NVIDIA VA-API 0.0.6 Driver Works On Multi-Threaded Decode, Improved GPU Selection
The open-source, unofficial project providing a NVIDIA VA-API driver on Linux systems built atop the proprietary driver's NVDEC video decode interface is out with a new feature release. This NVIDIA-VAAPI-Driver effort continues to be driven in large part for allowing GPU-accelerated video acceleration in Firefox and other software only targeting the open Video Acceleration API.
The nvidia-vaapi-driver project continues marching along for implementing VA-API atop NVDEC with particular emphasis on Firefox video support. This NVIDIA VA-API implementation works currently for H.264, H.265/HEVC, VP8, VP9, MPEG-2, and VC-1. There also remains experimental AV1 support too for use with newer NVIDIA RTX 30 graphics cards.
With today's nvidia-vaapi-driver 0.0.6 release there is multi-threaded decode where the final processing steps are punted off to a separate thread, improved GPU selection and honoring the "NVD_GPU" environment variable for selecting among multiple NVIDIA GPUs, an option to limit the maximum instances per-process, and a variety of other fixes and low-level improvements.
Downloads and more details on the v0.0.6 release for this unofficial NVIDIA VA-API driver implementation via GitHub.
The nvidia-vaapi-driver project continues marching along for implementing VA-API atop NVDEC with particular emphasis on Firefox video support. This NVIDIA VA-API implementation works currently for H.264, H.265/HEVC, VP8, VP9, MPEG-2, and VC-1. There also remains experimental AV1 support too for use with newer NVIDIA RTX 30 graphics cards.
👁 Image
NVIDIA's Linux driver has long been focused on the Video Decode and Presentation API for Unix (VDPAU) or more recently their cross-platform NVDEC interface. For software like Firefox only targeting the vendor-independent VA-API, this open-source project comes to the rescue for NVIDIA Linux customers.
With today's nvidia-vaapi-driver 0.0.6 release there is multi-threaded decode where the final processing steps are punted off to a separate thread, improved GPU selection and honoring the "NVD_GPU" environment variable for selecting among multiple NVIDIA GPUs, an option to limit the maximum instances per-process, and a variety of other fixes and low-level improvements.
Downloads and more details on the v0.0.6 release for this unofficial NVIDIA VA-API driver implementation via GitHub.