Topic 16: GPU and Accelerators Computing

Accelerator-based computing systems invest significant fractions of hardware real estate to execute critical computation with vastly higher efficiency than general-purpose CPUs. Amdahl’s Law of the Multi-core Era suggests that such an heterogeneous approach to parallel computing is bound to deliver better scalability and power-efficiency than homogeneous system scaling. While General Purpose Graphics Processing Units (GPGPUs) have catalyzed research in this area, new ideas emerge to help us model, deconstruct and analyze the performance of accelerators, develop new standards for programming accelerators at a high level of abstraction, and port end-to-end applications on accelerator-based systems. Topic 16 provides a forum to discuss advances in all aspects of GPUand accelerator-based computing.