Computational Modelling Group

Workshop  24th September 2012 9 a.m.  B25/1011

CUDA Fortran and OpenAcc GPU Development Workshop

Dave Norton
Portland Group International

Categories
C++, Computer Science, CUDA Fortran, Fortran, GPU, GPU-libs, HECToR, HPC, HPCx, Linux, OpenMP, Scientific Computing, Software Engineering
Submitter
Karl Wilkinson

This workshop has been postponed, the new date will be announced ASAP. Apologies for the inconvenience, Karl

Dave Norton from the Portland Group International (PGI) will be running a one day introductory workshop on the CUDA Fortran and OpenACC models for developing GPU accelerated code. Two models that provide a straightforward solution for porting code to GPUs.

CUDA Fortran is the Fortran counterpart to CUDA C whilst OpenACC uses a pragma based approach. Acceleration of scientific codes using these models has been reported to result in acceleration of up to a factor of 120x in exceptional cases. Examples can be found on the PGI newsletter: http://www.pgroup.com/resources/articles.htm#pginsider

This will be a hands on workshop utilizing the GPU accelerated nodes on Iridis3. Whilst the details of the course are not yet confirmed the following material will be covered:

  1. Programming with CUDA Fortran

  2. Programming with OpenACC (Approximately 2/3 of the course)

  3. Differences between OpenACC and the PGI Accelerator model

  4. Using OpenACC across multiple GPUs with OpenMP

Numbers will be limited, so reserve your place as soon as possible by contacting me at k.wilkinson@soton.ac.uk.

Please feel free to contact me with any questions about the workshop.

Best regards,

Dr. Karl A. Wilkinson