NvEncSharp provides a dotnet core and .net framework interface to Nvidia's NvEnc hardware video encoder (nvEncodeAPI), the CUDA video decoder (nvcuvid), and the CUDA APIs (nvcuda).
The CUDA API implementation is version 10020. Most (but some) deprecated APIs are not included.
There's additional CUDA kernels for frame resizing and decoding inside LibCudaLibrary.
- Add the nuget package.
- Reference the encoder sample code: Capturing the screen as a video.
- Reference the decoder sample code: Decoding container-free h264 NAL packets to .bmp files and the screen.
Testing support for linux and x86 would be great. This has only been used and tested on 64bit Windows.
Large sections of the CUDA API have not been tested. There's likely transcription errors.
Maintaining forward compatibility with CUDA APIs.
Sorry, this is way more a mess than it should be.
Note: The CUDA SDK is only needed for the sample projects. And is only needed for GPU based color space conversions.
- Install CUDA SDK 12.6, see notes below. CUDA SDK site.
- Select Visual Studio integration in the installer.
This originally targeted 10.2 CUDA SDK, and the API definitions still match that, but has been updated to target 12.6. I have not found a direct download link for 12.6.
The 12.6 documentation suggests the archived versions are now offered using conda but I have not tried this.
Updating the SDK version may be easier than finding the depricated version.
- Modify
CUDA_VERSIONinsideNvEncSharp/src/NvEncSharp.Cuda.Library/NvEncSharp.Cuda.Library.propsto your target version. - If it compiles it compiles. If not, it's possible that the CUDA SDK has introduced breaking changes that need to be addressed.