Braided Parallelism - A Programmers Perspective Benedict Gaster, Programming Models Architect, AMD
Building is at 4th and College. Room 86-01 is in the basement, take the elevator or stairs down to basement and follow the signs.
GPU architectures are often described in confusing and overly optimistic terms. The AMD Fusion APU architecture provides improvements in communication latency and bandwidth between devices, so the problem of choosing the right core on which to execute code becomes both more acute and more flexible as communication between tasks becomes easier. But as the CPU and GPU become ever closer, how does one program these machines? This talk takes a look at these emerging architectures; touches on the current parallel programming models for CPUs and GPGPUs; and follows this with a perspective on what future programming models for heterogeneous architectures will look like.