Show Your Support: Did you know that the hundreds of articles written on Phoronix each month are mostly authored by one individual? Phoronix.com doesn't have a whole news room with unlimited resources and relies upon people reading our content without blocking ads and alternatively by people subscribing to Phoronix Premium for our ad-free service with other extra features.
Better AMD Radeon VCE Video Encode Performance Coming To Linux
With a pending merge request to Mesa, AMD is updating the default motion estimation parameters to the Gallium3D video acceleration encode front-end. These updated values in turn should improve video encoding performance of H.264 with AMD Radeon GPUs having the VCE block.
The MR is under review but presumably will be merged still in time for Mesa 22.0. AMD's VCE 1.0 block premiered with Radeon HD 7000 series GPUs as well as Trinity/Richland APUs back in the day. VCE continued to be improved upon and the latest iteration was worked into Vega-based GPUs. But since Navi or Raven/Picasso on the APU front is now Video Core Next (VCN) as the successor to VCE.
Changing around a few values for speeding up AMD VCE H.264 encode performance on Linux...
This AMD VCE tuning for performance stems from long-open bug requests around slow VA-API H.264 encoding on Polaris GPUs, poor FFmpeg performance on Polaris, and related performance issues. With the updated parameters, these issues should be a problem of the past. One of those affected by the poor performance indicated that the encode speed is now much faster using OBS Studio for Polaris GPUs on Linux.