[FFmpeg-trac] #11271(undetermined:closed): How to choose the best hwaccel?

[FFmpeg]

#11271: How to choose the best hwaccel?
-------------------------------------+-------------------------------------
             Reporter:  Mads         |                    Owner:  (none)
  Johansen                           |
                 Type:  enhancement  |                   Status:  closed
             Priority:  normal       |                Component:
                                     |  undetermined
              Version:  git-master   |               Resolution:  fixed
             Keywords:               |               Blocked By:
             Blocking:               |  Reproduced by developer:  0
Analyzed by developer:  0            |
-------------------------------------+-------------------------------------
Changes (by Mads Johansen):

 * resolution:   => fixed
 * status:  new => closed

Comment:

 As stated above, I created a 3 minute Mandelbrot video in yuv420p,
 yuv422p, yuv444p, yuv420p10le, yuv422p10le and yuv444p10le in 1080p and
 2160p.
 3 minutes was chosen to allow the decoding to flatten out (as I noticed
 massive differences in the beginning vs end).
 I did 6 decodes of each file to remove jitter.

 On windows I used the following command, changing cuda with the other
 options written above.
 {{{
 ffmpeg -benchmark -hwaccel cuda -i mandelbrot-yuv420p-2160p.mov -f null -
 > mandelbrot-yuv420p-2160p1cuda.txt 2>&1
 }}}

 The results are as follows:
 {{{
         utime   stime   rtime
 CUDA    393.46  6.14    33.79
 d3d11va 396.74  9.35    34.60
 d3d12va 393.98  6.44    33.87
 dxva2   392.62  6.17    33.79
 vulkan  392.14  7.45    33.80
 opencl  446.94  6.30    33.69
 }}}

 The outlyer is opencl, which exclusively used CPU to decode the videos.

 The only conclusion I can come up with is that it doesn't matter what
 hwaccel you use.
-- 
Ticket URL: <https://trac.ffmpeg.org/ticket/11271#comment:2>
FFmpeg <https://ffmpeg.org>
FFmpeg issue tracker