Direct access to GPUs accelerates video-processing tasks
Graphics processing units (GPUs) are used across industry to accelerate complex and time-consuming tasks in a range of applications, allowing many previously slow and computationally complex problems to be solved in real time.
Historically, however, getting video into and out of the GPU from third-party hardware introduced unnecessary delays, because there was no direct synchronization or communication between devices.
Now, a solution to the problem has been developed by engineers at Nvidia (Santa Clara, CA, USA) in the form of an application programming interface (API) that allows third-party hardware to communicate directly with the company's GPUs.
By taking advantage of the new "GPUDirect for VideoAPI," a video I/O card can be fully synchronized with the GPU, eliminating CPU overhead and, most importantly, reducing latency, allowing third-party hardware to stream video to and from a GPU at sub-frame transfer rates (see image and click to enlarge details).
According to Greg Estes, industry executive, media and entertainment for Nvidia, enabling lower latency solutions for processing real-time video streams will benefit not just those working in broadcast and production facilities, but those developing applications in manufacturing, healthcare, and government agencies as well.
Nvidia GPUDirect for Video technology is available immediately, free of charge, directly from Nvidia for video I/O manufacturers, proprietary hardware providers, and video switcher manufacturers. It is supported on Windows 7 and Linux, and is designed to work with Nvidia Quadro 4000, Quadro 5000, Quadro 6000 and Tesla C-Series GPUs.
More information is available in the Nvidia developer zone.
-- Posted by Vision Systems Design