XSEDE Workshop: OpenACC | November 7
ODU’s Research Computing Services is hosting an XSEDE Workshop on OpenACC:
Tuesday, November 7
11:00 a.m. to 5:00 p.m.
Learning Commons @ Perry Library, Conference Room 1311
This hands-on workshop will introduce parallel computing on graphical processing units (GPU) using OpenACC. OpenACC is a directive-based approach similar to OpenMP to expose parallelism and offload computational work onto a GPU, which can speed up compute-intensive applications by a large factor. This workshop is presented by the Pittsburgh Supercomputing Center (PSC) and simulcast at ODU. The materials will be taught at an introductory level. Basic C or Fortran programming skill is necessary, but no prior knowledge of parallel programming is required. Attendees will have an opportunity to apply the lessons learned through hands-on exercises, which will be performed on a real supercomputer at PSC. Please bring your own laptop in order to complete the exercises.
Go to https://goo.gl/Y3FYfd to learn more and to register.
Seating is limited and the course is open to the public, so please register soon. Please contact hpc@odu.edu if you have questions regarding this workshop.
---------
NOTE: To register for the workshop, you need to have an XSEDE User Portal (XUP) account. If you don’t have one, please create one first at https://portal.xsede.org/my-xsede?p_p_id=58&p_p_lifecycle=0&p_p_state=maximized&p_p_mode=view&_58_struts_action=%2Flogin%2Fcreate_account , then sign up using the short link shown above.
PS: Why do we want to care about using GPUs? On Turing we currently have 15 nodes with state-of-the-art GPUs from NVIDIA, which have great potentials for increasing your calculation’s throughput by factors like 2-10x compared to using only CPUs (which translates to much less time waiting for your results). However, the CUDA programming language, which has become the most popular way of programming NVIDIA GPUs, has a rather steep learning curve and frequently ends up being quite tedious to use. OpenACC has emerged as an alternative and much easier method to use GPUs in a productive manner. By introducing OpenACC directives (that are very similar to OpenMP directives) to your existing C, C++, or Fortran programs, you can incrementally port your program to use GPUs and reap the speed-up of GPUs without being entangled by CUDA-style complexity.
Posted By: Dawn Midgette
Date: Fri Nov 03 08:46:03 EDT 2017