Name
Incremental Live Video Processing in IP-based GPU-leveraging Software
Date & Time
Monday, October 21, 2024, 3:30 PM - 4:00 PM
Description

Explore the shift from fixed-function appliances to general-purpose COTS hardware in live media production, emphasizing the flexibility and scalability of software-driven workflows. This track delves into optimizing software performance on COTS systems to maximize media processing density, reducing costs in data centers and cloud services. Learn about offloading media processing to GPUs and high-bandwidth NICs, and discover a novel approach to incrementally process SMPTE-2110-20 streams as frames are received. The session highlights how this method lowers latency and increases the number of streams per system, enhancing overall efficiency in live media workflows.

Technical Depth of Presentation
Intermediate
Take-Aways from this Presentation

The efficient use of COTS hardware permit additional media stream processing, effectively increasing the amount of media processing per unit of hardware Current IP-based GPU-leveraging software video workflows will wait for an entire video frame to arrive from the network before starting video frame processing. Taking advantage of the reliable pacing of a ST 2110-20 video essence stream and HEVC or AV1 slice specifications, the latency of GPU processing and subsequent encoding can be reduced thereby utilizing less resources allowing for the processing of additional streams in parallel making more efficient use of the COTS hardware system

Virtual Session Link