Tighter NIC/GPU Integration Yields Next-Level Media Processing Performance
As the media industry further consolidates building services based on commercial-off-the-shelf (COTS) hardware, the requirement for increasing performance continues to accelerate. The processing of video resolutions up to ultra-high definition (HD) (UHD) and 8K, higher frame rates, increased bit depth, and high-dynamic-range (HDR) imagery requires tighter integration and optimization between the network interface controller (NIC) and the graphical processing unit (GPU). Combined with simultaneous input and output of multiple streams; this creates the potential for performance bottlenecks that must be unlocked. This article describes how COTS hardware platforms running Graphics Processing Unit (GPU)s alongside ST 2059–2 Precision Time Protocol (PTP) locked Network Interface Cards (NIC)s, which are accurately pacing packets according to ST 2110–21 requirements, may further increase their performance throughput by reducing central processing unit (CPU)-driven data copy overhead and, in doing so, they reduce processing latency and jitter as well as more effectively use GPU resources while freeing up CPU resources.
- Print ISSN
- 1545-0279
- Electronic ISSN
- 2160-2492
- Published
- 2023-01
- Content type
- Original Research
- Keywords
- 2110, Data Processing Unit (DPU), Graphics Processing Unit (GPU), Network Interface Cards (NIC), Precision Time Protocol (PTP), timing
- DOI
- 10.5594/JMI.2022.3196732