NVIDIA Video Codec SDK 13.1 introduces several major improvements for GPU-accelerated video pipelines. Key encode additions include Hierarchical Reference Mode for AV1 supporting up to 31 B-frames for better quality without performance penalties, and UHQ tuning info combined with iterative encoding. On the decode side, per-macroblock statistics (QP, coding unit type, motion vectors) are now exposed for H.264/HEVC with no CPU overhead, enabling GPU-accelerated analytics. A new frame-accurate seek API allows random access to specific frames via a GOP-aware architecture, useful for AI inference and video editing workflows. The transcoder pipeline is redesigned with a modular, queue-based architecture and a new AppTransZeroCopy application that eliminates intermediate buffer copies by sharing CUarray surfaces between NVDEC and NVENC, reducing SM utilization and memory footprint. An official Docker-based development environment is also included for reproducible setup.
