Title: GPU Processors for data parallel solutions for high performance computation research
Duration: Half day
Description:
An introduction to a data-parallel programming model for GPU Computing using the CUDA architecture on NVIDIA GPUs. The talk will include an overview of the various application programming interfaces including OpenCL and CUDA language extensions, tools, and libraries, available for it. Optimization considerations and guidelines to achieve best performance will be discussed using real examples and case studies.
NVIDIA CUDA enabled GPU's have started to be appear in many HPC environments - and have given birth to the persona supercomputer concept. The parallelization using the CUDA architecture can scale to tens of thousands of cores. The tutorial is targeted to engineers and scientists who have interest in using parallel programming for computational science problems and will enable them to use this new breed of supercomputer using the CUDA architecture. The tutorial will cover key hardware and software technologies, heterogeneous computing, the kinds of problems that CUDA is well suited for, and the kinds of improvements you can expect to get, compute capabilities and the different features supported, various programming interfaces including C for CUDA, CUDA Driver API, OpenCL and Fortran integration, performance metrics and optimizations techniques relating to data structuring, execution configuration and instruction.
Presenters:
John Roberts joroberts at nivdia.com
and
Paulius Micikevicius pauliusm at nvidia.com
There is no fee to attend the tutorials, but please, fill out the registration form.
ICCS 2009 is organised by | ||
|
|
 |