NVIDIA Video Codec SDK
Type: Technology Tags: CUDA, NVIDIA, GPU, Video, Encoding, Decoding, NVENC, NVDEC, Media Related: NVIDIA-DALI, CV-CUDA, NPP, NVIDIA-Optical-Flow-SDK, nvImageCodec, NVIDIA-Video-Effects-SDK, NVIDIA-AI-for-Media-SDKs Sources: NVIDIA official documentation, https://docs.nvidia.com/maxine/vfx/latest/index.html Last Updated: 2026-04-29
Summary
NVIDIA Video Codec SDK is a comprehensive set of APIs providing hardware-accelerated video encode (NVENC) and decode (NVDEC) on NVIDIA GPUs for Windows and Linux. It supports H.264, HEVC (H.265), and AV1 encoding, and decoding for MPEG-2, VC-1, H.264, H.265, VP8, VP9, and AV1. On Blackwell GPUs it achieves up to 8K240 FPS with multiple hardware encoding engines, enabling faster-than-real-time video processing at all scales.
Detail
Purpose
Software video codecs consume substantial CPU resources and cannot achieve real-time performance at high resolutions or frame rates at scale. NVIDIA Video Codec SDK exposes dedicated NVENC and NVDEC hardware engines on the GPU, enabling faster-than-real-time video encode/decode that frees CPU for other tasks and dramatically reduces power consumption per encoded frame.
Key Features
- NVENC: hardware H.264, HEVC, and AV1 video encoding
- NVDEC: hardware decoding for MPEG-2, VC-1, H.264, H.265, VP8, VP9, AV1
- Up to 8K240 FPS on NVIDIA Blackwell GPUs with multiple encode engines
- Ultra-high quality (UHQ) AV1 mode approaching software codec quality at superior throughput
- 422 progressive and interlaced encode/decode for professional media workflows
- C-style APIs with C++ wrapper classes
- PyNvVideoCodec: Python bindings for easy integration
- DirectX and Vulkan Video API support
- Compatible with DALI, CV-CUDA, and deep learning frameworks
Use Cases
- Video transcoding for streaming platforms
- Deep learning training data preparation from video datasets
- Game broadcasting and cloud gaming (NVIDIA GeForce NOW)
- Virtual desktops and remote visualization (NVIDIA RTX Virtual Workstation)
- AI-powered video analytics
- Live video streaming and conferencing
- Secure video playback with DRM
Hardware Requirements
- NVIDIA GPU with NVENC/NVDEC hardware engines (Turing or newer recommended for AV1)
- Windows and Linux supported
- Blackwell GPUs for maximum performance (8K240 FPS)
Language Bindings
- C (primary C-style API)
- C++ (wrapper classes)
- Python (PyNvVideoCodec)
- DirectX 12 and Vulkan Video interfaces
Connections
- NVIDIA-DALI — DALI uses NVDEC hardware decode for GPU-accelerated video loading in training
- CV-CUDA — CV-CUDA preprocesses frames decoded by NVDEC for computer vision pipelines
- NPP — NPP processes raw decoded frames from NVDEC for further signal processing
- NVIDIA-Optical-Flow-SDK — Optical Flow SDK builds on decoded frames from Video Codec SDK
- nvImageCodec — nvImageCodec handles image codecs; Video Codec SDK handles video codecs
- NVIDIA-Video-Effects-SDK — video effects can sit between decode/capture and encode in media pipelines.
- NVIDIA-AI-for-Media-SDKs — AI media SDK hub for adjacent video enhancement workflows.