NVEnc Release Notes

8.00beta5

Fix --dolby-vision-rpu in AV1 encoding. ( #672 )

8.00beta4

Fix some codecs not being able to decode with avsw since 8.00b2.
Fix interlaced encoding not working when using filters since 8.00b1.
Add 10.0, 10.1, 10.2, 10.4 options to --dolby-vision-profile. ( #672 )

8.00beta3

Fix --dolby-vision-profile not working with readers other than avhw/avsw. ( #663 )
Fix memory allocation failure when using yuv422 output. ( #670 )

8.00beta2

Improve auto gpu selection for multi (NVIDIA) GPU environments.

8.00beta1

Add support for NVENC SDK 13.0. Driver 570.00 or later is required.
- add support for H.264/AV1 in --tf-level.
- Add forced_4 to --split-enc.
- Support HEVC --temporal-layers.
- Add experimental support for yuv422 output. (but untested, as I have no Blackwell GPUs)
Update CUDA for Windows build to CUDA 11.8.
- Requires driver 452.39 or later.
- Requires CC3.5 or later GPUs.
  - NVIDIA GPUs GeForce Maxwell generation or later and some Kepler (GTX Titan, 780(Ti), 730-710)
- GPUs that do not support CC3.5 (GTX770, 760, 740, 6xx) are no longer supported.
No longer supports builds with CUDA 10 or earlier.
- Builds for Ubuntu 18.04 removed.
Noe NVEnc will be able to handle "frame transfer from CPU to GPU", "filtering", and "frame submission to encoder" in parallel.
Merge AV1 --max-temporal-layers to --temporal-layers to be same as other codecs.
Improve dolby vision rpu handling. ( #663 )

7.82

Now AV1 --level can be specified again.
Added checks for --max-bitrate and --ref to avoid errors when specifying --level.
Avoid --qvbr not working properly when setting --max-bitrate too big. ( #486 )
Fix some case that audio not being able to play when writing to mkv using --audio-copy.
Now more detailed encoder error information will be shown.

7.81

Avoid width field in mp4 Track Header Box getting 0 when SAR is undefined. ( #680 )

7.80

- Fix some of the paramters of --vpp-libplacebo-tonemapping not working properly.

7.79

Fix documents for --vpp-resize nvvfx-superres ( #658 ).
Fix --trim being offset for a few frames when input file is a "cut" file (which does not start from key frame) and is coded using OpenGOP.