Performance difference in hardware encoding : V100 vs RTX3090


I have a RTX3090 and a V100 GPU. The V100 is a shared GPU. I have 8 GB of ram out of 32 GB. I ran some tests with NVENC and FFmpeg to compare the encoding speed of the two cards. On both cards, I encoded a video using these command line arguments :

ffmpeg -benchmark -vsync 0 -hwaccel nvdec -hwaccel_output_format cuda -i input.mp4 -c:v hevc_nvenc -c:a copy -qp 22 -preset <preset> output.mp4

For <preset>, I compared medium and fast.

The results :

RTX3090 :

  • Medium : 29.5 s
  • Fast : 9,2 s

V100 :

  • Medium : 29.6 s
  • Fast : 26.4 s

I don’t understand why the V100 is so much slower for the fast preset. Any ideas?