Jetson ffmpeg hardware acceleration usring nvdec but don't use GPU

[20171522]

Hi
I want to use hardware acceleration.
[****]
[Jetson Xavier]
[CUDA 10.2]

I met some problem but followed below the link and solved.

https://forums.developer.nvidia.com/t/jetson-tx2-and-ffmpeg-cant-initialize-nvrm-channel/165629/5
https://forums.developer.nvidia.com/t/hardware-accelerated-video-playback-with-l4t-ffmpeg/154019/7

And I used this command.

'ffmpeg -c:vh264_ntv4l2dec -i [input] [output]'

Run at ‘nvdec’ 700 MHz but do not use GPU.

So I looked into it (using GPU) and saw a post saying that Jetson Xavier doesn’t support hardware acceleration using GPU.

Summary

1. difference between -hwaccel cuda option and nvdec.
2. How can I use GPU to accelerate hardware if it’s true that it’s not supported and if it’s false.

Best regards.

[DaneLLL]

Hi,
On latest desktop GPUs and Jetson platforms, encoding/decoding tasks are done on individual NVENC/NVDEC engines. So that GPU is available for deep learning inference.

For capability of the hardware engines, please check
https://developer.nvidia.com/jetson-xavier-nx-data-sheet

[20171522]

Yes I checked data sheet.

NVDEC support 1080p60 10 stream but my video stumbles and is lagging.

My video
[ 1080p30 x1 ]
[ 720p35 x2 ]

I want to decrease CPU usage

Thanks for the reply.

[DaneLLL]

Hi,
For using ffmpeg there is additional buffer copy since it requires CPU buffer in the frameworks For optimal performance, we would suggest use jetson_multimedia_api or gstreamer. Please check developer guide and give it a try.

[20171522]

Okay, I will try.

Thank you.