P010le format in ffmpeg p010le is a 10-bit packed pixel format that stores the luminance and chrominance components in separate bytes. ffmpeg -init_hw_device qsv=hw -filter_hw_device hw -v verbose -f rawvideo -video_size 1920x1080 -pix_fmt p010le -i input. Previously you had to compile x264 with --bit-depth=10, and then link your ffmpeg to either an 8-bit or Without hardware acceleration, a typical command would be ffmpeg -i input. Try moving format=p010le before hwupload. Try scale_cuda=720:404:format=p010le and -profile:v main10: ffmpeg -vsync 0 -hwaccel cuda -hwaccel_output_format cuda -i input. (Kaby Lake applies Intel Core Gen 7). There is p010le and p010be. mkv -pix_fmt yuv420p10le -c:v hevc_nvenc -profile:v main10 -cq 21 out. mkv -vf scale_cuda=720:404:format=p010le -c:v hevc_nvenc -profile:v main10 -preset p7 -rc vbr -rc-lookahead 20 -b:v 2000k -bufsize 5M Here is the example command: time ffmpeg -hide_banner -hwaccel cuda -hwaccel_output_format cuda -i source. For example to play the 10bpp sequences provided by the HEVC standardization which are in 4:2:0 10bpp, you can use the following ffmpeg command: $ ffplay -video_size 2560x1600 -pixel_format yuv420p10le SteamLocomotiveTrain_2560x1600_60_10bit_crop. yuv -an \ -vf 'hwupload=extra_hw_frames=64,format=qsv' -c:v hevc_qsv -profile:v main10 output. The "p" in p010le stands for "planar," which means that the luminance and chrominance components are stored in separate planes. mkv -vf "hwdownload,format=p010le,format=yuv420p,hwupload_cuda" -c:v hevc_nvenc -preset slow test. So, it depends, you may wanna look how it works in ffmpeg and VAAPI. mkv "hwdownload,format=p010le,format=yuv420p,hwupload_cuda" downloads frames from hardware to system memory in the original format (in this According to the following page, "Encode a 1080p p010le raw yuv input as HEVC main10 profile (supported since Kaby Lake platform)". yuv For some reason when encoding with hevc_nvenc (I use an NVIDIA GTX 1660 Ti GPU), ffmpeg (the command line tool) changes the pixel format to a 10 or 12 bit variant (p010le, gray12le, yuv420p10le, yuv444p12le, ), but I would like to keep the full 16 bits, since this affects the quality of the depth stored. mkv -pix_fmt yuv420p10le -c:v libx265 -crf 21 -x265-params profile=main10 out. mkv. yuv420p10le is different from p010le, p010le has packed chroma and planar luma. Without hardware acceleration, a typical command would be ffmpeg -i input. Using CUDA (on a Pascal 1050 Ti), I expect the corresponding command to be ffmpeg -i input. The issue that hevc_nvenc uses pixel format cuda, and we can't use -pix_fmt p010le. You can also check x264 for supported bit depths: [] Output bit depth: 8/10. Little and big endian. Previously you had to compile x264 with --bit-depth=10, and then link your ffmpeg to either an 8-bit or . 10-bit pixel formats are: yuv420p10le, yuv422p10le, yuv444p10le. I don't know since which version FFmpeg support it. mp4 Also try ffmpeg -pix_fmts to list the available formats. nfh hroygp npjxup boxajn bmbzyv isq gbfkas vftpvt pwqfv orfk