Accelerating ffmpeg
Signalogic has an R&D project underway to accelerate ffmpeg using drop-in accelerator cards. Here are some of the features and objectives:
- Substantial increase in performance. Accelerators use arrays of Texas Instruments C66x multicore CPUs which integrate CPU functionality with compute intensive calculation, including instructions optimized for video processing
- Simple, maintainable programming interface based on an OpenMP superset. Avoid linear and hand-optimized assembly language, C source sections pragmatized (compute-intensive sectins)
- Achieve high performance and reduced heat (power consumption) in industry standard servers
ffmpeg subfolder for Texas Inst C66x architecture (using a superset of OpenMP) |
The table shows some expected video transcode channel capacity comparisons for a 2U box with dual Sandy Bridge.