Thursday, 29 November 16:00 - 16:45 | Conference Hall K
In recent years, GPU computing has evolved to deliver teraflops of floating-point compute power to workstations. This trend has necessitated that scientific and visualization applications require a mix of compute and graphics capabilities, in addition to efficiently processing large amounts of data. These applications include physically-based simulations, image processing, computer vision, augmented reality, etc. To maximize performance, the applications must be designed to allow data to be passed efficiently between compute and graphics contexts. Current compute APIs include functions dedicated to inter-operability with graphics APIs. In the first part of this talk, we will use CUDA C/C++ and OpenGL to illustrate this. The second part of the talk will focus on inter-operability on a system scale. In particular, the focus will be on the challenges and benefits of dedicating one GPU for compute and another for graphics, and how it translates to application-related design decisions which enable efficient, cross-GPU, compute, and graphics inter-operability.