Am Mo., 22. Juni 2020 um 16:23 Uhr schrieb Matthew Reus <[email protected]>:
> *ffmpeg -hwaccel_device 0 -hwaccel nvdec -hwaccel_output_format cuda -i > 'udp://@172.16.16.25:9802?overrun_nonfatal=1&fifo_size=1000000\&overrun_nonfatal > <http://172.16.16.25:9802?overrun_nonfatal=1&fifo_size=1000000\&overrun_nonfatal>' > -vf "format=yuv420p|cuda,yadif_cuda=0:-1:0,scale_npp=1280:720" -b:v 2.5M > -bf 1 -aspect 16:9 -profile:v main -preset:v medium -c:v h264_nvenc -c:a > aac -b:a 128k -ar 48000 -strict -2 -f flv > 'rtmp://admin:[email protected]/livetv/AnjanTv.stream > <http://admin:[email protected]/livetv/AnjanTv.stream>' -vf > "format=yuv420p|cuda,yadif_cuda=0:-1:0,scale_npp=-720:576" -b:v 600k -bf 1 > -aspect 16:9 -profile:v main -preset:v medium -c:v h264_nvenc -c:a aac -b:a > 128k -ar 48000 -strict -2 -f flv > 'rtmp://admin:[email protected]/ott/ottAnjanTv.stream > <http://admin:[email protected]/ott/ottAnjanTv.stream>'* I try to avoid network issues (I realized at some point that they are *always* unreproducible and impossible to debug) but aren't you explicitly asking FFmpeg to allocate huge amounts of memory? Or to ask differently: Is the issue reproducible with software decoding? Is the issue reproducible with software encoding? Is the issue reproducible with file output? Is the issue reproducible with file input? If this does not help, there is a tool called massif which tells you where exactly memory allocations happen. Carl Eugen _______________________________________________ ffmpeg-user mailing list [email protected] https://ffmpeg.org/mailman/listinfo/ffmpeg-user To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
