx-VP9 better than VAAPI VP9, albeit slower * libvpx-VP9 better and faster than x264 | navi_bitrates_red_kayak.nut.pdf navi_speed_red_kayak.nut.pdf |
| Rush Hour | bitrates_rush_hour.pdf speed_rush_hour.pdf VAAPI HEVC > VAAPI h.264 > libvpx-VP9 > VAAPI-VP9 > x264 | navi_bitrates_rush_hour.nut.pdf navi_speed_rush_hour.nut.pdf |
| Forest | navi_bitrates_rush_hour.nut.pdf navi_speed_rush_hour.nut.pdf |
Intel Conclusion
Keep in mind that this only applies to the stream-settings used
- Quality: VAAPI HEVC > VAAPI h.264 > libvpx-VP9 > VAAPI-VP9 > x264
- Speed: VAAPI h.264 > VAAPI VP9 > VAAPI HEVC >= libvpx-VP9 ~= x264
- VAAPI VP9 has lower speeds and quality than VAAPI h.264
- libvpx-VP9 has higher quality than x264 at the same speeds
- VAAPI-HEVC outperforms VAAPI-h.264 on hard content, but is 3x slower