Going Parallel (part III : GPU)

Europe/Brussels
Shanon (building Maxwell, first floor) (Louvain-La-Neuve)

Shanon (building Maxwell, first floor)

Louvain-La-Neuve

Place du Levant 3 1348 Louvain-la-Neuve Belgium
Description
Going parallel is crucial to speed up your gain and to use the full power of HPC center. In this session, we will cover parralelism with GPU (CUDA and OpenACC)

Contents:

 

  • Directive Based Parallel programming on GPU (OpenACC)
  • Parallel programming on GPU with CUDA
  • GPU optimization

Prerequisite:

  • Being able to use SSH with private keys 
  • Being familiar with a text editor 
  • Mastering the Linux command line and the GNU utilities (mkdir, cp, scp, etc.)
  • Passive knowledge of C
Type: Hands-on
Target audience: Rookie programmer
Must: This session is a must-have for anyone that want to move to GPU or use parralelism with shared memory
Registration
Registration
20 / 50
    • 09:30 11:00
      Parallel programming on GPU with CUDA 1h 30m

      Cuda is the most common way to code GPU application from scratch develloped by NVidia, it allows to push NVidia GPU to the maximum of their possibilities

      Contents:

      • Basic of GPU hardware
      • Basic of programming with CUDA
      • using nsight to profile your code

       

      Prerequisite:

      • Being able to use SSH with private keys 
      • Being familiar with a text editor 
      • Mastering the Linux command line and the GNU utilities (mkdir, cp, scp, etc.)
      • Basic knowledge of C

      Type: Hands-on
      Target audience: Rookie programmer
      Must: This session is a nice-to-have for those who need to develop highly parallel scientific software.

      Speaker: Olivier Mattelaer (UCLouvain/CISM)
    • 11:15 12:45
      Directive Based Parallel programming on GPU (OpenACC) 1h 30m

      This course is an introduction to OpenACC and OpenMP to target GPU. These programming models are based on programmers inserting hints into their C/C++ or Fortran programs on how the code can be parallelized. OpenMP, was initially designed for shared-memory system with CPU but has recently been extended to target heterogeneous systems with accelerators like GPU.  OpenACC, on the other end, was designed from the start to target accelerators.

      Contents:

      • Basics of OpenMP offloading and OpenACC
      • Compute and worksharing constructs
      • Data management
      • Asynchronous execution and synchronization

       

      Prerequisite:

      • Being able to use SSH with private keys 
      • Being familiar with a text editor 
      • Mastering the Linux command line and the GNU utilities (mkdir, cp, scp, etc.)
      • Basic knowledge of C/C++ or Fortran

      Type: Hands-on
      Target audience: Rookie programmer
      Must: This session is a nice-to-have for those who need to develop highly scientific software.

      Speaker: Orian Louant (Université de Liège)
    • 14:00 16:30
      GPU Optimization 2h 30m
      Speaker: Miguel De Le Court