How to use hardware acceleration with FFMPEG for trimming and concatenation purposes

Question

I am running FFMPEG on a Windows 10 (64-bit) laptop with Intel i3 (6th generation) processor and AMD R5 M430 GPU.

I shot many videos on a recent trip. Cumulatively, they are over 100 hours long. Each video occupies 2-3 GB minimum and is at least 1 hour long.

I need to trim 2-4 videos and then concatenate them into one single file. This operation needs to be performed for all these videos.

Right now, I am trimming individual videos using the following command:

ffmpeg.exe -i "INPUT_VIDEO" -ss START_TIME -to END_TIME -c copy "OUTPUT_VIDEO"

And then I concatenate the videos using this command:

ffmpeg.exe -f concat -safe 0 -i VideosToBeConcatenated.txt -c copy "CONCATENATED_OUTPUT_VIDEO"

where VideosToBeConcatenated.txt contains the file names of the videos to be concatenated.

Running these commands in a single Command Window gives me a speed of 20-25x. Running them in two Command Windows simultaneously drops the speed to 10-12x, while also making my system lag/freeze.

Is there a way I could use my GPU to speed up these operation? Say, using the i3's Quicksync or R5's AMF/VCE?

I tried using -hwaccel dxva2 in the above mentioned commands as follows:

ffmpeg.exe -hwaccel dxva2 -i "INPUT_VIDEO" -ss START_TIME -to END_TIME -c copy "OUTPUT_VIDEO"

and

ffmpeg.exe -hwaccel dxva2 -f concat -safe 0 -i VideosToBeConcatenated.txt -c copy "CONCATENATED_OUTPUT_VIDEO"

However, I don't notice any increase in speed.

I also set ffmpeg.exe and cmd.exe to "High Performance" in the AMD Switchable Graphics settings. But the in-built Switchable Graphics Monitor suggests that neither of the processes are using the dGPU.

Is there a way I could use my GPUs (i3's Quicksync or R5's AMF/VCE) to speed up these operation without compromising the quality of the output videos?

[Edit:] I am trying to make sense of FFMPEG's Quicksync documentation but it's too mature for an amateur like me.

Answer 1

Since you are using -c copy in your commands, there is no decoding, filtering or encoding taking place and so hardware acceleration isn't relevant at all. The main bottleneck is disk I/O. On a regular internal HDD (with a 3rd gen CPU) I get speeds of 400x so I suspect you're reading or writing to an external drive. Or your internal drive could be slow, have a smaller cache. The biggest boost will come from writing to a different drive than the drive on which the inputs reside, with ideally both drives being internal.