AMD AMF

https://github.com/GPUOpen-LibrariesAndSDKs/AMF/tree/master/amf/public/include/components

https://github.com/GPUOpen-LibrariesAndSDKs/AMF/blob/master/amf/doc/AMF_Video_Encode_HEVC_API.md

https://github.com/GPUOpen-LibrariesAndSDKs/AMF/blob/master/amf/doc/AMF_Video_PreAnalysis_API.md

OBS
https://github.com/obsproject/obs-studio/blob/master/plugins/obs-ffmpeg/texture-amf.cpp

---------------------------------------------------------------------------------------------------------


https://github.com/GPUOpen-LibrariesAndSDKs/AMF/issues/439#issuecomment-1917886326

https://github.com/GPUOpen-LibrariesAndSDKs/AMF/issues/417#issuecomment-2040289753

ENCODERS.vdprof

{
	"description": "VirtualDub external encoder profile collection",
	"externalEncoders": {
		"sets": {
			"ProRes": {
				"videoEncoder": "ProRes",
				"audioEncoder": "",
				"multiplexer": "",
				"description": "",
				"extension": "",
				"processPartial": true,
				"useOutputAsTemp": false
			},
			"ProRes Q4": {
				"videoEncoder": "ProRes Q4",
				"audioEncoder": "",
				"multiplexer": "",
				"description": "",
				"extension": "",
				"processPartial": true,
				"useOutputAsTemp": false
			}
		},
		"profiles": {
			"ProRes": {
				"name": "ProRes",
				"program": "ffmpeg.exe",
				"commandArguments": "-f rawvideo -pix_fmt %(pix_fmt) -s %(width)x%(height) -r %(fpsnum)/%(fpsden) -i - -c:v prores_ks -pix_fmt yuv420p10le %(tempvideofile)",
				"outputFilename": "%(outputname).mov",
				"type": 0,
				"pixelFormat": "yuv444p16le",
				"inputFormat": 0,
				"checkReturnCode": true,
				"logStdout": true,
				"logStderr": true,
				"bypassCompression": false,
				"predeleteOutputFile": false
			},
			"ProRes Q4": {
				"name": "ProRes Q4",
				"program": "ffmpeg.exe",
				"commandArguments": "-f rawvideo -pix_fmt %(pix_fmt) -s %(width)x%(height) -r %(fpsnum)/%(fpsden) -i - -c:v prores_ks -q:v 4 -pix_fmt yuv420p10le %(tempvideofile)",
				"outputFilename": "%(outputname).mov",
				"type": 0,
				"pixelFormat": "yuv444p16le",
				"inputFormat": 0,
				"checkReturnCode": true,
				"logStdout": true,
				"logStderr": true,
				"bypassCompression": false,
				"predeleteOutputFile": false
			}
		}
	}
}

Check encoder settings like this
ffmpeg -h encoder=libvpx-vp9

Say yes if asked
ffmpeg -y

Show encoding time in seconds
-benchmark

Write to logfile
-report

Hide ffmpeg (for piping)
-loglevel quiet

Or show warnings only
-loglevel warning

mkvpropedit --add-track-statistics-tags INPUT.mkv
Read more
Needed to write statistics for VP9 / Opus

mkvpropedit --tags all:"" --delete title INPUT.mkv
Delete all tags and title

ffmpeg -i INPUT.flac -c:a libopus -b:a 576K OUTPUT.opus

• 2CH = 160K
• 6CH = 432K
• 8CH = 576K

NTSC
-color_primaries smpte170m -colorspace smpte170m -color_trc smpte170m -color_range tv
PAL
-color_primaries bt470bg -colorspace bt470bg -color_trc gamma28 -color_range tv
NTSC HD
-color_primaries bt709 -color_trc bt709 -colorspace bt709 -color_range tv

How to concat: https://blog.feurious.com/concatenate-videos-together-using-ffmpeg

ffmpeg -y -benchmark -init_hw_device opencl=GPU -filter_hw_device GPU -i SAMPLE.mkv -map_metadata -1 -vf "format=p010,hwupload,tonemap_opencl=tonemap=hable:tonemap=hable:primaries=bt709:transfer=bt709:matrix=bt709:range=tv:format=p010,hwdownload,format=p010" -c:v prores_ks -qscale:v 4 TonemapGPU.mov

ffmpeg -y -benchmark -i SAMPLE.mkv -map_metadata -1 -vf "zscale=transfer=linear,tonemap=hable,zscale=primaries=bt709:transfer=bt709:matrix=bt709:range=tv,format=yuv420p10" -c:v prores_ks -qscale:v 4 TonemapCPU.mov

ffmpeg -y -benchmark -i 0001.mov -map_metadata -1 -c:v libx264 -x264-params "ref=4:keyint=60:scenecut=0" -profile:v high -level:v 4.2 -preset veryslow -crf 12 -vf scale=1920:-1 -pix_fmt nv12 -color_primaries bt709 -color_trc bt709 -colorspace bt709 -color_range tv -an OUTPUT.mkv

Note: Level 4.2 is limited to a maximum of 4 reference frames.

https://silentaperture.gitlab.io/mdbook-guide/encoding/x264.html

ffmpeg -y -benchmark -i INPUT.mkv -c:v prores_ks -q:v 4 -vf "dejudder, fps=30000/1001, fieldmatch=order=tff:mode=pcn_ub:combmatch=full, bwdif=mode=0:deint=interlaced, decimate" -an OUTPUT.mkv

ffmpeg -y -benchmark -i INPUT.mkv -c:v h264_amf -quality balanced -qp_i 32 -qp_p 32 -g 120 -an OUTPUT_AVC_QP32.mkv

460 FPS - 152 MB - 8452 kb/s avg, 21934 kb/s Max

ffmpeg -y -benchmark -i INPUT.mkv -c:v hevc_amf -quality balanced -qp_i 30 -qp_p 30 -g 120 -an OUTPUT_HEVC_QP30.mkv

460 FPS - 155 MB - 8612 kb/s, 26722 kb/s Max

ffmpeg -y -benchmark -i INPUT.mkv -c:v libx264 -x264-params "keyint=120:scenecut=0:nal-hrd=none:filler=0:vbv-maxrate=16000:vbv-bufsize=16000:bitrate=8000:threads=24" -profile:v high -preset medium -an x264.mkv

127 FPS - 143 MB - 7946 kb/s, 18114 kb/s Max

lookahead cannot surpass keyint

ffmpeg -y -benchmark -init_hw_device vulkan=GPU -hwaccel vulkan -hwaccel_output_format vulkan -extra_hw_frames 30 -i "C:\Users\Admin\Downloads\New folder\OPM.avs" -filter_hw_device GPU -vf "format=nv12,hwupload,libplacebo=w=iw/2:h=ih/2:downscaler=sinc,hwdownload,format=yuv420p" -c:v hevc_amf -qp_i 18 -qp_p 18 -g 240 -an TEST.mkv

ffmpeg -y -benchmark -init_hw_device vulkan -i "C:\Users\Admin\Downloads\New folder\OPM.avs" -vf "libplacebo=w=iw*2:h=ih*2:upscaler=sinc" -c:v libx264 -preset ultrafast -qp 0 -g 1 -an test1234.mkv

ffmpeg -y -benchmark -i INPUT.mkv -c:v libx264 -x264-params "ref=4:psy-rd=1,0.15:merange=32:threads=24:lookahead-threads=6:bframes=3:b-pyramid=none:keyint=48:scenecut=0:rc-lookahead=48:crf=19:qcomp=0.75:vbv-maxrate=5000:vbv-bufsize=10000:nal-hrd=none:filler=0:aq-strength=0.8" -preset veryslow -pix_fmt yuv420p -an H264.mkv

ffmpeg -y -benchmark -i Bleach02.avs -pix_fmt yuv422p10le -c:v prores_ks cut005_48.mkv

ffmpeg -y -benchmark -i cut03.mkv -pix_fmt yuv422p10le -c:v prores_ks -vf mpdecimate,setpts=N/24/TB -vsync vfr cut003.mkv

ffmpeg -y -f concat -i list.txt -c copy CUTS_48.mkv

ffmpeg -y -f concat -i list.txt -c:v prores_ks CUTS_TEST.mkv

ffmpeg -i cut003_48.mkv -vf fps=48000/1001 cut03\%04d.png

ffmpeg -y -r 48 -i cut03\%04d.png -c:v prores_ks -pix_fmt yuv422p10le cut03_FIXED_48.mkv

list.txt

file 'cut01_FIXED_48.mkv'
file 'cut02_FIXED_48.mkv'
file 'cut03_FIXED_48.mkv'
file 'cut04_FIXED_48.mkv'
file 'cut05_FIXED_48.mkv'

If you set TargetBitrate before setting PeakBitrate AMF tries to adjust peak bitrate

GPUOpen-LibrariesAndSDKs/AMF#488 (comment)

---

If you would like to have more control over bitrate accuracy, please consider using PEAK_CONSTRAINED_VBR

GPUOpen-LibrariesAndSDKs/AMF#489 (comment)

---

GPUOpen-LibrariesAndSDKs/AMF#439 (comment)

PAEngineType=

• Host/CPU (1)
• DX11 (3)
• OpenCL (4)
• Vulkan (10)
• DX12 (11)

---

GPUOpen-LibrariesAndSDKs/AMF#435

LTR requires MaxConsecutiveBPictures=0

---

https://visionular.ai/what-are-av1-golden-frames/

Golden Frames / Long Term Reference Frames

Golden frames are useful in boosting compression efficiency because not only are high quality and are long-term references, but they significantly improve the coding efficiency of the subsequent frames specially in scenes with low motion or repetitive elements, resulting in lower bit-rates without reducing video quality.

See also:
https://docs.nvidia.com/video-technologies/video-codec-sdk/11.1/nvenc-video-encoder-api-prog-guide/index.html#long-term-reference

---

GPUOpen-LibrariesAndSDKs/AMF#500 (comment)

The main reason for adding low latency parameters is to force clocks to stay high and keep latency low.

LowLatencyInternal=true

---

NEVER use "Alternative Reference Frames"
ALT-REF (temporally filtered) frames
VERY blurry
Used by VP9 (2-pass) & VP10 (AV1)
https://gitlab.com/AOMediaCodec/SVT-AV1/-/blob/master/Docs/Parameters.md#av1-specific-options

While OBS currently overrides the rate control method to CBR for streaming

https://github.com/obsproject/obs-studio/pull/9352#discussion_r1515202908

🤡🤡🤡

...

Most AMF parameters explained:
https://github.com/GPUOpen-LibrariesAndSDKs/AMF/wiki/AMF-Encoder-Settings-and-Tuning-in-FFmpeg

AMD Display Library dithering:
https://github.com/GPUOpen-LibrariesAndSDKs/display-library/blame/29beb5a1217ab5805c5032284fdcdc56418bce36/include/adl_defines.h#L1013

• MaxConsecutiveBPictures supersedes BPicturesPattern
• MaxConsecutiveBPictures=1 or greater supersedes PALongTermReferenceEnable=true
• PAPerceptualAQMode=1 supersedes EnableVBAQ=true
• PATemporalAQMode=1 or 2 supersedes EnableVBAQ=true and PAPerceptualAQMode=1
• PAHighMotionQualityBoostMode=1 supersedes HighMotionQualityBoostEnable=true

x265

• ctu=32 for 1920x1080
• no-sao=1
• limit-sao=1
• b-pyramid=0
• aq-mode=3
• mcstf=1 Motion-compensated spatio-temporal filtering. Give it a try

x264

too many lookahead threads significantly degrades lookahead accuracy
https://code.videolan.org/videolan/x264/-/blob/master/encoder/encoder.c?ref_type=heads#L1294

x264 has a complex lookahead module designed to estimate the coding cost of frames that have not yet been
analyzed by the main encoder module. It uses these estimations to make a variety of decisions, such as adaptive B-frame
placement, explicit weighted prediction, and bit allocation for buffer-constrained ratecontrol. For performance reasons, it
operates on a half-resolution version of the frame and calculates SATD residuals only, doing no quantization or
reconstruction.

It is important to note that, in x264, the regular B-frame quantizer offsets (--pbratio) are disabled
when MBtree is on, since they serve the same role.

One might assume similarly that macroblock-tree can replace keyframe quantizer offsets. However, testing
suggested this was not the case

Benchmark Software Decoder

ffmpeg -benchmark -i H264.mkv -an -f null -

Important AMD encoder settings

PASceneChangeDetectionEnable=false PALookAheadBufferDepth=41 MaxNumRefFrames=4

---

• PASceneChangeDetectionEnable=false (default=true) - adaptive I-frames, Scenecut, requires Lookahead
• AdaptiveMiniGOP=true (default=true) - adaptive B-frames, B-adapt, requires Lookahead
• BReferenceEnable=true (default=true) - Use B-frames as Reference frames

---

MaxNumRefFrames=4 =4 for High 4.2 Profile (1920x1080@60)

https://github.com/obsproject/obs-studio/blob/master/plugins/obs-nvenc/nvenc.c

https://github.com/obsproject/obs-studio/blob/master/plugins/obs-ffmpeg/obs-ffmpeg-nvenc.c

obsproject/rfcs#7

Important NVenc settings

AVC
lookaheadDepth=32 useBFramesAsRef=1

HEVC
lookaheadLevel=3 lookaheadDepth=32

Note: NV_ENC_LOOKAHEAD_LEVEL_0 has the highest performance,
while NV_ENC_LOOKAHEAD_LEVEL_3 has the highest quality. It increases bitrate significantly and should not be used.

---

https://github.com/obsproject/obs-studio/blob/ef93ef6950df0a5435c2756c11ace92d15e00d52/plugins/obs-nvenc/nvenc-properties.c#L213

useBFramesAsRef=

=1 Any B-frame can be a reference frame

=2 Halved - Uses the following logic:

4 B-frames = 2 reference frames
3 B-frames = 1 reference frame
2 B-frames = 1 reference frame
1 B-frame = 1 reference frame <-- untested, but should be logical to clamp to 1

Show all supported encoders in FFMPEG:

ffmpeg -encoders

Show all encoder options:

ffmpeg -h encoder=hevc_nvenc

Lossless: NVenc

ffmpeg -y -benchmark -i INPUT.mkv -c:v hevc_nvenc -tune lossless -pix_fmt p010le -an OUTPUT.mkv

Almost Lossless: Apple ProRes 422 High Quality

ffmpeg -y -benchmark -i INPUT.mkv -c:v prores_ks -q:v 0 -profile:v 3 -bits_per_mb 1024 -vendor apl0 -pix_fmt yuv422p10le 1024.mov

Note:-bits_per_mb can go up to 8192. Higher = less lossy.

Lossless: UT Video

ffmpeg -y -benchmark -i INPUT.mkv -c:v utvideo -color_primaries bt709 -color_trc bt709 -colorspace bt709 -color_range tv OUTPUT.mkv

AV1 Basically lossless settings:

ffmpeg -y -benchmark -i INPUT.mkv -c:v libsvtav1 -svtav1-params "enable-tf=0:mbr=10M:tile-columns=2:tile-rows=0" -g 240 -qp 14 -preset 4 -pix_fmt yuv420p10le SVT_AV1.mkv

Libaom AV1 settings:

ffmpeg -y -benchmark -i TEST.mkv -c:v libaom-av1 -pix_fmt yuv420p10le -g 240 -cpu-used 4 -auto-alt-ref 0 -aq-mode 0 -tile-columns 2 -tile-rows 0 -row-mt 1 -crf 16 -pass 1 libaom-pass1.webm
ffmpeg -y -benchmark -i TEST.mkv -c:v libaom-av1 -pix_fmt yuv420p10le -g 240 -cpu-used 4 -auto-alt-ref 0 -aq-mode 0 -tile-columns 2 -tile-rows 0 -row-mt 1 -crf 16 -pass 2 libaom-pass2.webm

NVEnc by rigaya
Docs

High Quality Example:

NVEncC64.exe --codec hevc --cqp 18 --preset p7 --output-depth 10 --lookahead 32 --ref 8 --split-enc forced_2 --colormatrix bt709 --colorprim bt709 --transfer bt709 --colorrange limited -i INPUT.mkv -o OUTPUT.mkv

190 FPS with 2560x1440 input from the RTX 5070 Ti

Lossless:

NVEncC64.exe --lossless --colormatrix bt709 --colorprim bt709 --transfer bt709 --colorrange limited -i INPUT.mkv -o OUTPUT_Lossless-avc.mkv

ffmpeg -y -benchmark -i TEST.mkv -c:v libx264 -x264-params "ref=8:keyint=240:bframes=3:b-adapt=2:rc-lookahead=60:aq-mode=3:me=umh:merange=24:subme=9:trellis=2:lookahead-threads=2:threads=16" -crf 14 -pix_fmt yuv420p10le AVC_14.mkv