Smart Video Transcoding Solution

Accelerate your transcoding pipelines for optimal density and quality save costs on hardware equipment

Meet us on IBC

Overview

In conventional pipelines the source AVC/HEVC stream is decoded to a raw YUV, then encoded again with different bitrate/resolution.

In our pipeline we reuse various information from the source stream to increase performance of encoder. It could be macroblock-level data like motion, intra mode, etc.

Optimal stat muxing

Multiple outputs

Fully hardware assisted

Ultra low latency

Principal diagram

Density

Build on the top of Intel Media SDK Vq SvT outperform vanilla hardware transcoding

Session: 720p transcodings 2500 kpbs into 2 outputs: 1500 kbps and 750 kbps

Plot shows average number of parallel sessions at 30fps on the same hardware

Quality

Rd-plots comparison between Vq SvT, Intel QSV and x264

High speed case is very fast/TU7 profile for x264/QSV

High quality - medium/TU4 profiles

Vq SvT trasrating based on a intel media sdk outperforms conventional transcoding in the if the bitrate difference between source and destination is moderate, like no more thant two times.

Monitor tool

Multifunctional utility for quality assessment, hardware montoring and performance evaluation.

Monitor stream data: PSNR, bitrate, speed

Monitor hardware metrics: GPU, VDBOX, RENDER loads

Configure various transcoding pipelines: 1:N, N:N, 1:1 via intuitive GUI

Assess visual quality in realtime for any transcoding

Conclusion

  • X264:
    ffmpeg -i stream.mp4 -tune zerolatency -bf 1 -preset medium -g 50 -profile:v baseline
    		-map 0 -c:v libx264 -b:v 750K -f flv out1
    		-map 0 -c:v libx264 -b:v 1500K -f flv out2
  • QSV:
    ffmpeg -hwaccel qsv -c:v h264_qsv -i stream.mp4 
    		-map 0 -c:v h264_qsv -b:v 750K -maxrate 750K -minrate 750K -f flv out1 
    		-map 0 -c:v h264_qsv -b:v 1500K -maxrate 1500K -minrate 1500K -f flv out2
  • Vq SvT:
    ffmpeg -i stream.mp4 -codec:v copy -bsf:v h264_mp4toannexb 
    		-map 0 -f tee [f=flv,bitrate=750]out1|[f=flv,bitrate=1500]out2
  • Hardware transcoding shows significant increase in density of transcoding channels per unit with insignificant quality loss
  • VQ SvT show even higher density but with limitation
  • 2 VDBOX GEN is desired for maximum performance(GT3 graphics):
    • VQ SvT is bound to VDBOX throughput
    • Intel QSV is bound to RENDER throughput
    • More VDBOX’s gives higher performance
  • Conventional transcoding by x264 and QSV shows comparable quality. While x264 is slightly better
  • VQ SvT should be split into two cases:
    1. Output bitrate reduction is less than 2x compare to source – in this case quality of transrate is higher than x264/QSV
    2. Output bitrate reduction is less than 2x compare to source – significant quality drop in transrate pipeline