Can FFMPEG use without using CPU

[kucing]

sorry for my bad English,
I know I cannot use multiple GPUs as an FFMPEG accelerator,
so my question is :
can the FFMPEG encoding (hevc_nvenc) work using GPU only without (or minimum) CPU usage?
so I can do the encoding simultaneously?

Note : I am Already Using Multiple GPUs for Python Torch (CUDA)

thank you for reading, have a nice day

[titoCA321]

You can encode with CPU or GPU. If you are using GeForce cards, there is no support for multiple encoders on the gaming GPUs. If you want multiple encoders on NVIDIA GPUs you will need to switch over to Nvidia compute or professional cards on the Tesla or Quadro class.
If you are just encoding or decoding video streams the the T4 will serve you best. The gaming cards sacrificed video encoders for gaming performance.

[pandy]

sorry for my bad English,
I know I cannot use multiple GPUs as an FFMPEG accelerator,
so my question is :
can the FFMPEG encoding (hevc_nvenc) work using GPU only without (or minimum) CPU usage?
so I can do the encoding simultaneously?

You can use multiple GPU's - everything depends on proper configuration - check https://developer.nvidia.com/blog/nvidia-ffmpeg-transcoding-guide/ or https://stackoverflow.com/questions/65983094/how-use-ffmpeg-multi-gpu-proccess
NVidia provide also guides how to minimize CPU usage (keeping all video processing close to GPU as much as possible so CPU will be used in few % perhaps).

Also some non Quadro cards support multiple NVEnc instances (technically you encode not on GPU but on dedicated SoC being part of GPU - this SoC is called NVEnc)
check https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new , beside to this most of NVEnc instances can support concurrent sessions (so multiple streams encoded at a cost of speed).

https://docs.nvidia.com/video-technologies/video-codec-sdk/12.1/ffmpeg-with-nvidia-gpu/index.html

[kucing]

You can encode with CPU or GPU. If you are using GeForce cards, there is no support for multiple encoders on the gaming GPUs. If you want multiple encoders on NVIDIA GPUs you will need to switch over to Nvidia compute or professional cards on the Tesla or Quadro class.
If you are just encoding or decoding video streams the the T4 will serve you best. The gaming cards sacrificed video encoders for gaming performance.

thank you, titoCA321
i use GeForce cards for gaming and Python Torch (CUDA), so it impossible for current device to do that,
thank you again

[kucing]

sorry for my bad English,
I know I cannot use multiple GPUs as an FFMPEG accelerator,
so my question is :
can the FFMPEG encoding (hevc_nvenc) work using GPU only without (or minimum) CPU usage?
so I can do the encoding simultaneously?

You can use multiple GPU's - everything depends on proper configuration - check https://developer.nvidia.com/blog/nvidia-ffmpeg-transcoding-guide/ or https://stackoverflow.com/questions/65983094/how-use-ffmpeg-multi-gpu-proccess
NVidia provide also guides how to minimize CPU usage (keeping all video processing close to GPU as much as possible so CPU will be used in few % perhaps).

Also some non Quadro cards support multiple NVEnc instances (technically you encode not on GPU but on dedicated SoC being part of GPU - this SoC is called NVEnc)
check https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new , beside to this most of NVEnc instances can support concurrent sessions (so multiple streams encoded at a cost of speed).

https://docs.nvidia.com/video-technologies/video-codec-sdk/12.1/ffmpeg-with-nvidia-gpu/index.html

thank you for the info pandy