CUDA Programming and Performance Tuning with IBM XL Compilers on POWER8 Systems

Abstract

The exploitation of general-purpose computing on a heterogeneous parallel system proves to be highly efficient for running several technical computing workloads.  This approach can also apply to a wide range of workloads, such as chemistry, bioinformatics, molecular biology, engineering, and big data analytics.  The Power S822LC featured NVIDIA GPU is built on industry standards and incorporating community innovation from the OpenPOWER Foundation.  This presentation will give an overview of CUDA programming with IBM C/C++ and Fortran compilers and provide performance running tips for different workloads on Power S822LC systems.

Speaker Bio

Istvan Reguly is a Lecturer at PPCU, Hungary. His research focus is high performance computing, he designs and develops domain specific programming methods for scientific computing that enable the efficient use of modern hardware architectures. Istvan was previously a Research Associate at the Oxford e-Research Centre, University of Oxford, and he holds a PhD and an MSc in computer science.

Shereen Ghobriais the Offering Manager for XL compilers on Power systems with the mission of improving performance for OpenPower platform. He started his career in compiler development at IBM Toronto lab in 1997 and took different technical and management roles in C/C++ compilers, Java, Automatic Binary Optimizer for z/OS, and WebSphere Application Server. His experience includes leading teams working on different platforms with focus on Power and zSystems. Shereen holds an MSc in computer science.