FFMPEG build with NVDEC or CUVID for HW Accelerated decoding

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

I have learn’t that latest FFMPEG 4 supports HW acceleration which is compatible for Maxwell arch GPUs like on Jetson Nano. However, I have found official tutorial from NVIDIA for compiling FFMPEG from source with NVENC enabled. But I haven’t found any guide for NVDEC/CUVID.

adding --enable-nvdec returns unrecognized option during ./configure. Any help is appreciated.

2

Edit: I don’t think hwaccel is supported at all with ffmpeg on Nano.

3

I’ve seen a demo from Nvidia Deepstream to run 8 channel 1080p video at 30fps. I wonder how that could become possible without hardware acceleration!!
Maybe TX1 has its own video decoder/encoder hardware elements. They aren’t available on ffmpeg but You can leverage it via “Multimedia API” mentioned in JetPack SDK | NVIDIA Developer .

4

Without ffmpeg. I can’t find the thread (I have been searching) but I recall a mod stating that ffmpeg hardware decode isn’t supported and the 8 channel level of performance is only with their MMAPI. Gstreamer has accelerated decode and encode on nano as well.

5

Hi,
ffmpeg with hardware acceleration is not supported on TX2, Jetson Xavier and Jetson Nano.
It is supported on desktop GPUs.