GPUDirect for Video

  • Published August 2015

Back in the days where USB2.0, FireWire and Gigabit Ethernet were the primary interfaces of industrial cameras, the bandwidth for video transmission was the bottleneck of professional image processing systems.

Today, with bandwidths of up to 25 Gpbs of 4x CoaXPress, even the latest CPUs are overwhelmed by the enormous amount of data. However, many filter, convolution and matrix-vector-operations can easily be performed on modern GPUs.

With Active Silicon’s frame grabbers and NVIDIA’s CUDA programming model for parallel computing, it is possible to leverage the GPUDirect™ for Video technology, which bypasses the CPU and optimizes the transfer of video frames from our grabber to the GPU of NVIDIA Tesla™ and Quadro™ cards. Our well documented API and SDK sample code allow for the hassle-free integration of parallel computing techniques on standard computer hardware. Thus, system developers can fully benefit from the high resolution and frame rate of modern industrial cameras and minimize computation time and hardware costs.

Learn more about NVIDIA’S GPUDirect for Video here.