NVDEC and CUVID API names

NVDEC names NVIDIA’s decoding engine while CUVID and NVDECODE names persist in software interfaces, library symbols and media-program wrappers that reach its supported decode paths.

NVIDIA video decoding appears under several related names. NVDEC names the dedicated engine. NVDECODE API names its software interface, while CUVID remains visible in function, header and library names.

Reading those names at their own levels avoids inventing separate codec formats or hardware blocks.

Current interfaces retain CUVID symbols

The SDK 13.1 guide still documents these names:

Name Role
cuviddec.h and nvcuvid.h Decoder-interface headers
cuvidCreateDecoder Create the selected decoder
cuvidDecodePicture Submit picture reconstruction
cuvidMapVideoFrame Obtain the traditional mapped/post-processed output
cuvidUnmapVideoFrame Release that mapped output
nvcuvid.dll or libnvcuvid.so Platform library naming in the documented stack

The names are part of the software contract. Seeing cuvid in a trace does not establish that the application is using a separate ancient decoder engine.

The NVDEC article develops the surface and ownership behavior behind those calls.

A media program can expose different wrappers

An application can integrate dedicated codec wrappers with cuvid-related names, or use a hardware-acceleration path that describes its device frames as CUDA.

For example, h264_cuvid names a H.264 decoding wrapper in builds that provide it. A -hwaccel cuda option requests a supported hardware-acceleration arrangement under the media program's own configuration.

These names do not replace the actual source codec. H.264 remains H.264, and HEVC needs its own suitable decode configuration.

They also do not identify the encoder. A later h264_nvenc name concerns the H.264 output encoding stage.

Device context and engine execution are different facts

A decoder API can use a CUDA context and expose CUDA-addressable output. That makes the picture useful to GPU processing or a compatible encoder.

The compressed reconstruction can still be carried out by NVDEC's specialized hardware. A CUDA memory/interface label and a compute-engine activity counter have different scopes.

See Also

References